GNU Emacs integrated computing environment

My comprehensive "dotemacs" (.emacs) for GNU/Linux

Created: 2019-08-15
Updated: 2020-05-31, 12:26 +0300.
See this file's upstream git history. Everything is part of my dotfiles' repository.

Table of Contents

1 Overview

1.1 Canonical links to this document

1.2 What is this

The present document, referred to in the source code version as, contains the bulk of my configurations for GNU Emacs. It is designed using principles of "literate programming": a combination of ordinary language and inline code blocks. Emacs knows how to parse this file properly so as to evaluate only the Elisp ("Emacs Lisp") included herein. The rest is for humans to make sense of my additions and their underlying rationale.

Literate programming allows us to be more expressive and deliberate. Not only can we use typography to its maximum potential, but may also employ techniques such as internal links between sections. This makes the final product much more useful for end users than a terse script.

Each section provides information about the code it contains. In case you feel something is missing, I maintain a Frequently Asked Questions section (when in doubt, or to offer feedback, suggestions, further comments, etc., do contact me).

In more practical terms, this document is written using org-mode. It contains all package configurations for my Emacs setup. To actually work, it needs to be initialised from another file that only covers the absolute essentials.

1.2.1 Contents of my init.el

The is actually loaded from an other file, named init.el as per the Emacs convention. Mine is designed to add the community-driven MELPA archive to the list of package repositories, configure use-package (see comprehensive explanation below) and then load the file with my configurations (i.e. the present document). Old version prior to Emacs 27

For reference, these were the contents of my init.el prior to Emacs 27.1.

(require 'package)
(setq package-enable-at-startup nil)
(add-to-list 'package-archives
             '("melpa" . ""))
(unless package--initialized (package-initialize))

(require 'org)
(org-babel-load-file (expand-file-name "~/.emacs.d/")) Current version for Emacs 27+

Whereas in Emacs 27.1 and onward, they are modified thus:

(require 'package)

(add-to-list 'package-archives
             '("melpa" . ""))

;; Initialise the packages, avoiding a re-initialisation.
(unless (bound-and-true-p package--initialized)
  (setq package-enable-at-startup nil)

;; Make sure `use-package' is available.
(unless (package-installed-p 'use-package)
  (package-install 'use-package))

;; Configure `use-package' prior to loading it.
  (setq use-package-always-ensure nil)
  (setq use-package-always-defer nil)
  (setq use-package-always-demand nil)
  (setq use-package-expand-minimally nil)
  (setq use-package-enable-imenu-support t)
  ;; The following is VERY IMPORTANT.  Write hooks using their real name
  ;; instead of a shorter version: after-init ==> `after-init-hook'.
  ;; This is to empower help commands with their contextual awareness,
  ;; such as `describe-symbol'.
  (setq use-package-hook-name-suffix nil))

  (require 'use-package))

(require 'org)
(setq vc-follow-symlinks t)
(org-babel-load-file (expand-file-name "~/.emacs.d/")) The "early init" for Emacs 27+

Starting with Emacs 27.1, an early-init.el is now required to control things with greater precision. Its code is as follows:

;; Do not initialise the package manager.  This is done in `init.el'.
(setq package-enable-at-startup nil)

;; Allow loading from the package cache.
(setq package-quickstart t)

;; Do not resize the frame at this early stage.
(setq frame-inhibit-implied-resize t)

These adjustments are of paramount importance due to changes in the way Emacs initialises the package manager. Prior to Emacs 27.1, the init.el was supposed to handle that task by means of calling package-initialize. Whereas for Emacs 27.1, the default behaviour is to start the package manager before loading the user's init file. This can create unexpected results with regard to how existing configuration files are parsed—or at least that was my experience with certain settings not being parsed consistently (was not able to reproduce it reliably). I prefer the old behaviour so I simply tell the early-init.el to defer the process of initialising the package manager to when init.el is evaluated.

1.2.2 About `use-package'

This is a tool that streamlines the configuration of packages. It handles everything from assigning key bindings, setting the value of customisation options, writing hooks, declaring a package as a dependency for another, and so on.

Though it might not be readily apparent, a "package" in Emacs parlance is any elisp file that is evaluated by Emacs. This includes libraries that are shipped with the upstream distribution as well as code that comes from other sources.

As such use-package is not a package manager, in the sense of installing, removing, listing packages. It only configures things using a declarative syntax. The package manager of Emacs is package.el while there are other tools available from third parties, such as straight.el.

Unlike a typical extensible program, there is no real distinction between native Emacs code and the one that comes from third parties. There is no externally-facing limited set of features that other tools can plug into. Emacs is an interpreter of lisp (Emacs Lisp), meaning that any elisp is evaluated in real time, making Emacs behave in accordance with it.

I have an hour long presentation about switching to Emacs, where this and other topics are discussed in greater detail. It is good to understand the context in order to appreciate the differences between the various use-package declarations documented herein.

The three types of use-package declarations that I use:

  1. To set up external packages. Those are denoted by the inclusion of :ensure which means that the package manager should make sure the package is installed.
  2. To configure default packages. No :ensure is needed for them.
  3. To declare custom or otherwise experimental packages that are not available in any repository and which I handle manually and plan to review at a later date. Those include a :load-path that makes their code available to my environment.

In several package declarations you will see a :diminish keyword that leverages the diminish package. This affects the so-called "lighter" that each package may define. The lighter is the piece of text that a tool will append to the mode line. For example, Flyspell's lighter is "Fly". With :diminish we demand that the lighter be removed (the information is still available when running C-h m).

With use-package we can improve the start-up performance of Emacs in a few fairly simple ways. Whenever a command is bound to a key it is configured to be loaded only once invoked. Otherwise we can specify which functions should be autoloaded by means of the :commands keyword.

Furthermore, and if absolutely necessary, I define all variables that are supposed to be immutable with the :custom keyword. This writes them to the custom.el that I specify further below. You should, however, consider that to be the exception, as all minor modes, custom functions, or other configurations are normally specified under the :config keyword. The activation of a mode should always be the very last thing, once all variables have been set. Make sure to read the manual for more information on the individual keywords.

I set the variable use-package-hook-name-suffix to nil in order to always type in the proper name of a hook. The default behaviour is to omit the suffix for convenience. But that means that we can no longer benefit from the contextual awareness of help/documentation commands (e.g. C-h o over any of the symbols/functions/variables below will put the thing at point as the first completion option).

Last but not least, you should be warned of a common error with handling package installs (with or without use-package): if Emacs complains that the package you want no longer exists, it means that you must refresh your package index because there is a new version of that package, so the old one that is still registered on your list has been removed from the source. Do that with either M-x package-refresh-contents or the package browser M-x list-packages.

The following snippet of elisp sets up and configures use-package to my liking. It is already referenced in the previous section concerning the contents of my init.el. This is due to changes in how Emacs 27.1 starts up. Whereas before I used to configure use-package from inside this document.

;; Make sure `use-package' is available.
(unless (package-installed-p 'use-package)
  (package-install 'use-package))

;; Configure `use-package' prior to loading it.
  (setq use-package-always-ensure nil)
  (setq use-package-always-defer nil)
  (setq use-package-always-demand nil)
  (setq use-package-expand-minimally nil)
  (setq use-package-enable-imenu-support t)
  ;; The following is VERY IMPORTANT.  Write hooks using their real name
  ;; instead of a shorter version: after-init ==> `after-init-hook'.
  ;; This is to empower help commands with their contextual awareness,
  ;; such as `describe-symbol'.
  (setq use-package-hook-name-suffix nil))

  (require 'use-package))

Settings that do not have a corresponding package are declared using the special use-package emacs notation.

1.2.3 About the source code version of this document

In the org-mode version of this document, I make sure that the above-referenced code blocks are not declared as an emacs-lisp source but rather as mere examples, so they are not accidentally parsed by the actual setup.

Actual code blocks are wrapped between #+begin_src and #+end_src tags (not visible in the website version of this page). For Emacs 27.1, such templates can be quickly inserted with C-c C-, (this works both for empty blocks and active regions). For more on the matter, refer to Org's section further below.

As for the various settings included herein, you can learn even more about them by using Emacs' built-in documentation (great for discovering new features and pieces of functionality).

Additionally, you will notice some metadata tags specific to org-mode below each heading. These are generated by the functions that are defined in the package declaration for org-id. The idea is to keep anchor tags consistent when generating a new HTML version of this document.

This metadata also makes it possible to create immutable internal links, whenever a reference is needed. To create such links, you can use C-c l to capture the unique ID of the current section and then C-c C-l to create a link (the former is defined in the Org package declaration—this is an internal link in action).


Copyright (c) 2019-2020 Protesilaos Stavrou <>

This file is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

This file is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with this file. If not, see

2 Base settings

This section contains the relatively few configurations that are needed prior to the setup of everything else.

2.1 Always rebuild init and refresh package cache

When Emacs expands this org-mode file into the actual elisp code, it creates a new document: derives emacs-init.el. The latter holds my customisations in the state they were in at the time the document was created. Any updates require a rewrite.

To make sure that I do not load older settings after having made some tweaks to my dotemacs, I want to delete that derived file when I instruct Emacs to terminate its process. This ensures that edits I made to are parsed into a new emacs-init.el at the next startup.

Similarly, I want to maintain an up-to-date cache with the installed packages, which can be used to speed up boot times, ceteris paribus. This "quickstart" method is part of Emacs 27+.

(use-package emacs
  (defun prot/delete-emacs-init ()
    (let ((configs "~/.emacs.d/emacs-init.el"))
      (when configs
        (delete-file configs))))
  :hook ((kill-emacs-hook . prot/delete-emacs-init)
         (kill-emacs-hook . package-quickstart-refresh)))

2.2 Remove modeline "lighters"

As was noted in the section about `use-package' we can remove the so-called "lighter" text that modes append to the mode line. This is all that Diminish does and you will see a :diminish keyword in the relevant packages.

If you actually want to edit the lighters, check the Delight package.

(use-package diminish
  :after use-package)

2.3 Put customisation settings in a "custom.el"

When you install a package or use the various customisation interfaces to tweak things to your liking, Emacs will append a piece of elisp to your init file. I prefer to have that stored in a separate file.

Note that a common source of inconsistencies with configurations arises from a conflict between your code and what is stored in the custom snippet. When something does not seem to work as intended, make sure to check custom.el and edit or delete it if necessary (if deleted, it will be created again the next time you start Emacs).

(use-package cus-edit
  (setq custom-file "~/.emacs.d/custom.el")

  (unless (file-exists-p custom-file)
    (write-region "" nil custom-file))

  (load custom-file))

2.4 Fundamentals for Hydras

A "hydra" is an ad-hoc mode for mapping commands to a group of more convenient key bindings. In an active hydra, all of its keys—the hydra's "heads"—can be typed in any order, keeping the mode active in the process (if wanted), while at least one key performs the function of exiting the hydra.

In practice, I do not use hydras as substitutes for mode maps. Instead, I treat them as complementary to minor modes that offer enhancements to a particular major mode but are otherwise not very useful outside that niche. Think of cases such as a code linter (the minor mode) as part of a programming session (major mode).

The reason I use hydras this way is due to the official key binding conventions that I [generally] follow (as documented in the Emacs manual). The idea of using C-c [a-z] for all these minor modes is good only if you have a few of them and only if they are more useful outside a particular task. In other words, it does not scale.

I thus employ C-c h as a common prefix for all my hydras (knowing that it is meant for private use), binding each sequence to the relevant minor mode's map. An extra key completes the chord following mnemonics. For instance, a hydra about the linter when that is active is C-c h l, for the outline viewer C-c h o, for git merge conflicts C-c h g, and so on.

Look for the defhydra macro in the relevant package declarations (these will evolve over time). Hydras are colour-coded as a means of denoting their behaviour with regard to the use of non-defined keys and the criteria for exiting.

Colour Hydra heads Other keys
red Accept and Continue Accept and Exit
pink Accept and Continue Accept and Continue
amaranth Accept and Continue Reject and Continue
teal Exit Reject and Continue
blue Exit Accept and Exit

Also check the project's source for more information.

(use-package hydra :ensure)

2.5 Base typeface configurations

Any font I choose must support Latin and Greek character sets, be readable at both small and large sizes, preferably offer roman and italic variants with corresponding bold weights, not be too thin, not have too short of an x-height, not be too wide, not have a name that directly advertises some brand, not try to call too much attention to its details, and be equally readable against light and dark backdrops.

While there are many good free/libre options available, only a handful of them cope well with my fairly demanding needs. Some look good at large point sizes. Others lack Greek characters. While a few of them are virtually unreadable when cast on a light background (bitmap fonts in particular, including the otherwise sublime Terminus).

As of 2020-05-18, my choice of primary monospaced font is a variant of Iosevka (version 3.0 or higher), which fixes some of the issues I had with earlier iterations of this typeface. Prior to that, I was using "Hack" and then my patched version of Hack, the latter of which I now qualify as second-best due to its (i) slightly worse italics than Iosevka, and (ii) its lack of a companion proportional typeface for mixed font setups.

For a proportionately-spaced typeface, I go with Iosevka Sparkle because it combines nicely with the monospaced version and gives a neat typewriter look and feel (further below I specify settings that govern the Mode for proportional fonts and link to other relevant sections).

I use the *.ttf files of Iosevka, while I make sure that all variations I do not intend to use are removed from my system. Doing this through dired (see section on Dired (directory editor, file manager)):

  • % m to mark files matching a regular expression.
  • .*\(semi\|medium\|thin\|light\|extra\|heavy\|oblique\|extended\).*
  • D to delete the marked items.

This process leaves me with regular, italic, bold, and bold-italic.

Lastly, note that on a modern GNU/Linux system that uses the fontconfig library, per-user fonts are stored in ~/.local/share/fonts.

2.5.1 Primary font settings

Moving on to my configurations, the functions that follow the pattern prot/SCOPE-font allow me to conveniently define the settings I need to use in the given context. Their docstrings should offer you all the information you need.

While there are many ways to define a font family in Emacs, I find that the best way is to do it at the "face" level. Faces are constructs that specify attributes for a piece of text. There range from a foreground and background colour, to text properties such as the font family, its height, weight, etc. (Emacs themes are, in essence, programs that configure those faces).

To set our baseline typeface, we configure the default face. To understand the syntax used here, do C-h f set-face-attribute. In essence we are specifying the font that should be used in case no other font applies for the given face. This is actually a good thing because there are many cases where you want a face to retain its own attributes (e.g. let org-mode inline code be presented in its monospace font while using a sans-serif for the main text—see Mode for proportional fonts).

The two main ways to set a font using set-face-attribute are as follows:

Set individual attributes
Define the values of keywords like :family, :height, :weight on a per-attribute basis.
Set font parameters
Pass a single string that is consistent with how fontconfig (GNU/Linux) reads fonts. This allows us to apply parameters that are specified in the fontconfig user spec, such as whether to apply an antialiasing effect or tweak the hinting style. Note that all fontconfig parameters are optional and will default to the rules that apply across the operating system

I choose the latter because it is more flexible. For completeness, this is the syntax, replacing all capitalised words with the appropriate values (read the docs with C-h f set-face-attribute):

  • (set-face-attribute 'FACE nil :family "NAME" :height "HEIGHT")
  • (set-face-attribute 'FACE nil :font "FAMILY-SIZE:PARAM1=PARAMVALUE1")
(use-package emacs
  (setq x-underline-at-descent-line nil)
  (setq underline-minimum-offset 0)

  (defconst prot/default-font "Iosevka Curly"
    "The default typeface.")

  (defconst prot/font-params "autohint=false:hintstyle=hintslight:embeddedbitmap=false"
    "The default fixed-pitch typeface.")

  (defun prot/set-face-attribute-font (family size)
    "Set `default' face font to FAMILY at SIZE."
    (set-face-attribute 'default nil :font (concat family
                                                   (number-to-string size)

  (defun prot/laptop-fonts ()
    "Fonts for the small laptop screen.

Pass desired argument to `prot/font-sizes' for use on my
small laptop monitor."
    (when window-system
      (prot/set-face-attribute-font prot/default-font 12)))

  (defun prot/desktop-fonts ()
    "Fonts for the larger desktop screen.

Pass desired argument to `prot/font-sizes' for use on my larger
desktop monitor (external display connected to my laptop)."
    (when window-system
      (prot/set-face-attribute-font prot/default-font 13.5)))

  (defun prot/screencast-fonts ()
    "Fonts for screen casts and video demos.

Pass desired argument to `prot/font-sizes' for use during screen
casting.  The idea is to make it easier for viewers to see what I
am doing."
    (when window-system
      (prot/set-face-attribute-font prot/default-font 16.5)))

  (defun prot/presentation-fonts ()
    "Fonts for presentations and video blogs.

Pass desired argument to `prot/font-sizes' for use during
presentations.  Also see `prot/org-presentation'."
    (when window-system
      (prot/set-face-attribute-font prot/default-font 20)))

  (defun prot/fonts-per-monitor ()
  "Use font settings based on screen size.

Choose between `prot/laptop-fonts' and `prot/desktop-fonts'
depending on the width of the monitor.  The calculation is based
on the maximum width of my laptop's screen.  So if an external
display is attached, then it is considered a desktop scenario.

While this function is interactive, it is best to run it with the
`after-init-hook' or perhaps some other event that tracks
monitor-related events."
    (when window-system
      (if (<= (display-pixel-width) 1366)

  :hook (after-init-hook . prot/fonts-per-monitor))

2.5.2 Font families for other generic face groups

What I am doing here is define the exact font family to be used by these generic faces, which are designed to be inherited by other faces in order to keep things consistent across a theme. We could do the same for practically any face, but I think that is a step too far.

Do not forget that my standard font family for the default face and also for fixed-pitch is declared in the previous section (I normally use a monospaced font for everything, so there is no real need to decouple default and fixed-pitch).

Note that the :height in this snippet is specified as a floating point, which means that it is a multiple of the base size (defined above). For the sake of completeness, when the :height property is an integer it is understood as the point size × 100. So 10.5pt is expressed as 105.

Also bear in mind that on a GNU/Linux system there normally is no need to test for the presence of a font before setting it. This is handled by fontconfig with rules that can be defined at the user or system level (things work out-of-the-box on every distro I tried).

(use-package emacs
  (defconst prot/variable-pitch-font "Iosevka Sparkle"
    "The default variable-pitch typeface.")

  (set-face-attribute 'variable-pitch nil :family prot/variable-pitch-font :height 1.0)
  (set-face-attribute 'fixed-pitch nil :family prot/default-font :height 1.0))

2.5.3 Mode for proportional fonts

When Emacs operates in a graphical terminal, it can display text using mixed font settings: a variety of heights and families (in addition to other typegraphic attributes). This means that it can draw text on the same line that has potentially profoundly different properties character-by-character. For example, a sans-serif typeface for prose with a monospaced font for inline code (for the underlying faces that may be used, see Font families for other generic face groups).

While we can handle things on our own by calling set-face-attribute for each face we are interested in, there is a built-in mechanism to quickly toggle the use of proportionately-spaced fonts (in contrast to the monospaced ones which usually are the standard). Depending on your theme this will not yield good results in Org mode, so be warned.

For what it's worth, my Modus themes are designed to take full advantage of this facility (see Modus themes and other visuals). Any face that is sensitive to spacing or that would break the layout if displayed in a variable width is designed to explicitly inherit from the fixed-pitch face (the exact font family was specified in the previous section, though the only thing that matters is to be monospaced).

With the following package declaration my idea is to integrate prot/variable-pitch-mode into some "reading mode" configurations that I have, mutatis mutandis. Likely candidates are the toggle for olivetti-mode (see the section on “Focus mode” for writing) and my simple-yet-super-effective plain text presentations with Org mode (see simple presentations inside of Emacs). Please note though that this may be reviewed upon further testing and experimentation.

(use-package face-remap
  :diminish buffer-face-mode            ; the actual mode
  (defun prot/variable-pitch-mode (&optional size)
    "Toggle `variable-pitch-mode' and additional parameters.
SIZE is intented for one of my functions for setting the primary
font size.

Example: (prot/variable-pitch-mode (prot/screencast-fonts))"
    (if (bound-and-true-p buffer-face-mode)
          (variable-pitch-mode -1)
          (setq-local cursor-type 'box)   ; TODO better restore original value
          (when size                      ; TODO restore previous value
      (variable-pitch-mode 1)
      (setq-local cursor-type 'bar)
      (when size

2.5.4 Typeface suitability test

Here is a simple test I have come up with to make an initial assessment of the overall quality of the font: can you discern the character at a quick glance? If yes, your choice of typeface is good prima facie, else search for something else.

Note that this test is not perfect, since many typefaces fall short in less obvious ways, such as the space between the characters. Also note that the website version of this document may not accurately represent the typeface I am using.


Sample character set
Check for monospacing and Greek glyphs


3 Selection candidates and search methods

3.1 Completion framework and extras

As discussed in my video about Emacs' buffer and window management, the optimal way of using Emacs is through searching and narrowing selection candidates. Spend less time worrying about where things are on the screen and more on how fast you can bring them into focus. This is, of course, a matter of realigning priorities, as we still wish to control every aspect of the interface.

Since the day I switched to Emacs (July 2019), I was using some completion framework other than the default. I started out with ivy and its companion packages, switched to the built-in ido and then went back to the former. I never experimented with any sort of customisations to the generic minibuffer experience. Nor did I ever bother with the oldest built-in tool of the sort (icomplete) that is designed to complement the minibuffer's internal mechanisms for matching items. Not until ~10 February 2020…

It turns out that, despite appearances to the contrary, the defaults are very powerful, opening up a range of possibilities to those eager to learn and experiment (a common theme in Emacs).

In the following package declarations I am defining several functions that enhance the experience of icomplete. These are part of a learning process to (i) explore the internals of Emacs and study how various problems are solved with elisp, and (ii) determine how far one can go, in terms of efficient functionality, without deviating from the norms inherent to the tools that are shipped with Emacs.

What this also means is that I am deprecating Ivy and its dependants as well as reviewing any other package that expected their presence or somehow contributed to them. I do understand, however, that some users may still need to see the code and customisations I had for those, thus I am keeping everything in place with the :disabled keyword. Everything is under the heading of important configurations that I stopped using.

3.1.1 Minibuffer essentials and Icomplete (built-in completion)

For newcomers, watch my video demo of Icomplete (2020-02-26). This section and subsquent ones contain a lot of documentation and several elements. Make sure to also follow any links from here, or contact me in case something is amiss. There is also an entry in the "Frequently Asked Questions" that might help you make sense of the complete picture: Why keep many completion UIs? (typology of my completion interfaces).

The minibuffer is the locus of extended command interaction. Whether it is about offering input to a prompt, performing a search, executing a function by its name, the minibuffer remains at the epicentre. The minibuffer and Icomplete are meant to work in tandem as part of a singular framework. The default experience is far more powerful than it seems to be. It can get even better by tweaking the available customisation options and defining our own extensions.

Icomplete is the tool that offers incremental completion feedback for what the minibuffer is doing (where appropriate). There is no added layer of complexity, no need for another metaphor of interactivity beside what you already use the minibuffer for. In essence, Icomplete is just a visualisation of what is going on under the hood, meaning that all the completion mechanisms actually work even if icomplete-mode is disabled—though you obviously lose the visual feedback.

Now some comments about my implementation:

  • The, dare I say, sublime “orderless” package is developed by Omar Antolín Camarena and you will find it on MELPA (I use more packages from Omar in my dotemacs, such as Icomplete vertical mode). It provides the orderless completion style for efficient out-of-order grouped pattern matching. The components can be determined using several styles, such as regexp, flex, prefix, initialism (check the README because there are lots of variations and extras). Delimiters are literal spaces by default, but can be configured to match other characters, with hyphens and slashes being likely choices. As such, Orderless can supersede—and greatly improve upon—the completion styles that come built into Emacs, adding to them the powerful out-of-order capability.
  • My prot/orderless-literal-dispatcher is based on an example from the Orderless README. By appending an equals sign to the end of a component I am instructing Orderless to match that component literally. Obviously one can use the style dispatchers for each of the matching patterns, though I find this to be sufficient for my needs.
  • The completion-styles try to match candidates using one style at a time (assuming you have more than one), moving from the first to the last until something is matched. In practice though, orderless is so powerful that there is no pressing need to use the built-in styles. An exception is a niche functionality of the partial-completion: with it you can navigate to a filesystem path like ~/.l/s/fo for ~/.local/share/fonts. So my modest recommendation is to use those two styles to cover every case.
  • The completions-format concerns the layout of the *Completions* buffer that pops up after trying to complete a non-unique match (or by typing ? from inside the minibuffer). By default, it can be focused directly with M-v while inside a minibuffer prompt.
  • To enhance the experience of the Completions buffer, I define several keys that make motions easier and consistent with other read-only interfaces. The h key calls a command of mine to offer help (documentation) for the item at point, typically a function or a variable. I also define M-v to take me back to the minibuffer while inside the *Completions* (and s-v to do it from anywhere else).
    • The placement of the Completions, Help, and other buffers is defined in detail in the section about Window rules and basic tweaks, specifically within the display-buffer-alist.
  • I enable recursive minibuffers. This practically means that you can start something in the minibuffer, switch to another window, call the minibuffer again, run some commands, and then move back to what you initiated in the original minibuffer. To exit such recursive edits, hit C-] (abort-recursive-edit), though the regular C-g should also do the trick. The minibuffer-depth-indicate-mode will show an indicator next to the minibuffer prompt if a recursive edit is in progress.
  • UPDATE 2020-05-16: Deprecated in favour of the new "embark" package, though the functionality is exactly the same from an end-user perspective (see Embark (actions for completion candidates)). ORIGINAL: My prot/complete-kill-or-insert-candidate is a simple yet effective tool that I bind to three key chords. With M-o w you copy the current candidate to the kill ring and continue with the minibuffer session. With M-o i you insert the candidate in the buffer from where the minibuffer was called, without appending to the kill ring, while still keeping the session active. And M-o j will insert the candidate in the same way and immediately exit all recursive edits.
  • Note the nuances in the behaviour between RET and C-j. With the Return key, we instruct the minibuffer to expand the current candidate and then exit the session, if possible. Whereas C-j is meant to insert the minibuffer's contents exactly as they are and exit immediately. You need the latter in cases where you want foo but the match is for foobar.
  • The key bindings in the pattern of s-KEY follow the principles I outlined in my note about the use of the Super key.

Also check my configurations pertaining to the minibuffer history. After about two months of full time usage (as of 2020-04-03), I am confident in the built-in mechanism's ability to sort things well enough and to surface the results I am most likely interested in, based on previous selections. This means that we do not need a third-party scoring and filtering library like prescient or amx.

Now here is the actual code for the minibuffer part (icomplete is further below):

(use-package minibuffer

  ;; Super-powerful completion style for out-of-order groups of matches
  ;; using a comprehensive set of matching styles.
  (use-package orderless
    (setq orderless-regexp-separator "[/\s_-]+")
    (setq orderless-matching-styles

    (defun prot/orderless-literal-dispatcher (pattern _index _total)
      (when (string-suffix-p "=" pattern)
        `(orderless-literal . ,(substring pattern 0 -1))))

    (defun prot/orderless-initialism-dispatcher (pattern _index _total)
      (when (string-suffix-p "," pattern)
        `(orderless-strict-leading-initialism . ,(substring pattern 0 -1))))

    (setq orderless-style-dispatchers '(prot/orderless-literal-dispatcher
    :bind (:map minibuffer-local-completion-map
                ("SPC" . nil)))         ; space should never complete

  (setq completion-styles
        '(orderless partial-completion))
  (setq completion-category-defaults nil)
  (setq completion-cycle-threshold 3)
  (setq completion-flex-nospace nil)
  (setq completion-pcm-complete-word-inserts-delimiters t)
  (setq completion-pcm-word-delimiters "-_./:| ")
  (setq completion-show-help nil)
  (setq completion-ignore-case t)
  (setq read-buffer-completion-ignore-case t)
  (setq read-file-name-completion-ignore-case t)
  (setq completions-format 'vertical)   ; *Completions* buffer
  (setq enable-recursive-minibuffers t)
  (setq read-answer-short t)
  (setq resize-mini-windows t)

  (file-name-shadow-mode 1)
  (minibuffer-depth-indicate-mode 1)
  (minibuffer-electric-default-mode 1)

  (defun prot/focus-minibuffer ()
    "Focus the active minibuffer.

Bind this to `completion-list-mode-map' to M-v to easily jump
between the list of candidates present in the \\*Completions\\*
buffer and the minibuffer (because by default M-v switches to the
completions if invoked from inside the minibuffer."
    (let ((mini (active-minibuffer-window)))
      (when mini
        (select-window mini))))

  (defun prot/focus-minibuffer-or-completions ()
    "Focus the active minibuffer or the \\*Completions\\*.

If both the minibuffer and the Completions are present, this
command will first move per invocation to the former, then the
latter, and then continue to switch between the two.

The continuous switch is essentially the same as running
`prot/focus-minibuffer' and `switch-to-completions' in
    (let* ((mini (active-minibuffer-window))
           ;; This could be hardened a bit, but I am okay with it.
           (completions (or (get-buffer-window "*Completions*")
                            (get-buffer-window "*Embark Occur*")
                            (get-buffer-window "*Embark Live Occur*"))))
      (cond ((and mini
                  (not (minibufferp)))
             (select-window mini nil))
            ((and completions
                  (not (eq (selected-window)
             (select-window completions nil)))))

  ;; Technically, this is not specific to the minibuffer, but I define
  ;; it here so that you can see how it is also used from inside the
  ;; "Completions" buffer
  (defun prot/describe-symbol-at-point (&optional arg)
    "Get help (documentation) for the symbol at point.

With a prefix argument, switch to the *Help* window.  If that is
already focused, switch to the most recently used window
    (interactive "P")
    (let ((symbol (symbol-at-point)))
      (when symbol
        (describe-symbol symbol)))
    (when arg
      (let ((help (get-buffer-window "*Help*")))
        (when help
          (if (not (eq (selected-window) help))
              (select-window help)
            (select-window (get-mru-window)))))))

  ;; This will be deprecated in favour of the `embark' package
  (defun prot/completions-kill-save-symbol ()
    "Add symbol-at-point to the kill ring.

Intended for use in the \\*Completions\\* buffer.  Bind this to a
key in `completion-list-mode-map'."
    (kill-new (thing-at-point 'symbol)))

;;;; DEPRECATED in favour of the `embark' package (see further below),
;;;; which implements the same functionality in a more efficient way.
;;  (defun prot/complete-kill-or-insert-candidate (&optional arg)
;;     "Place the matching candidate to the top of the `kill-ring'.
;; This will keep the minibuffer session active.
;; With \\[universal-argument] insert the candidate in the most
;; recently used buffer, while keeping focus on the minibuffer.
;; With \\[universal-argument] \\[universal-argument] insert the
;; candidate and immediately exit all recursive editing levels and
;; active minibuffers.
;; Bind this function in `icomplete-minibuffer-map'."
;;     (interactive "*P")
;;     (let ((candidate (car completion-all-sorted-completions)))
;;       (when (and (minibufferp)
;;                  (or (bound-and-true-p icomplete-mode)
;;                      (bound-and-true-p live-completions-mode))) ; see next section
;;         (cond ((eq arg nil)
;;                (kill-new candidate))
;;               ((= (prefix-numeric-value arg) 4)
;;                (with-minibuffer-selected-window (insert candidate)))
;;               ((= (prefix-numeric-value arg) 16)
;;                (with-minibuffer-selected-window (insert candidate))
;;                (top-level))))))

  ;; Defines, among others, aliases for common actions to Super-KEY.
  ;; Normally these should go in individual package declarations, but
  ;; their grouping here makes things easier to understand.
  :bind (("s-f" . find-file)
         ("s-F" . find-file-other-window)
         ("s-d" . dired)
         ("s-D" . dired-other-window)
         ("s-b" . switch-to-buffer)
         ("s-B" . switch-to-buffer-other-window)
         ("s-h" . prot/describe-symbol-at-point)
         ("s-H" . (lambda ()
                    (prot/describe-symbol-at-point '(4))))
         ("s-v" . prot/focus-minibuffer-or-completions)
         :map minibuffer-local-completion-map
         ("<return>" . minibuffer-force-complete-and-exit)
         ("C-j" . exit-minibuffer)
         ;;;; DEPRECATED in favour of the `embark' package
         ;; ("M-o w" . prot/complete-kill-or-insert-candidate)
         ;; ("M-o i" . (lambda ()
         ;;              (interactive)
         ;;              (prot/complete-kill-or-insert-candidate '(4))))
         ;; ("M-o j" . (lambda ()
         ;;              (interactive)
         ;;              (prot/complete-kill-or-insert-candidate '(16))))
         :map completion-list-mode-map
         ("h" . prot/describe-symbol-at-point)
         ("w" . prot/completions-kill-save-symbol)
         ("n" . next-line)
         ("p" . previous-line)
         ("f" . next-completion)
         ("b" . previous-completion)
         ("M-v" . prot/focus-minibuffer)))

And the following package declaration is for the interactive completion interface: icomplete (remember, Icomplete just offers the interface, not the underlying mechanisms). As such, do not forget to also check the entire section above this message, the part on minibuffer history, and my docs+configs for ad-hoc verticality.

Overview of the following package declaration:

  • The values of all variables that pertain to the delay of feedback are tentative. My initial tests suggest that they behave exactly the way I want, but this might change once I test them further. In short, I want to introduce some delay in the feedback I get when the list of candidates is long. This paradoxically makes things feel faster while I am still typing (because by that time the list has been narrowed to my input).
  • For versions of Emacs above 27, there is a minor mode called fido (Fake IDO, where ido is an alternative option to Icomplete that tries to do more than just completion). This new mode changes some of the primary key bindings and commands of icomplete so that it meets the expectations of Ido users. It is not meant as a fully fledged replacement of Ido, as its scope is much narrower, at least for the time being. If you are curious, check the source code for both icomplete and ido with the help of M-x find-library. I do not use Fido, though I think it is a welcome step in the right direction. For the sake completeness, note that you can always bind some Fido command without activating the minor mode.
  • The keybindings I specify define motions that ensure consistency between regular editing and the rotation of the candidates' list. The default icomplete key bindings leave something to be desired and are therefore repurposed for my extensions.
  • As with the keys for the minibuffer, note the nuances in the behaviour between RET and C-j. With the Return key, we tell Icomplete to expand the current candidate and then exit the minibuffer if possible. Whereas C-j is meant to insert exactly what is in the minibuffer, which can be tricky in some circumstances. Use the latter when Icomplete is matching foobar but you only want foo (needed when renaming files or when creating links in Org to non-existent candidates).
(use-package icomplete
  :after minibuffer                     ; Read that section as well
  (setq icomplete-delay-completions-threshold 100)
  (setq icomplete-max-delay-chars 2)
  (setq icomplete-compute-delay 0.2)
  (setq icomplete-show-matches-on-no-input t)
  (setq icomplete-hide-common-prefix nil)
  (setq icomplete-prospects-height 1)
  ;; (setq icomplete-separator " · ")
  ;; (setq icomplete-separator " │ ")
  ;; (setq icomplete-separator " ┆ ")
  ;; (setq icomplete-separator " ¦ ")
  (setq icomplete-separator (propertize " ┆ " 'face 'shadow))
  (setq icomplete-with-completion-tables t)
  (setq icomplete-in-buffer t)
  (setq icomplete-tidy-shadowed-file-names nil)

  (fido-mode -1)                        ; Emacs 27.1
  (icomplete-mode 1)

  (defun prot/icomplete-minibuffer-truncate ()
    "Truncate minibuffer lines in `icomplete-mode'.
  This should only affect the horizontal layout and is meant to
  enforce `icomplete-prospects-height' being set to 1.

  Hook it to `icomplete-minibuffer-setup-hook'."
    (when (and (minibufferp)
               (bound-and-true-p icomplete-mode))
      (setq truncate-lines t)))

  ;; Note that the the syntax for `use-package' hooks is controlled by
  ;; the `use-package-hook-name-suffix' variable.  The "-hook" suffix is
  ;; not an error of mine.
  :hook (icomplete-minibuffer-setup-hook . prot/icomplete-minibuffer-truncate)
  :bind (:map icomplete-minibuffer-map
              ("<tab>" . icomplete-force-complete)
              ("<return>" . icomplete-force-complete-and-exit) ; exit with completion
              ("C-j" . exit-minibuffer) ; force input unconditionally
              ("C-n" . icomplete-forward-completions)
              ("<right>" . icomplete-forward-completions)
              ("<down>" . icomplete-forward-completions)
              ("C-p" . icomplete-backward-completions)
              ("<left>" . icomplete-backward-completions)
              ("<up>" . icomplete-backward-completions)
              ;; The following command is from Emacs 27.1
              ("<C-backspace>" . icomplete-fido-backward-updir))) TODO can registers be inserted via completion? TODO can mark-ring positions be selected? Icomplete vertical mode

When I first switched to icomplete some time in late January to early February 2020, I had to implement my own admittedly sub-par tweaks for displaying candidates vertically and for tailoring that presentation to my particular needs. There was no package "ecosystem" around Icomplete that I could leverage.

Thankfully, Omar Antolín Camarena's icomplete-vertical is here (since early April 2020) to fill in this gap. The package provides a global minor mode for displaying the list of candidates vertically by default: simple and effective.

While fairly young, icomplete-vertical already feels like a mature, feature-complete tool: it offers a robust experience out-of-the-box, but also provides facilities for users to introduce an element of ad-hoc verticality to their bespoke completion functions. More specifically:

  • With icomplete-vertical-toggle, which should be bound to a key inside the minibuffer, we can use whatever layout we want whenever we need it. Excellent!
  • While the icomplete-vertical-do macro can be used to parametrise a custom function with an optional height and unique separator. The latter comes with the option to pick from a list of presets: {solid,dashed,dotted}line.

The project's README should offer all the information you need. Several of my functions offer real-world implementations of the aforementioned (I did, after all, contribute some minor patches and user feedback in the early stages of this package, though all the real work is done by Omar—and it is a lot of work as confirmed by the commit logs and the scope of the diffs).

With regard to verticality, I am the kind of user that actually likes the standard horizontal view as a default presentation. It works splendidly for all my common workflows of executing a command by name, switching to a buffer, changing git branches, and the like. Verticality should, in my opinion, be the default only for lists that present naturally long candidates. A good example is the recentf-list that consists of full filesystem paths, with the kill-ring being another one.

Couched in those terms, the following package declaration does not enable icomplete-vertical-mode globally. Instead, it defines the elements that are necessary for activating verticality on an ad-hoc or per-function basis (and there are lots of them).

In prot/icomplete-yank-kill-ring you will spot a function that disables the sorting of the list. Without it, the kills appear in a seemingly random order, which is highly undesired. I adapted that piece of functionality from the dotemacs of GitHub user jixiuf, following a comment I got from them on my video demo of Icomplete (2020-02-26).

With icomplete-vertical, I can now remove all the poor code I had in place when I got started, relying instead on the thoughtful design and features that Omar has so meticulously developed. In short: the package is a near must-have for any icomplete user, while I expect it to single-handedly convince many users to give Emacs' built-in completion mechanism a fair chance.

Finally, not all "vertical" functions of mine are defined here. Some are also found in completion for projects and directory trees and others still across this document (search for icomplete-vertical-do).

(use-package icomplete-vertical
  :after (minibuffer icomplete) ; do not forget to check those as well
  (setq icomplete-vertical-prospects-height (/ (frame-height) 6))
  (icomplete-vertical-mode -1)

  (defun prot/icomplete-yank-kill-ring ()
    "Insert the selected `kill-ring' item directly at point.
When region is active, `delete-region'.

Sorting of the `kill-ring' is disabled.  Items appear as they
normally would when calling `yank' followed by `yank-pop'."
    (let ((kills                    ; do not sort items
           (lambda (string pred action)
             (if (eq action 'metadata)
                 '(metadata (display-sort-function . identity)
                            (cycle-sort-function . identity))
                action kill-ring string pred)))))
          (:separator 'dotted-line :height (/ (frame-height) 4))
        (when (use-region-p)
          (delete-region (region-beginning) (region-end)))
         (completing-read "Yank from kill ring: " kills nil t)))))

  :bind (("s-y" . prot/icomplete-yank-kill-ring)
         :map icomplete-minibuffer-map
         ("C-v" . icomplete-vertical-toggle)))

3.1.2 Embark (actions for completion candidates)

2020-05-28: In active development. Things are stabilising but may still change considerably.

Embark (Emacs Minibuffer Actions Rooted in Keymaps) is another great contribution by Omar Antolín Camarena. At its core, this tool allows you to perform actions on the completion framework's list of candidates. These can be commands like saving the matching item to the kill ring, inserting it to the current buffer, or even producing a dired buffer from the full list of relevant file names.

As its expanded name suggests, Embark is based on the principle of key maps. A "key map", which is standard in the Emacs space, is the list that holds the association between key chords and the commands they invoke. Key maps provide a form of modality: they apply only where they are enabled, which allows them to be implemented on a context-dependent basis. And they can be layered on top of each other.

This means that you can bind an "entry point" to Embark, which you may then call from the minibuffer (minibuffer-local-completion-map). I set that to M-o (embark-act). This gives you access to all the relevant actions you may call for the given context.

What constitutes an "action" can either be one of the default options provided by embark or a custom function you define yourself. Same for the keymaps you may use.

For all available default actions, check the project's wiki page, or just browse the source code (I prefer the latter and it can be done easily with M-x find-library). This is what it looks like:

(defvar embark-general-map
   '(("i" . embark-insert)
     ("w" . embark-save)
     ("RET" . embark-default-action)
     ("C-g" . embark-cancel)
     ("C-h" . embark-keymap-help)
     ([remap self-insert-command] . embark-undefined)
     ("C-u" . universal-argument))

Embark also provides facilities for placing the list of candidates in either a bespoke buffer (embark-occur) or in a content-aware buffer like dired and ibuffer (embark-export for files and buffers respectively). To make things ever better, there is an option for a "live completion" style, where the "Embark Live Occur" buffer will be updated to reflect your minibuffer input. I choose to use this presciently for commands where I intend to carefully inspect the results themselves. Also read: Why keep many completion UIs? (typology of my completion interfaces).

As for how Embark fits in my completion framework, it quite simply offers a piece of functionality that is altogether absent from the upstream Emacs distribution. This is, by the by, why I greatly appreciate Omar's contributions: they work with standard Emacs mechanisms, extending them in sensible ways, so you have the liberty to mix and match modules from core Emacs, your homegrown functions, and third-party packages, to suit your preferences and expected workflows, without ever committing to an all-encompassing framework (not that those are bad per se).

The embark package is under active development, so expect things to be in a state of flux for the immediate future. I use :load-path because I like to check the source code and try to learn some Elisp along the way—otherwise you should go with something like the quelpa package.

Finally, make sure to check my Window rules and basic tweaks to see how the buffers for "Embark Occur" and its "live" counterpart are positioned as part of the general window layout.

(use-package embark
  :load-path "~/.emacs.d/prot-dev/embark/" ; in development
  ;; This uses a bespoke face from my themes
  (setq embark-indicator
        (propertize "Act" 'face 'modus-theme-refine-magenta))

  (defun prot/embark-insert-exit ()
    "Like `embark-insert' but exits current recursive minibuffer."
    (with-minibuffer-selected-window (insert (embark-target)))

  (defun prot/embark-insert-exit-all ()
    "Like `embark-insert' but exits all recursive minibuffers."
    (with-minibuffer-selected-window (insert (embark-target)))

  (defun prot/embark-live-occur-describe-function ()
    "Enable `embark-live-occur' for `describe-function'."
    (let ((icomplete-mode nil)
          (completing-read-function 'embark-completing-read))
      (call-interactively 'describe-function)))

  (defun prot/embark-live-occur-describe-variable ()
    "Enable `embark-live-occur' for `describe-variable'."
    (let ((icomplete-mode nil)
          (completing-read-function 'embark-completing-read))
      (call-interactively 'describe-variable)))

  (defun prot/embark-live-occur-describe-symbol ()
    "Enable `embark-live-occur' for `describe-symbol'."
    (let ((icomplete-mode nil)
          (completing-read-function 'embark-completing-read))
      (call-interactively 'describe-symbol)))

  :hook (embark-occur-mode-hook . embark-occur-direct-action-minor-mode)
  :bind (("C-h f" . prot/embark-live-occur-describe-function)
         ("C-h v" . prot/embark-live-occur-describe-variable)
         ("C-h o" . prot/embark-live-occur-describe-symbol)
         :map minibuffer-local-completion-map
         ("M-o" . embark-act)
         :map embark-general-map
         ("o" . embark-occur)
         ("O" . embark-export)  ; uses `embark-occur' as a fallback
         ("?" . embark-live-occur) ; same principle as `minibuffer-completion-help'
         ("m" . prot/embark-insert-exit)
         ("j" . prot/embark-insert-exit-all))) TODO Complete review of Embark package declaration [3/6] DONE Review embark-occur buffer display repositioning DONE Check all key maps DONE Update the docs TODO Ensure embark-occur works with my custom file-related commands TODO Explore possibility for dedicated windows of embark-occur (dired) TODO Embark actions for imenu, especially for Org buffers

3.1.3 Imenu (dynamic completion-based buffer navigation)

The imenu is a built-in library that builds an index of buffer positions pointing to semantically-relevant constructs. It then displays the list through a completion interface. Selecting an item repositions the point there.

This is a great tool for quickly jumping to a position in the buffer. Combine it with the sheer power of the completion framework or the possibility to further extend it with little helper snippets and you get a whole new way of thinking about moving around in a buffer.

By default, imenu is designed to not refresh the index it builds, offering a "Rescan" option instead. I find that too conservative for my case, opting instead for automatic rescaning. This has the desirable side-effect of removing the manual refresh command from the list.

With prot/imenu-vertical I provide much-needed verticality to the list, because it typically consists of naturally long candidates that make for a poor fit in the horizontal layout of Icomplete. Notice that it configures the orderless completion style, while it also relies on icomplete-vertical. Review my Completion framework and extras.

The :hook here will simply scroll the buffer so that the landing position of an Imenu interaction becomes the first visible line at the top. I find this to be the easiest way to know where I am.

(use-package imenu
  (setq imenu-use-markers t)
  (setq imenu-auto-rescan t)
  (setq imenu-auto-rescan-maxout 600000)
  (setq imenu-max-item-length 100)
  (setq imenu-use-popup-menu nil)
  (setq imenu-eager-completion-buffer t)
  (setq imenu-space-replacement " ")
  (setq imenu-level-separator "/")

  (defun prot/imenu-vertical ()
    "Use a vertical Icomplete layout for `imenu'.
Also configure the value of `orderless-matching-styles' to avoid
aggressive fuzzy-style matching for this particular command."
    (let ((orderless-matching-styles    ; make sure to check `orderless'
      (icomplete-vertical-do (:height (/ (frame-height) 4))
        (call-interactively 'imenu))))

  :hook (imenu-after-jump-hook . (lambda () (recenter 0)))
  :bind ("C-." . prot/imenu-vertical)) Imenu-list (dedicated sidebar)

While completion for imenu is wonderful in its own right, you may still need to have a sense of the bigger picture. With imenu-list we can place the contents of the current list in a dedicated buffer that sits on the right side of the frame. We may then use that as an index to further support us in the task of navigating through the current buffer.

(use-package imenu-list
  (defun prot/imenu-list-dwim (&optional arg)
    "Convenience wrapper for `imenu-list'.
Move between the current buffer and a dedicated window with the
contents of `imenu'.

The dedicated window is created if it does not exist, while it is
updated once it is focused again through this command.

With \\[universal-argument] toggle the display of the window."
    (interactive "P")
    (if arg
          (if (eq major-mode 'imenu-list-major-mode)
              (pop-to-buffer (other-buffer (current-buffer) t))

  :bind ("C-," . prot/imenu-list-dwim)) Flat imenu index

This package offers a global and a local minor mode for flattening the index of imenu. By default Imenu produces a multi-level index, where appropriate. For example, it will put the heading of this section under that of its parent: Imenu (dynamic completion-based buffer navigation). So to navigate to this point, you need to select the parent and then the child path. Good for a tree view (see Imenu-list (dedicated sidebar)). But not great for fuzzy-style search through the completion UI, because it slows down things considerably, while not offering any typographic or layout means of recognising the structure at first sight.

In other words, a multi-level completion interface leaves much to be desired. By enabling the flimenu-global-mode we get a flat list for the completion-based interaction with imenu. Now we can just search directly for any item on the list. Great!

To make things even better, we can still access the fully fledged tree presentation of imenu-list, as it operates independently.

Lastly, flimenu could also be used as a toggle, by virtue of its local minor mode, though I cannot think of a scenario where I would want that, given the aforementioned.

(use-package flimenu
  (flimenu-global-mode 1))

3.1.4 Generic completion for projects and directory trees

Bear in mind that this section only covers a set of generic commands for querying version-controlled directories, aka "projects", or directory trees in general. They do not cover every type of interaction with projects or directories, but only those that are not specific to other tools, such as dired and ibuffer.

Overview of "advanced searches" as of 2020-05-31 (subject to further refinements):

Function name Key Description
prot/find-file-vc-or-dir M-s f Recursive file search from project root or dir
prot/find-project M-s p Switch to project (projects across many paths)
prot/rg-vc-or-dir M-s g Grep regexp recursively (editable buffer)
prot/rg-ref-in-dir M-s r Grep for ref in present dir (edit buf)
prot/dired-fd-dirs M-s d Recursive search for subdirs
prot/dired-fd-files-and-dirs M-s z Recursive files and dirs
prot/buffers-major-mode M-s b Buffers matching current's major mode
prot/buffers-vc-root M-s v Buffers in current project

Only the first two are in this section. For the rest see:

Many of those accept a prefix argument (C-u), which will put the results in a Dired/Ibuffer. Otherwise they use the completion framework (Icomplete in my case, though technically the underlying completing-read should work with any compliant front-end).

Concerning the design of these key bindings, they are consistent with all my "advanced search methods" (e.g. the default M-s o for occur). If I ever identify a conflict, such as M-s f in dired-mode-map, I disable that function altogether in favour of my own (or remap its key if the function it calls is useful overall).

Recall that my motivation for writing those is to practice some Elisp. Otherwise you may be better off with a library like Projectile.

(use-package project
  (defun prot/find-file-vc-or-dir (&optional arg)
    "Find file by name that belongs to the current project or dir.
With \\[universal-argument] match files by contents.  This
requires the command-line executable called 'rg' or 'ripgrep'."
    (interactive "P")
    (let* ((default-directory (file-name-directory
                               (or (locate-dominating-file "." ".git" )
      (if arg
          (let* ((regexp (read-regexp
                          (concat "File contents matching REGEXP in "
                                  (propertize default-directory 'face 'bold)
                                  ": ")))
                 (results (process-lines "rg" "-l" "--hidden" "-m" "1" "-M" "120" regexp)))
             (icomplete-vertical-do ()
               (completing-read (concat
                                 "Files with contents matching "
                                 (propertize regexp 'face 'success)
                                 (format " (%s)" (length results))
                                 ": ")
                                results nil t))))
        (let* ((filenames-all (directory-files-recursively default-directory ".*" nil t))
               (filenames (cl-remove-if (lambda (x)
                                          (string-match-p "\\.git" x))
          (icomplete-vertical-do ()
             (completing-read "Find file recursively: " filenames nil t)))))))

  (defun prot/find-project (&optional arg)
    "Switch to sub-directory at the specified locations.
With \\[universal-argument] produce a `dired' buffer instead with
all the possible candidates."
    (interactive "P")
    (let* ((dirs (list "~/Git/Projects/" "~/.emacs.d/prot-dev/"))
           (dotless directory-files-no-dot-files-regexp)
           (cands (mapcan (lambda (d)
                            (directory-files d t dotless))
           (projects (mapcar 'abbreviate-file-name cands))
           (buf "*Projects Dired*"))
      (if arg
          (dired (cons (generate-new-buffer-name buf) projects))
        (icomplete-vertical-do ()
           (completing-read "Find project: " projects nil t))))))

  :bind (("M-s p" . prot/find-project)
         ("M-s f" . prot/find-file-vc-or-dir)
         ("M-s l" . find-library)))

3.1.5 In-buffer completions

After trying the popular third-party "Company" package, I felt that it did not offer much of an added value to my typing experience, while its popup feature detracted from the otherwise frugal aesthetics of my setup. Furthermore, I felt like it was adding a second type of completion paradigm while ignoring the original one, i.e. the minibuffer—again, an offense against simplicity.

What I have in this section is a few simple tweaks and built-in ways to complete terms while typing text in a buffer. I think that, for most cases, the minibuffer can be used effectively to perfom in-buffer completion: start from my Minibuffer essentials and Icomplete to understand how these fit into the broader framework.

I yanked contrib/completing-read-in-region straight from Omar Antolín Camarena's dotemacs repo. It provides a minibuffer-completion interface to the standard tab-completion mechanism (also check what I have on the topic of Tabs, indentation, and the TAB key). It does the same for dabbrev-completion (see next section).

(use-package emacs
  (defun contrib/completing-read-in-region (start end collection &optional predicate)
    "Prompt for completion of region in the minibuffer if non-unique.
Use as a value for `completion-in-region-function'."
    (if (and (minibufferp) (not (string= (minibuffer-prompt) "Eval: ")))
        (completion--in-region start end collection predicate)
      (let* ((initial (buffer-substring-no-properties start end))
             (limit (car (completion-boundaries initial collection predicate "")))
             (all (completion-all-completions initial collection predicate
                                              (length initial)))
             (completion (cond
                          ((atom all) nil)
                          ((and (consp all) (atom (cdr all)))
                           (concat (substring initial 0 limit) (car all)))
                          (t (completing-read
                              "Completion: " collection predicate t initial)))))
        (if (null completion)
            (progn (message "No completion") nil)
          (delete-region start end)
          (insert completion)

  (setq completion-in-region-function #'contrib/completing-read-in-region)
  :bind (:map minibuffer-local-completion-map
              ("<tab>" . minibuffer-force-complete))) Dabbrev (dynamic word completion)

This is Emacs' own approach to dynamic/arbitrary text completion inside the buffer: "dynamic abbreviation" or else dabbrev. This mechanism works by reading all text before point to find a suitable match. Different scenaria determine whether it should also look forward and in other buffers.

In essence, Dabbrev can help you type again what you already have. It will not draw findings from some knowledge bank, nor will it try to read your mind (though Emacs will definitely have an M-x conduit to such a technology).

With dabbrev-expand we make an attempt to complete the text at point. Repeated invocations will cycle through the candidates. No feedback is provided, much in the same way yanking from the kill-ring works. Whereas dabbrev-completion benefits from minibuffer interactivity, courtesy of contrib/completing-read-in-region that was defined in the section right above.

The dabbrev-abbrev-char-regexp is configured to match both regular words and symbols (e.g. words separated by hyphens). This makes it suitable both for code and ordinary language.

While the dabbrev-abbrev-skip-leading-regexp is instructed to also expand words and symbols that start with any of these: $, *, /, =. This regexp may be expanded in the future, but the idea is to be able to perform completion in contexts where the known word/symbol is preceded by a special characters. For example, in the org-mode version of this document, all inline code must be placed between the equals sign. So now typing the =, then a letter, will still allow me to expand text based on that input.

To check what I have on regular expressions, see further below my configurations and documentation for re-builder and visual-regexp.

(use-package dabbrev
  :after (minibuffer icomplete icomplete-vertical) ; read those as well
  (setq dabbrev-abbrev-char-regexp "\\sw\\|\\s_")
  (setq dabbrev-abbrev-skip-leading-regexp "\\$\\|\\*\\|/\\|=")
  (setq dabbrev-backward-only nil)
  (setq dabbrev-case-distinction nil)
  (setq dabbrev-case-fold-search t)
  (setq dabbrev-case-replace nil)
  (setq dabbrev-check-other-buffers t)
  (setq dabbrev-eliminate-newlines nil)
  (setq dabbrev-upcase-means-case-search t)
  :bind (("M-/" . dabbrev-expand)
         ("C-M-/" . dabbrev-completion)
         ("s-/" . dabbrev-completion))) Skeletons and abbreviations

NOTE 2020-05-26: Pending review.

This section stores all the "skeletons" I define. These are snippets of text, typically templates or code statements, that are meant to speed up typing. While abbreviations are shorter versions of terms that automatically expand into what they correspond to. I combine skeletons with abbreviations.

Please note that these will be very simplistic at first. I am aware that they can be abstracted using elisp—need to learn more on that front. Also note that wherever you see " _ " it signifies the position of the cursor after the skeleton has been inserted.

(use-package abbrev
  (setq abbrev-file-name "~/.emacs.d/abbrevs")
  (setq only-global-abbrevs nil)

  ;; simple skeletons ;;
  (define-skeleton protesilaos-com-skeleton
    "Adds a link to my website while prompting for a possible
    "Insert website extension: "
    "" str "")
  (define-abbrev global-abbrev-table "meweb"
    "" 'protesilaos-com-skeleton)

  (define-skeleton protesilaos-gitlab-skeleton
    "Adds a link to my GitLab account while prompting for a
  possible extension.  Makes it easy to link to my various git
    "Website extension: "
    "" str "")
  (define-abbrev global-abbrev-table "megit"
    "" 'protesilaos-gitlab-skeleton)
  :hook ((text-mode-hook . abbrev-mode)
         (git-commit-mode-hook . abbrev-mode)))

3.2 Configurations for—or extensions to—built-in search commands

These are meant to enhance the functionality of tools that are already shipped with Emacs.

3.2.1 Isearch enhancements

The built-in search mechanism is a thing of beauty: minimal in its presentation, powerful in its applications.

I use isearch all the time for quick navigation, either to a visible part of the buffer or to some specific string I am aware of. It also is essential when used in the context of a keyboard macro, as demonstrated in my video about Isearch powers in keyboard macros (2020-01-21).

Run C-h k C-s to get an awesome help menu with all the extra keys you can use with isearch. These are the ones I use the most:

Key chord Description
C-s C-w Search char or word at point
M-s . Similar, but broader match
M-s o Run `occur' on regexp
M-s h r Highlight regexp
M-s h u Undo the highlight
C-s M-r Toggle regexp search
M-% Run `query-replace'
C-M-% `query-replace-regexp'

Many commands can be invoked while running isearch to operate on the current match. For example, C-s SEARCH M-s o will produce an "Occur" buffer with the contents of the search terms. Absolutely great!

With regard to the replace commands, note that you can use them on the active region. Furthermore, you do not need to confirm each action, but can instead type ! to answer "yes" to all possible replacements. Better only use this while having already limited the results to the active region, to some specialised editable buffer like the one of occur, or by using Emacs' narrowing techniques, such as narrow-to-region.

In the package declaration below, the combined effect of the variables for whitespace is a valuable hack: typing a space is the same as inserting a wildcard, which is much more useful as far as I am concerned. A single space represents a wildcard that matches items in a non-greedy fashion. This affects regular searches (the standard C-s and C-r). The regexp-sensitive functions C-M-s and C-M-r remain in tact. You can always toggle whitespace matching behaviour while performing a search, with M-s SPC (revert back to just literal spaces).

Now on to some custom functions, all of which are derived from the source code of isearch (do it with M-x find-library RET isearch RET). Here is an overview of what goes into this package declaration.

Mark isearch match
Replaces the default mark command following a successful search. I prefer to mark the match. This can be then used to insert multiple cursors (if you are using it), kill the region, etc. Besides, it is always possible to mark a region from point to search string by running C-x C-x following a successful search.
Move to opposite end
Isearch places the point at either the beginning or the end of the match, depending on the direction it is moving in. For single words or balanced expressions this is not an issue because you can always confirm a search by using a motion key (so, for example, move to the end of the matching word with M-f). There are, however, matches that are not limited to such boundaries. For those cases moving to the opposite end might require multiple key presses, which is bad when trying to record an efficient keyboard macro. prot/isearch-other-end addresses the issue. It is bound to C-RET while running a successful search. The direct inspiration is this forum answer. Note though that you can achieve the same result by changing the direction the search is moving towards with C-s or C-r (though I still prefer my minor addition).
Delete non-match
The built-in method to remove the entirety of a mismatched input is to hit C-g following a failed search. This keeps the valid part and allows you to continue searching. However, I find that the choice of key binding can prove problematic, since C-g also exits a standard/successful search. As such, the simple function prot/isearch-abort is designed to remove the entirety of a mismatch, just by hitting backspace (aka DEL). For valid searches, backspace functions exactly as expected, deleting one character at a time. Note, though, that it is no longer possible to delete part of a failed search, just by hitting backspace: you can still rely on C-M-d for that (or edit the input with M-e).
Replace symbol at point
Combine the built-in functions of isearch-forward-symbol-at-point and isearch-query-replace-regexp into a single command that is bound to the key chord M-s %. Simple and super effective (pro tip: hit ! to answer "yes" to all possible matches, which is possible in all cases where Emacs asks you for multiple confirmations).

The variables about the lazy count that are commented as "Emacs 27.1" effectively supersede the functionality of anzu, a package I once used.

(use-package isearch
  (setq search-highlight t)
  (setq search-whitespace-regexp ".*?")
  (setq isearch-lax-whitespace t)
  (setq isearch-regexp-lax-whitespace nil)
  (setq isearch-lazy-highlight t)
  ;; All of the following variables were introduced in Emacs 27.1.
  (setq isearch-lazy-count t)
  (setq lazy-count-prefix-format nil)
  (setq lazy-count-suffix-format " (%s/%s)")
  (setq isearch-yank-on-move 'shift)
  (setq isearch-allow-scroll 'unlimited)

  (defun prot/isearch-mark-and-exit ()
    "Mark the current search string and exit the search."
    (push-mark isearch-other-end t 'activate)
    (setq deactivate-mark nil)

  (defun prot/isearch-other-end ()
    "End current search in the opposite side of the match.
Particularly useful when the match does not fall within the
confines of word boundaries (e.g. multiple words)."
    (when isearch-other-end
      (goto-char isearch-other-end)))

  (defun prot/isearch-abort-dwim ()
    "Delete failed `isearch' input, single char, or cancel search.

This is a modified variant of `isearch-abort' that allows us to
perform the following, based on the specifics of the case: (i)
delete the entirety of a non-matching part, when present; (ii)
delete a single character, when possible; (iii) exit current
search if no character is present and go back to point where the
search started."
    (if (eq (length isearch-string) 0)
      (while (or (not isearch-success) isearch-error)

  (defun prot/isearch-query-replace-symbol-at-point ()
    "Run `query-replace-regexp' for the symbol at point."

  :bind (("M-s M-o" . multi-occur)
         ("M-s %" . prot/isearch-query-replace-symbol-at-point)
         :map minibuffer-local-isearch-map
         ("M-/" . isearch-complete-edit)
         :map isearch-mode-map
         ("C-g" . isearch-cancel)       ; instead of `isearch-abort'
         ("M-/" . isearch-complete)
         ("C-SPC" . prot/isearch-mark-and-exit)
         ("<backspace>" . prot/isearch-abort-dwim)
         ("<C-return>" . prot/isearch-other-end)))

3.2.2 Occur (replace.el)

These are a few stylistic tweaks for the "Occur" buffer (M-x occur). The faces are part of my Modus themes (see relevant section).

(use-package replace
  (setq list-matching-lines-jump-to-current-line t)
  ;; See my "Modus themes" for these inherited faces
  (setq list-matching-lines-buffer-name-face
        '(:inherit modus-theme-intense-neutral :weight bold))
  (setq list-matching-lines-current-line-face
        '(:inherit modus-theme-special-mild))
  :hook (occur-mode-hook . hl-line-mode))

3.2.3 Regular expressions: re-builder and visual-regexp

To learn more about regular expressions, read the relevant pages in the official manual. Assuming you have this installed properly on your system, run C-h r i regexp to get to the starting chapter.

Also watch my ~35 minute-long primer on Emacs regexp (2020-01-23).

Emacs offers a built-in package for practising regular expressions. By default, re-builder uses Emacs-style escape notation, in the form of double backslashes. You can switch between the various styles by using C-c TAB inside of the regexp builder's buffer. I choose to keep this style as the default. Other options are string and rx.

(use-package re-builder
  (setq reb-re-syntax 'read))

Another option (though the two are not mutually exclusive) is to use the third-party package visual-regexp. This one is meant as a drop-in replacement for query-replace (and the regexp variant). I prefer not to use it that way, but only invoke it via M-x when I need to test a regular expression that I would then replace with something else. The major upside of this tool is that it highlights groups individually and offers a live preview of the replacement, making it absolutely great when dealing with complex sets of regexp constructs.

(use-package visual-regexp
  (setq vr/default-replace-preview nil)
  (setq vr/match-separator-use-custom-face t))

3.2.4 wgrep (writable grep)

With wgrep we can directly edit the results of a grep and save the changes to all affected buffers. In principle, this is the same as what the built-in occur offers. We can use it to operate on a list of matches by leveraging the full power of Emacs' editing capabilities (e.g. keyboard macros, multiple cursors…).

(use-package wgrep
  (setq wgrep-auto-save-buffer t)
  (setq wgrep-change-readonly-file t))

3.2.5 ripgrep (rg.el)

This is a package that allows us to interface with the external command line program called "ripgrep". My video demo of rg.el (2020-03-25) covers the main features of this tool.

What I find particularly appealing about rg.el is that it follows the interface paradigms of built-in Emacs functions, such as grep or occur. With regard to the latter, it even uses the same key to convert the results' buffer into an editable one: e (the ability to write changes is provided by the wgrep package that I define right above).

Furthermore, rg.el interfaces with ibuffer, another built-in package, to list saved searches (see my prot/rg-save-search-as-name in the package declaration below). Saved searches are regular buffers. You can switch to any of them the normal way.

While inside of an rg.el buffer, hit m to produce a transient menu from where you can refine your search. This works just like magit. In addition, you can consult the universal C-h m for documentation concerning the major mode you are in.

Concerning the key bindings for navigating the results buffer, I find that the standard motions should retain their general function, while moving between file headings can be done with M-{n,p}.

rg.el is designed in such a way that it offers useful functionality without depending on a particular completion framework (e.g. Ivy, Helm). I consider this an advantage, especially when combined with the overall alignment of this package with standard Emacs tools.

Also see my configurations for project-related commands.

(use-package rg
  :after wgrep
  (setq rg-group-result t)
  (setq rg-hide-command t)
  (setq rg-show-columns nil)
  (setq rg-show-header t)
  (setq rg-custom-type-aliases nil)
  (setq rg-default-alias-fallback "all")

  (rg-define-search prot/rg-vc-or-dir
    "RipGrep in project root or present directory."
    :query ask
    :format regexp
    :files "everything"
    :dir (let ((vc (vc-root-dir)))
           (if vc
               vc                         ; search root project dir
             default-directory))          ; or from the current dir
    :confirm prefix
    :flags ("--hidden -g !.git"))

  (rg-define-search prot/rg-ref-in-dir
    "RipGrep for thing at point in present directory."
    :query point
    :format regexp
    :files "everything"
    :dir default-directory
    :confirm prefix
    :flags ("--hidden -g !.git"))

  (defun prot/rg-save-search-as-name ()
    "Save `rg' buffer, naming it after the current search query.

This function is meant to be mapped to a key in `rg-mode-map'."
    (let ((pattern (car rg-pattern-history)))
      (rg-save-search-as-name (concat "«" pattern "»"))))

  :bind (("M-s g" . prot/rg-vc-or-dir)
         ("M-s r" . prot/rg-ref-in-dir)
         :map rg-mode-map
         ("s" . prot/rg-save-search-as-name)
         ("C-n" . next-line)
         ("C-p" . previous-line)
         ("M-n" . rg-next-file)
         ("M-p" . rg-prev-file)))

4 Directory, buffer, window management

4.1 Dired (directory editor, file manager)

The directory editor abbreviated as "Dired" (which I pronounce like "tired", "mired", etc.) is a built-in tool that performs file management operations inside of an Emacs buffer. It is simply superb! I use it daily for a number of tasks.

You can interactively copy, move (rename), symlink, delete files and directories, handle permissions, compress or extract archives, run shell commands, combine Dired with regular editing capabilities as part of a keyboard macro, search[+replace] across multiple files, encrypt/decrypt files, and more. Combine that with the possibility of matching items with regular expressions or creating an editable Dired buffer and you have everything you need to maximise your productivity.

Check some of my videos:

4.1.1 Base settings for Dired

The options here are meant to do the following:

  • Copy and delete recursively. Do not ask about it.
  • Search only file names while point is there, else the rest (useful when using the detailed view).
  • Deletion sends items to the system's Trash, making it safer than the standard rm.
  • Prettify output. Sort directories first. Show dotfiles first. Omit implicit directories (the single and double dots). Use human-readable size units. There are also options for tweaking the behaviour of find-name-dired, in the same spirit. To learn everything about these switches, you need to read the manpage of ls. You can do it with M-x man RET ls.
  • Hide all the verbose details by default (permissions, size, etc.). These can easily be toggled on using the left parenthesis ( inside a dired buffer. Also enable highlighting of the current line, which makes it even easier to spot the current item (I do not enable this globally, because I only want it for per-line interfaces, such as Dired's, but not for per-character ones, such as text editing).
  • While having two dired buffers side-by-side, the rename and copy operations of one are easily propagated to the other. Dired is smart about your intentions and uses the adjacent Dired buffer's path as a prefix when performing such actions.
  • For Emacs 27.1, Dired can automatically create destination directories for its copy and remove operations. So you can, for example, rename file to /non-existent-path/file and you will get what you want right away.
  • For Emacs 27.1, renaming a file of a version-controlled repository (git) will be done using the appropriate VC mechanism.
  • Let the relevant find commands use case-insensitive names.
  • Enable asynchronous mode. This is subject to change, as I need to test it a bit more.

Note that dired-listing-switches and find-ls-option are configured to show hidden directories and files before their non-hidden counterparts. If you want to reverse this order, you must append the -X option (such as -AFXhlv --group-directories-first).

The commands with the contrib/ prefix in dired-aux are copied from the Emacs configurations of Omar Antolín Camarena. They let you insert the path of a bookmarked directory while performing an action such as copying and renaming/moving a file. While my prot/dired-fd-* are conceptually similar to functions in projects and directory trees as well as the ones in the Ibuffer section.

(use-package dired
  (setq dired-recursive-copies 'always)
  (setq dired-recursive-deletes 'always)
  (setq delete-by-moving-to-trash t)
  (setq dired-listing-switches
        "-AGFhlv --group-directories-first --time-style=long-iso")
  (setq dired-dwim-target t)
  ;; Note that the the syntax for `use-package' hooks is controlled by
  ;; the `use-package-hook-name-suffix' variable.  The "-hook" suffix is
  ;; not an error of mine.
  :hook ((dired-mode-hook . dired-hide-details-mode)
         (dired-mode-hook . hl-line-mode)))

(use-package dired-aux
  (setq dired-isearch-filenames 'dwim)
  ;; The following variables were introduced in Emacs 27.1
  (setq dired-create-destination-dirs 'ask)
  (setq dired-vc-rename-file t)

  ;; TODO defmacro to avoid duplication of code in `fd' functions
  ;; TODO how can a defmacro produce named functions that are then
  ;; mapped to keys?
  (defun prot/dired-fd-dirs (&optional arg)
    "Search for directories in VC root or PWD.
With \\[universal-argument] put the results in a `dired' buffer.
This relies on the external 'fd' executable."
    (interactive "P")
    (let* ((vc (vc-root-dir))
           (dir (expand-file-name (if vc vc default-directory)))
           (regexp (read-regexp
                    (concat "Subdirectories matching REGEXP in "
                            (propertize dir 'face 'bold)
                            ": ")))
           (names (process-lines "fd" "-i" "-H" "-a" "-t" "d" "-c" "never" regexp dir))
           (buf "*FD Dired*"))
      (if names
          (if arg
              (dired (cons (generate-new-buffer-name buf) names))
            (icomplete-vertical-do ()
               (completing-read (concat
                                 "Files or directories matching "
                                 (propertize regexp 'face 'success)
                                 (format " (%s)" (length names))
                                 ": ")
                                names nil t)))))
      (user-error (concat "No matches for " "«" regexp "»" " in " dir))))

  (defun prot/dired-fd-files-and-dirs (&optional arg)
    "Search for files and directories in VC root or PWD.
With \\[universal-argument] put the results in a `dired' buffer.
This relies on the external 'fd' executable."
    (interactive "P")
    (let* ((vc (vc-root-dir))
           (dir (expand-file-name (if vc vc default-directory)))
           (regexp (read-regexp
                    (concat "Files and dirs matching REGEXP in "
                            (propertize dir 'face 'bold)
                            ": ")))
           (names (process-lines "fd" "-i" "-H" "-a" "-t" "d" "-t" "f" "-c" "never" regexp dir))
           (buf "*FD Dired*"))
      (if names
          (if arg
              (dired (cons (generate-new-buffer-name buf) names))
            (icomplete-vertical-do ()
               (completing-read (concat
                                 "Files and directories matching "
                                 (propertize regexp 'face 'success)
                                 (format " (%s)" (length names))
                                 ": ")
                                names nil t)))))
      (user-error (concat "No matches for " "«" regexp "»" " in " dir))))

  (defun contrib/cdb--bookmarked-directories ()
    (cl-loop for (name . props) in bookmark-alist
             for fn = (cdr (assq 'filename props))
             when (and fn (string-suffix-p "/" fn))
             collect (cons name fn)))

  (defun contrib/cd-bookmark (bm)
    "Insert the path of a bookmarked directory."
     (list (let ((enable-recursive-minibuffers t))
              "Directory: " (contrib/cdb--bookmarked-directories) nil t))))
    (when (minibufferp)
      (delete-region (minibuffer-prompt-end) (point-max)))
    (insert (cdr (assoc bm (contrib/cdb--bookmarked-directories)))))
  :bind (("M-s d" .  prot/dired-fd-dirs)
         ("M-s z" . prot/dired-fd-files-and-dirs)
         :map dired-mode-map
         ("C-+" . dired-create-empty-file)
         ("M-s f" . nil)
         :map minibuffer-local-filename-completion-map
         ("C-c d" . contrib/cd-bookmark)))

(use-package find-dired
  :after dired
  (setq find-ls-option
        '("-ls" . "-AGFhlv --group-directories-first --time-style=long-iso"))
  (setq find-name-arg "-iname"))

(use-package async :ensure)

(use-package dired-async
  :after (dired async)
  :hook (dired-mode-hook . dired-async-mode))

Pro tip while renaming or copying a file, M-n will return its original name, thus allowing you to easily {pre,ap}pend to it. This leverages an intriguing concept of Emacs' design called "future history" (because M-p goes back to your previous entries). The notion of the future history, when applied, is basically an educated guess of what the user would want to do in the current context, given that they are not searching through their previous actions. TODO extend core Dired [1/3] DONE front-end for `fd' executable TODO better handling of `xdg-open' TODO combine arbitrary dirs with selection

4.1.2 Narrowed dired

The standard way to produce a Dired buffer with only a handful of files is to mark them, either manually or with % m, then toggle the mark with t, and then remove (just from the view) everything with k. This will leave you with only the files you need to focus on. With g you get back to the unfiltered listing.

For dynamic filtering, use this package. It offers several commands, but I find that I only ever need to narrow by a regular expression (check the source code for all of them M-x find-library dired-narrow).

(use-package dired-narrow
  :after dired
  (setq dired-narrow-exit-when-one-left t)
  (setq dired-narrow-enable-blinking t)
  (setq dired-narrow-blink-time 0.3)
  :bind (:map dired-mode-map
         ("/" . dired-narrow-regexp)))

4.1.3 wdired (writable dired)

This is the editable state of a dired buffer. You can access it with C-x C-q. Write changes to files or directories, as if it were a regular buffer, then confirm them with C-c C-c.

  • While in writable state, allow the changing of permissions.
  • While renaming a file, any forward slash is treated like a directory and is created directly upon successful exit.
(use-package wdired
  :after dired
  :commands wdired-change-to-wdired-mode
  (setq wdired-allow-to-change-permissions t)
  (setq wdired-create-parent-directories t))

4.1.4 peep-dired (file previews including images)

By default, dired does not show previews of files, while image-dired is intended for a different purpose. We just want to toggle the behaviour while inside a regular dired buffer.

(use-package peep-dired
  :after dired
  (setq peep-dired-cleanup-on-disable t)
  (setq peep-dired-enable-on-directories nil)
  (setq peep-dired-ignored-extensions
        '("mkv" "webm" "mp4" "mp3" "ogg" "iso"))
  :bind (:map dired-mode-map
              ("P" . peep-dired)))

4.1.5 image-dired (image thumbnails and previews)

This tool offers facilities for generating thumbnails out of a selection of images and displaying them in a separate buffer. An external program is needed for converting the images into thumbnails: imagemagick. Other useful external packages are optipng and sxiv. The former is for operating on PNG files, while the latter is a lightweight image viewer.

I feel this process is a bit cumbersome and can be very slow if you try to generate lots of images at once. The culprit is the image converter.

(use-package image-dired
  (setq image-dired-external-viewer "xdg-open")
  (setq image-dired-thumb-size 80)
  (setq image-dired-thumb-margin 2)
  (setq image-dired-thumb-relief 0)
  (setq image-dired-thumbs-per-row 4)
  :bind (:map image-dired-thumbnail-mode-map
              ("<return>" . image-dired-thumbnail-display-external)))

4.1.6 dired-subtree (tree-style view/navigation)

Tree-style navigation means that the subdirectories of the current Dired buffer can be expanded and contracted in place. It then is possible to perform the same kind of folding on their subdirectories, and so on.

This is, in my opinion, a far more intuitive interaction than the default way of inserting subdirectories in the current buffer below their parent (type i over the target dir). There still are uses for that technique, but tree-style navigation is easier for day-to-day tasks.

What I have here:

  • The tab key will expand or contract the subdirectory at point.
  • C-TAB will behave just like org-mode handles its headings: hit it once to expand a subdir at point, twice to do it recursively, thrice to contract the tree.
  • I also have Shift-TAB for contracting the subtree when the point is inside of it.
(use-package dired-subtree
  :after dired
  (setq dired-subtree-use-backgrounds nil)
  :bind (:map dired-mode-map
              ("<tab>" . dired-subtree-toggle)
              ("<C-tab>" . dired-subtree-cycle)
              ("<S-iso-lefttab>" . dired-subtree-remove)))

4.1.7 dired-x (extra Dired functions)

These are some additional features that are shipped with Emacs. The one I need the most is dired-jump and its "other window" variant. These are among my favourite commands. They will always take you to the directory that contains the current buffer.

'Jumping' works even when you are inside buffers that do not visit files, such as Magit, Diff, or Eshell. This is its most valuable quality! Edit a file then proceed to do some file management, then invoke previous-buffer or winner-undo to go back to where you were (I have a few key bindings for those in the Window configuration section). Everything happens naturally. Emacs' interconnectedness at its best!

I keep dired-clean-confirm-killing-deleted-buffers to t as a safety mechanism: if a file is ever deleted by accident I can use its buffer to restore it (never happened in practice).

With regard to binding keys, I choose to handle things myself. There has never been a case where I had to run info or man inside of a directory listing and wished there was some keyboard shortcut readily available.

While in dired-mode, if you need to open all marked files at once, you can hit F. It calls dired-do-find-marked-files.

(use-package dired-x
  :after dired
  (setq dired-clean-up-buffers-too t)
  (setq dired-clean-confirm-killing-deleted-buffers t)
  (setq dired-x-hands-off-my-keys t)
  (setq dired-bind-man nil)
  (setq dired-bind-info nil)

  (defun prot/dired-jump-extra (&optional arg)
    "Switch directories comprising context and bookmarks.
    (interactive "P")
    (let* ((vc (vc-root-dir))
           (buf-name (buffer-file-name))
           (path (if buf-name
           (file (abbreviate-file-name path))
           (bookmarks (mapcar (lambda (b)
                                (cdr b))
           (collection (append bookmarks
                             (list (file-name-directory file)
                                   (when vc vc))))
           (files (cl-remove-if (lambda (f)
                                  (eq f nil))
      (icomplete-vertical-do ()
         (completing-read "Jump to context or bookmark: " files nil t)))))

  :bind (("C-c j" . prot/dired-jump-extra)
         ("C-x C-j" . dired-jump)
         ("s-j" . dired-jump)
         ("C-x 4 C-j" . dired-jump-other-window)
         ("s-J" . dired-jump-other-window)))

4.1.8 diredfl (more dired colours)

This package defines a few more colours for Dired, especially while in the detailed view. My themes support it, as well as a ton of other packages (see the section on my Modus themes).

(use-package diredfl
  :hook (dired-mode-hook . diredfl-mode))

4.1.9 dired-like view for the trash directory

trashed applies the principles of dired to the management of the user's filesystem trash. Use C-h m to see the docs and keybindings for its major mode.

Basically, its interaction model is as follows:

  • m to mark for some deferred action, such as D to delete, R to restore.
  • t to toggle the status of all items as marked. Use this without marks to m (mark) all items, then call a deferred action to operate on them.
  • d to mark for permanent deletion.
  • r to mark for restoration.
  • x to execute these special marks.
(use-package trashed
  (setq trashed-action-confirmer 'y-or-n-p)
  (setq trashed-use-header-line t)
  (setq trashed-sort-key '("Date deleted" . t))
  (setq trashed-date-format "%Y-%m-%d %H:%M:%S"))

4.2 Working with buffers

4.2.1 Unique names for buffers

These settings make it easier to work with multiple buffers. When two buffers have the same name, Emacs will try to disambiguate them by displaying their unique path inside angled brackets. With the addition of uniquify-strip-common-suffix it will also remove the part of the file system path they have in common.

All such operations are reversed once an offending buffer is removed from the list, allowing Emacs to revert to the standard of displaying only the buffer's name.

(use-package uniquify
  (setq uniquify-buffer-name-style 'post-forward-angle-brackets)
  (setq uniquify-strip-common-suffix t)
  (setq uniquify-after-kill-buffer-p t)) TODO make uniquify better for Help, Info buffers

4.2.2 Ibuffer and extras (dired-like buffer list manager)

ibuffer is a built-in replacement for list-buffers that allows for fine-grained control over the buffer list. For this reason I bind it to C-x C-b.

Overview of its features:

  • mark and delete buffers same way you do in dired (see the previous sections on dired (directory editor, file manager));
  • mark by a predicate, such as name, major mode, etc.;
  • sort buffers by name, filesystem path, major mode, size;
  • run occur on the marked buffers (remember: Occur produces a buffer that you can edit once you enable the editable state with e);
  • run query-replace on marked buffers or its regular-expression-aware equivalent.

Run the universal help command for major mode documentation (C-h m) while inside ibuffer to get a detailed list of all available commands and their key bindings.

With regard to the following package declaration, these are my tweaks to the default behaviour and presentation:

  • Prompt for confirmation only when deleting a modified buffer.
  • Hide the summary.
  • Do not open on the other window (not focused window).
  • Do not show empty filter groups.
  • Do not cycle movements. So do not go to the top when moving downward at the last item on the list.

Also watch my introduction to Ibuffer (2020-04-02).

Now some extras after I published that video: prot/buffers-major-mode and prot/buffers-vc-root will produce a filtered list based on the current buffer's major mode or root version-control directory respectively. Their standard mode of interaction is through minibuffer completions (see Minibuffer essentials and Icomplete). With a prefix argument C-u they switch to a dedicated Ibuffer view instead.

For those two I received help from Omar Antolín Camarena with regard to the use of read-buffer and the lambda passed to it: my prototype was using the generic completing-read with cl-remove-if-not for filtering the list of candidates (all possible errors are my own). The upside of using Omar's approach is that it informs other tools that this kind of completion concerns buffers, so they can adapt accordingly. This is done, for example, by Omar's powerful embark package (see section on Embark (actions for completion candidates)).

I consider these two commands to be conceptually related to what I have defined in Completion for projects and directory trees. The key chords follow the same pattern, with the M-s prefix being considered an entry point for "advanced search" methods, inspired by the default for occur (M-s o) as well as a few others.

(use-package ibuffer
  (setq ibuffer-expert t)
  (setq ibuffer-display-summary nil)
  (setq ibuffer-use-other-window nil)
  (setq ibuffer-show-empty-filter-groups nil)
  (setq ibuffer-movement-cycle nil)
  (setq ibuffer-default-sorting-mode 'filename/process)
  (setq ibuffer-use-header-line t)
  (setq ibuffer-default-shrink-to-minimum-size nil)
  (setq ibuffer-formats
        '((mark modified read-only locked " "
                (name 30 30 :left :elide)
                " "
                (size 9 -1 :right)
                " "
                (mode 16 16 :left :elide)
                " " filename-and-process)
          (mark " "
                (name 16 -1)
                " " filename)))
  (setq ibuffer-saved-filter-groups nil)

  (defun prot/buffers-major-mode (&optional arg)
    "Select buffers that match the current buffer's major mode.
With \\[universal-argument] produce an `ibuffer' filtered
accordingly.  Else use standard completion."
    (interactive "P")
    (let* ((major major-mode)
           (prompt "Buffers for ")
           (mode-string (format "%s" major))
           (mode-string-pretty (propertize mode-string 'face 'success)))
      (if arg
          (ibuffer t (concat "*" prompt mode-string "*")
                   (list (cons 'used-mode major)))
          (concat prompt mode-string-pretty ": ") nil t
          (lambda (pair) ; pair is (name-string . buffer-object)
            (with-current-buffer (cdr pair) (derived-mode-p major))))))))

  (defun prot/buffers-vc-root (&optional arg)
    "Select buffers that match the present `vc-root-dir'.
With \\[universal-argument] produce an `ibuffer' filtered
accordingly.  Else use standard completion.

When no VC root is available, use standard `switch-to-buffer'."
    (interactive "P")
    (let* ((root (vc-root-dir))
           (prompt "Buffers for VC ")
           (vc-string (format "%s" root))
           (vc-string-pretty (propertize vc-string 'face 'success)))
      (if root
          (if arg
              (ibuffer t (concat "*" prompt vc-string "*")
                       (list (cons 'filename (expand-file-name root))))
              (concat prompt vc-string-pretty ": ") nil t
              (lambda (pair) ; pair is (name-string . buffer-object)
                (with-current-buffer (cdr pair) (string= (vc-root-dir) root))))))
        (call-interactively 'switch-to-buffer))))

  :hook (ibuffer-mode-hook . hl-line-mode)
  :bind (("M-s b" . prot/buffers-major-mode)
         ("M-s v" . prot/buffers-vc-root)
         ("C-x C-b" . ibuffer)
         :map ibuffer-mode-map
         ("* f" . ibuffer-mark-by-file-name-regexp)
         ("* g" . ibuffer-mark-by-content-regexp) ; "g" is for "grep"
         ("* n" . ibuffer-mark-by-name-regexp)
         ("s n" . ibuffer-do-sort-by-alphabetic)  ; "sort name" mnemonic
         ("/ g" . ibuffer-filter-by-content))) Ibuffer integration with VC (version control framework)

This package offers a few functions for operating on ibuffer items based on their corresponding version control data.

To me the most common case is to establish filter groups on a per-project basis with ibuffer-vc-set-filter-groups-by-vc-root. This makes it easier to get an overview of where each buffer belongs, especially in cases where you have similar looking names.

In terms of functionality, filter groups allow for per-group actions:

  • Move between group headings with M-n and M-p.
  • Toggle the visibility of the group with RET, while the point is over the heading.
  • With point over a heading, m will mark all its buffers, while d will mark them for deletion (confirm the latter with x, same as with dired).

The above granted, I prefer to create such filter groups manually via a convenient key binding, rather than calling the function through a relevant hook.

Make sure to also read the section on version control tools.

(use-package ibuffer-vc
  :after (ibuffer vc)
  :bind (:map ibuffer-mode-map
              ("/ V" . ibuffer-vc-set-filter-groups-by-vc-root)
              ("/ <deletechar>" . ibuffer-clear-filter-groups)))

4.2.3 Scratch buffers per-major-mode

This package will produce a buffer that matches the major mode of the one you are currently in. Use it with M-x scratch. Doing that with a prefix argument (C-u) will prompt for a major mode instead. Simple yet super effective!

The prot/scratch-buffer-setup simply adds some text in the buffer and renames it appropriately for the sake of easier discovery.

(use-package scratch
  (defun prot/scratch-buffer-setup ()
    "Add contents to `scratch' buffer and name it accordingly."
    (let* ((mode (format "%s" major-mode))
           (string (concat "Scratch buffer for: " mode "\n\n")))
      (when scratch-buffer
          (insert string)
          (goto-char (point-min))
          (comment-region (point-at-bol) (point-at-eol)))
        (next-line 2))
      (rename-buffer (concat "*Scratch for " mode "*") t)))
  :hook (scratch-create-buffer-hook . prot/scratch-buffer-setup)
  :bind ("C-c s" . scratch))

4.3 Window configuration

I believe that Emacs' true power lies in its buffer management rather than its multiplexing. The latter becomes inefficient at scale, since it tries to emulate the limitations of the real world, namely, the placement of things on a desk.

By leveraging the power of the computer, we can use search methods to easily reach any item. There is no need to remain confined to the idea of a finite space (screen real estate) that needs to be carefully managed.

That granted, Emacs' multiplexing can be turned into a powerhouse as well, covering everything from window placement rules, to the recording of history and layouts, as well as directional or direct window navigation.

The prot/window-single-toggle is based on Pierre Neidhardt's windower.

(use-package emacs
  (defvar prot/window-configuration nil
    "Current window configuration.
Intended for use by `prot/window-monocle'.")

  (defun prot/window-single-toggle ()
    "Toggle between multiple windows and single window.
This is the equivalent of maximising a window.  Tiling window
managers such as DWM, BSPWM refer to this state as 'monocle'."
    (if (one-window-p)
        (when prot/window-configuration
          (set-window-configuration prot/window-configuration))
      (setq prot/window-configuration (current-window-configuration))
  :bind ("s-m" . prot/window-single-toggle))

4.3.1 Window rules and basic tweaks

The display-buffer-alist and all other functions grouped together with prot/window-dired-vc-root-left are considered experimental and subject to review. The former is intended as a rule-set for controlling the display of windows. While the latter serves as a series of tangible examples of passing certain rules programmatically, in combination with a few relevant extras. The objective is to create a more intuitive workflow where targeted buffer groups or types are always shown in a given location, on the premise that predictability improves usability.

For each buffer action in display-buffer-alist we can define several functions for selecting the appropriate window. These are executed in sequence, but my usage thus far suggests that a simpler method is just as effective for my case.

Everything pertaining to buffer actions is documented at length in the GNU Emacs Lisp Reference Manual, currently corresponding to version 26.3. Information can also be found via C-h f display-buffer and, for my settings, C-h f display-buffer-in-side-window.

With regard to the contents of the :bind keyword of the window library, most key combinations are complementary to the standard ones, such as C-x 1 becoming s-1, C-x o turning into s-o and the like. They do not replace the defaults: they just provide more convenient access to their corresponding functions. They all involve the Super key, following the norms described in the introductory note on the matter. Concerning the balance-windows-area I find that it is less intrusive than the original balance-windows normally bound to the same C-x +.

For a demo of the display-buffer-alist and the functions that accompany it, watch my video on rules for buffer placement (2020-01-07).

(use-package window
  (setq display-buffer-alist
        '(;; top side window
          ("\\*\\(Flycheck\\|Flymake\\|Package-Lint\\|vc-git :\\).*"
           (window-height . 0.16)
           (side . top)
           (slot . 0)
           (window-parameters . ((no-other-window . t))))
           (window-height . 0.16)
           (side . top)
           (slot . 1)
           (window-parameters . ((no-other-window . t))))
           (window-height . 0.16)
           (side . top)
           (slot . 2)
           (window-parameters . ((no-other-window . t))))
          ;; bottom side window
          ("\\*\\(Output\\|Register Preview\\).*"
           (window-width . 0.16)       ; See the :hook
           (side . bottom)
           (slot . -1)
           (window-parameters . ((no-other-window . t))))
           (window-height . 0.16)
           (side . bottom)
           (slot . 0)
           (window-parameters . ((no-other-window . t))))
           (window-height . 0.16)
           (side . bottom)
           (slot . 1))
          ;; left side window
           (window-width . 0.20)       ; See the :hook
           (side . left)
           (slot . 0)
           (window-parameters . ((no-other-window . t))))
          ;; right side window
           (window-width . 0.25)
           (side . right)
           (slot . 0)
           (window-parameters . ((no-other-window . t)
                                 (mode-line-format . (" "
           (window-width . 0.25)
           (side . right)
           (slot . 1))
          ;; bottom buffer (NOT side window)
  (setq window-combination-resize t)
  (setq even-window-sizes 'height-only)
  (setq window-sides-vertical nil)
  ;; Note that the the syntax for `use-package' hooks is controlled by
  ;; the `use-package-hook-name-suffix' variable.  The "-hook" suffix is
  ;; not an error of mine.
  :hook ((help-mode-hook . visual-line-mode)
         (custom-mode-hook . visual-line-mode))
  :bind (("s-n" . next-buffer)
         ("s-p" . previous-buffer)
         ("s-o" . other-window)
         ("s-2" . split-window-below)
         ("s-3" . split-window-right)
         ("s-0" . delete-window)
         ("s-1" . delete-other-windows)
         ("s-5" . delete-frame)
         ("C-x +" . balance-windows-area)
         ("s-q" . window-toggle-side-windows)))

;; These are all experimental.  Just showcasing the power of passing
;; parameters to windows or frames.
(use-package emacs
  :commands (prot/window-dired-vc-root-left
  (defun prot/window-dired-vc-root-left ()
    "Open root directory of current version-controlled repository
or the present working directory with `dired' and bespoke window
parameters.  This is meant as a proof-of-concept function,
illustrating how to leverage window rules to display a buffer,
plus a few concomitant extras."
    (let ((dir (if (eq (vc-root-dir) nil)
                   (dired-noselect default-directory)
                 (dired-noselect (vc-root-dir)))))
       dir `((side . left)
             (slot . -1)
             (window-width . 0.16)
             (window-parameters . ((no-other-window . t)
                                   (no-delete-other-windows . t)
                                   (mode-line-format . (" "
      (with-current-buffer dir
        (rename-buffer "*Dired-Side*")
        (setq-local window-size-fixed 'width)))
    (with-eval-after-load 'ace-window
      (when (boundp 'aw-ignored-buffers)
        (add-to-list 'aw-ignored-buffers "*Dired-Side*"))))

  (defun prot/make-frame-floating-with-current-buffer ()
    "Display the current buffer in a new floating frame.

This passes certain parameters to the newly created frame:

- use a different name than the default;
- use a graphical frame;
- do not display the minibuffer.

The name is meant to be used by the external rules of my tiling
window manager (BSPWM) to present the frame in a floating state."
    (make-frame '((name . "my_float_window")
                  (window-system . x)
                  (minibuffer . nil))))

  (defun prot/display-buffer-at-bottom ()
    "Move the current buffer to the bottom of the frame.  This is
useful to take a buffer out of a side window.

The window parameters of this function are provided mostly for
didactic purposes."
    (let ((buffer (current-buffer)))
      (with-current-buffer buffer
         buffer `((window-parameters . ((mode-line-format . (" "

4.3.2 Window history (winner-mode)

Winner is a built-in tool that keeps a record of buffer and window layout changes. It then allows us to move back and forth in the history of said changes. I have it enabled by default, while I assign its two main functions to Super and the right/left arrow keys.

(use-package winner
  :hook (after-init-hook . winner-mode)
  :bind ("<s-right>" . winner-redo)
         ("<s-left>" . winner-undo))

Windmove is also built into Emacs. It provides functions for selecting a window in any of the cardinal directions. I use the Vim keys while holding down Super and Meta because other mnemonics-based actions involving just Super or Meta are already occupied.

4.3.3 Tabs for window layouts and buffers (Emacs 27.1)

Starting with version 27.1, Emacs has built-in support for two distinct concepts of "tabs":

  1. Work spaces that contain windows in any given layout.
  2. A list of buffers presented as buttons at the top of the window.

The former, represented by the tab-bar library, is best understood as the equivalent of "virtual desktops", as these are used in most desktop environments or window managers.

The latter, implemented in tab-line, is the same as the tabs you are used to in web browsers. Each buffer is assigned to a single tab. Clicking on the tab takes you to the corresponding buffer.

I do not need the tab-line as I find such tabs to be inefficient at scale. Finding a buffer through search mechanisms is generally faster: it does not matter whether you have ten or a hundred buffers on the list (unless, of course, they all have similar names in which case you are in trouble either way—do not forget to check my Ibuffer settings).

On the other hand, the work spaces (tab-bar) are very useful for organising the various applications that are running inside of Emacs. You can, for example, have your current project on tab (workspace) 1, your email and news reader on 2, music on 3, and so on. Of course, this can also be achieved by using separate frames for each of these, though I generally prefer working in a single frame (plus you can define a window configuration or frameset in a register—see relevant section).

For me tabs are useful as groups of buffers in a given window configuration. I do not want a persistent bar with buttons that introduces extra visual clutter. Switching to tabs is done through completion, specifically prot/icomplete-tab-bar-select-tab. This extends the configurations in my minibuffer essentials and Icomplete. Otherwise, we can use tab-switcher which produces a buffer with the entire list, plus a basic command for marking an item for deletion (same principle as with, e.g., dired).

All settings I configure here are meant to work in accordance with this abstract conception of "tabs are work spaces". Here are the main key chords for tab-bar (they will all work properly if you keep the mode active):

Key Description
C-x t b Open a buffer in a new tab
C-x t d Open a directory in a new tab
C-x t f Open a file in a new tab
C-x t 0 Close current tab
C-x t 1 Close all other tabs
C-x t 2 Open current buffer in new tab

These are consistent with the standard commands for handling windows and accessing buffers/files in the "other window" (the C-x 4 KEY pattern). There is also a command for giving a name to the current tab, accessed via C-x t r, though I find I do not use it.

(use-package tab-bar
  (setq tab-bar-close-button-show nil)
  (setq tab-bar-close-last-tab-choice 'tab-bar-mode-disable)
  (setq tab-bar-close-tab-select 'recent)
  (setq tab-bar-new-tab-choice t)
  (setq tab-bar-new-tab-to 'right)
  (setq tab-bar-position nil)
  (setq tab-bar-show nil)
  (setq tab-bar-tab-hints nil)
  (setq tab-bar-tab-name-function 'tab-bar-tab-name-all)

  (tab-bar-mode -1)
  (tab-bar-history-mode -1)

  (defun prot/icomplete-tab-bar-tab-dwim ()
    "Do-What-I-Mean function for getting to a `tab-bar-mode' tab.
If no other tab exists, create one and switch to it.  If there is
one other tab (so two in total) switch to it without further
questions.  Else use completion to select the tab to switch to."
    (let ((tabs (mapcar (lambda (tab)
                          (alist-get 'name tab))
      (cond ((eq tabs nil)
            ((eq (length tabs) 1)
             (icomplete-vertical-do ()
                (completing-read "Select tab: " tabs nil t)))))))

  :bind (("C-x t t" . prot/icomplete-tab-bar-tab-dwim)
         ("s-t" . prot/icomplete-tab-bar-tab-dwim)
         ("C-x t s" . tab-switcher)))

;; This is only included as a reference.
(use-package tab-line
  :commands (tab-line-mode global-tab-line-mode)
  (global-tab-line-mode -1)) TODO tab-line for groups that make sense, such as EWW

4.3.4 Directional window motions (windmove)

While C-x o (other-window) is very useful when working with two or three windows, it can become tiresome. Thankfully, Emacs comes with a built-in package to move to a window in the given direction. I bind the cardinal directions to Super+Ctrl plus the Vim keys (heresy!), with aliases for the arrows you find on a standard keyboard's number pad.

The windmove-create-window specifies what should happen when trying to move past the edge of the frame. The idea with this is to allow it to create a new window with the contents of the current buffer. I tried it for a while but felt that the times it would interfere with my layout where more than those it would actually speed up my workflow.

(use-package windmove
  (setq windmove-create-window nil)     ; Emacs 27.1
  :bind (("C-s-k" . windmove-up)
         ("C-s-l" . windmove-right)
         ("C-s-j" . windmove-down)
         ("C-s-h" . windmove-left)
         ;; numpad keys clockwise: 8 6 2 4
         ("<kp-up>" . windmove-up)
         ("<kp-right>" . windmove-right)
         ("<kp-down>" . windmove-down)
         ("<kp-left>" . windmove-left)))

5 Applications and utilities

This section includes configurations for programs like email clients, news reader, music players… Anything you would normally see in a standalone application. The end goal is to eventually integrate every aspect of my computing inside of Emacs.

5.1 Calendar

Some basic settings for this tool. It is used by the diary (next section), as well as all Org-mode facilities that require date/time input (see following sections).

(use-package calendar
  (setq calendar-mark-diary-entries-flag t)
  (setq calendar-time-display-form
        '(24-hours ":" minutes
                   (when time-zone
                     (concat " (" time-zone ")"))))
  (setq calendar-week-start-day 1)      ; Monday
  (setq calendar-date-style 'iso)
  (setq calendar-holidays
        (append holiday-general-holidays holiday-local-holidays
                holiday-other-holidays holiday-christian-holidays
                holiday-islamic-holidays holiday-oriental-holidays
  :hook (calendar-today-visible-hook . calendar-mark-today))

5.2 Diary

Emacs comes with a built-in facility to record tasks and create notifications for them. It is simply called diary. I am still assessing its overall utility in my workflow, though these configurations should be good enough at this stage.

(use-package diary-lib
  (setq diary-file "~/.emacs.d/diary")
  (setq diary-entry-marker "diary")
  (setq diary-show-holidays-flag t)
  (setq diary-header-line-flag nil)
  (setq diary-mail-addr "")
  (setq diary-mail-days 3)
  (setq diary-number-of-entries 3)
  (setq diary-comment-start ";")
  (setq diary-comment-end "")
  (setq diary-date-forms
        '((day "/" month "[^/0-9]")
          (day "/" month "/" year "[^0-9]")
          (day " *" monthname " *" year "[^0-9]")
          (monthname " *" day "[^,0-9]")
          (monthname " *" day ", *" year "[^0-9]")
          (year "[-/]" month "[-/]" day "[^0-9]")
          (dayname "\\W"))))

5.3 Org-mode (personal information manager)

Org offers you the basic tools to organise your life in super-efficient ways using nothing but plain text.

In its purest form, Org is a markup language that is similar to Markdown: symbols are used to denote the meaning of a construct in its context, such as what may represent a headline element or a phrase that calls for emphasis.

What lends Org its super powers though is everything else built around it: a rich corpus of elisp functions that automate, link, combine, enhance, structure, or otherwise enrich the process of using this rather simple markup language. This very document is written in org-mode while its website version is produced by a function that exports Org notation into its HTML equivalent.

The present section contains several sub-sections, each dedicated to a particular aspect of Org. Unless explicitly stated, everything should be considered a work-in-progress as I gradually build up my knowledge of this killer app.

5.3.1 Org basic configurations

These are the base settings that other more specialised functions of Org depend on. Here is an overview:

Agenda and default setup
The bulk of the org-agenda configurations is defined in a subsequent section. Here we just declare the default file system paths for searching for relevant files. The "notes" file is meant as a fallback option for when org-capture has not been given a file to write to (also see the org-capture section).
Re-filing items
This is done with C-c C-w which then prompts us for a heading under which the current item should be positioned. I set my Org agenda files as one possible target and the current buffer as the other. The maximum depth should be 2 levels. Re-filing can also be done from inside an org-capture interface. Any new entry should go at the end of the heading it is filled under.
To-do settings
I generally use a very simple system of writing tasks, in the sense that I do not really care about time-tracking or assigning an intermediate state, etc. I let tags and the description further qualify the meaning. The letter inside parentheses is for faster access when using the C-c C-t method. The upside of having lots of specialised keywords is that it becomes easier to filter your tasks in the relevant agenda views.
Logging meta data
I do not really care about tracking all the minutia of why a deadline was reviewed or whatnot. Though it is kind of nice to have a timestamp of when a task was concluded (still don't care about it).

Now on to the miscellaneous settings:

  • With the t value of org-special-ctrl-a/e we assign a special meaning to the motions that take us to the beginning or end of the line when those are performed over a heading. The idea here is that we can always perform changes to the absolute beginning of the line, such as by increasing the heading's depth with M-right.
  • I do not enable this sort of contextual awareness for the C-k command, because I do consider it rather unpredictable.
  • All the markup characters should be hidden from view, in the same way links are. This generally reduces the distractions in the document.
  • The org-structure-template-alist had its value and functionality changed in Org version 9.2, which ships with Emacs 27. To insert a template you must now use C-c C-,.
  • The return key should never follow a link because I sometimes call it by accident. Use C-c C-o instead.
  • With org-loop-over-headlines-in-active-region we can perform actions such as tagging and scheduling on the items within the active region. I configure it to only apply to headings of the same level, in order to avoid possible inconveniences.
(use-package org
  ;; agenda and basic directory structure
  (setq org-directory "~/Org")
  (setq org-default-notes-file "~/Org/")
  (setq org-agenda-files
  (setq org-deadline-warning-days 3)
  ;; refile, todo
  (setq org-refile-targets
        '((org-agenda-files . (:maxlevel . 2))
          (nil . (:maxlevel . 2))))
  (setq org-refile-use-outline-path t)
  (setq org-refile-allow-creating-parent-nodes 'confirm)
  (setq org-refile-use-cache t)
  (setq org-reverse-note-order nil)
  (setq org-todo-keywords
        '((sequence "TODO(t)" "|" "DONE(D)" "CANCELLED(C)")
          (sequence "ACT(a)" "|" "ACTED(A)")
          (sequence "BUY(b)" "|" "BOUGHT(B)")
          (sequence "MEET(m)" "|" "MET(M)" "POSTPONED(P)")
          (sequence "STUDY(s)" "|" "STUDIED(S)")
          (sequence "RECORD(r)" "|" "RECORDED(R)")))
  (setq org-fontify-done-headline t)
  (setq org-fontify-quote-and-verse-blocks t)
  (setq org-fontify-whole-heading-line nil)
  (setq org-enforce-todo-dependencies t)
  (setq org-enforce-todo-checkbox-dependencies t)
  (setq org-track-ordered-property-with-tag t)
  (setq org-highest-priority ?A)
  (setq org-lowest-priority ?C)
  (setq org-default-priority ?A)
  ;; code blocks
  (setq org-confirm-babel-evaluate nil)
  ;; log
  (setq org-log-done 'time)
  (setq org-log-note-clock-out nil)
  (setq org-log-redeadline nil)
  (setq org-log-reschedule nil)
  (setq org-read-date-prefer-future 'time)
  ;; general
  (setq org-special-ctrl-a/e t)
  (setq org-special-ctrl-k nil)
  (setq org-hide-emphasis-markers t)
  (setq org-structure-template-alist    ; CHANGED in Org 9.2, Emacs 27.1
        '(("s" . "src")
          ("E" . "src emacs-lisp")
          ("e" . "example")
          ("q" . "quote")
          ("v" . "verse")
          ("V" . "verbatim")
          ("c" . "center")
          ("C" . "comment")))
  (setq org-catch-invisible-edits 'show)
  (setq org-return-follows-link nil)
  (setq org-loop-over-headlines-in-active-region 'start-level)
  (setq org-imenu-depth 7)
  :bind (:map org-mode-map
         ("<C-return>" . nil)
         ("<C-S-return>" . nil)))

5.3.2 Org links

The org-store-link is one of the commands I use the most, as it allows me to, inter alia, connect the various sections of this document. Use it to store a direct link to the heading you are currently under (also see Consistent Org heading IDs (and anchor tags)).

There are several ways to insert such links. With C-c C-l (which calls org-insert-link) you will be prompted to select a stored link from the link. It will be inserted at point, using the right markup, but will first ask you for a description text. Otherwise you can invoke C-c C-l with an active region, to create a link to that location with the selected text becoming the description. Otherwise you can just call org-insert-last-stored-link to skip the interactive process and just insert the last link.

In addition to these, org-insert-link can be used to create references on demand. Say you have a URL on the kill-ring: C-c C-l, then C-y followed by RET to confirm your input. Complete the process with a description and you are good to go.

(use-package ol
  (setq org-link-keep-stored-after-insertion t)
  :bind (:map org-mode-map
              ("C-c l" . org-store-link)
              ("C-c S-l" . org-toggle-link-display)
              ("C-c C-S-l" . org-insert-last-stored-link))) TODO prot/org-store-link-dwim

5.3.3 Org-capture templates

The org-capture tool is a powerful way to quickly produce some kind of structured information. The type of data and the way to store is determined by a system of templates which accepts a series of possible specifiers as well as the evaluated part of arbitrary elisp code.

Each template is accessed via a key. These are listed in a temporary buffer when you call org-capture. Unique keys give direct access to their template, whereas templates that share a common initial key will produce a second selection list with the remaining options. In the latter case, the initial key entry has no call to an actual function, but is just written as a heading. For an example, look how I do the "idea" section: all templates whose keys follow the pattern iX are only visible after hitting i and are then accessed via X.

The visibility of a template is further controlled by another variable: org-capture-templates-contexts. This allows us to tell Org the context in which we want certain options to appear in. Otherwise they remain concealed from our view. Equipped with this piece of functionality, we can freely write highly-specialised templates that capture structured text when viewing some particular item, but are not needed for more general purposes. I do this for certain actions that only come into effect when reading email inside of the relevant gnus buffers (also check my comprehensive configurations for email and the Gnus news/mail reader).

Speaking of mail, you will notice some specifiers like :fromname. This refers to the From field in emails and will capture the name part only. Other similar keywords are :from (name and email), :fromaddress (email only), :subject.

Specifiers that start with the caret sign (^) represent prompts for further user input. The pattern ^{TEXT} is a prompt whose name is TEXT. To offer possible options, use ^{Initial|ONE|TWO|THREE}, where the first entry is the text of the prompt and all the rest are the available choices (depending on your completion framework, you may need to add an empty option as well, with ||, should you ever want to insert nothing). In some templates I use the ^t specifier, which is a built-in method to ask for a specific date.

The text that goes into a template can be written as part of a string or inside a function that is then evaluated. I generally prefer to use simple strings, though I might revise this approach going forward. To insert a new line inside of a string, use \n.

The %? determines where the point should be once the template is filled in. While %i will insert the contents of the active region.

As things currently stand, my capture templates always write to headings inside of files. Note though that there are more possibilities, as described in the manual.

A file can be specified by its absolute path or just a name. In the latter case, its location is understood relative to org-directory. When using the file+headline pattern, non-existing files are created automatically once you call the relevant template. Same for their respective headings.

Finally, the contrib/org-capture-no-delete-windows and relevant advice address a problem I have when org-capture fails to conclude its actions when called from inside of a side window (for more on those, refer to the section on Window rules and basic tweaks). The code is taken directly from this Stack Overflow thread.

Consider watching my primer on org-capture (2020-02-04) which shows all of the above in action.

(use-package org-capture
  :after org
  (setq org-capture-templates
        '(("b" "Basic task" entry
           (file+headline "" "Basic tasks that need to be reviewed")
           "* %?")
          ("c" "Capture some concise actionable item and exit immediately" entry
           (file+headline "" "Task list without a defined date")
           "* TODO [#B] %^{Title}\n :PROPERTIES:\n :CAPTURED: %U\n :END:\n\n %i %l" :immediate-finish t)
          ("t" "Task of importance with a tag, deadline, and further editable space" entry
           (file+headline "" "Task list with a date")
           "* %^{Scope of task||TODO [#A]|STUDY [#A]|MEET with} %^{Title} %^g\n DEADLINE: %^t\n :PROPERTIES:\n :CONTEXT: %a\n :CAPTURED: %U\n :END:\n\n %i %?")
          ("r" "Reply to an email" entry
           (file+headline "" "Mail correspondence")
           "* TODO [#B] %:subject\n SCHEDULED: %t\n :PROPERTIES:\n :CONTEXT: %a\n :END:\n\n %i %?")
          ("i" "Idea")
          ("ia" "Activity or event" entry
           (file+headline "" "Activities or events")
           "* ACT %^{Act about what}%? :private:\n :PROPERTIES:\n :CAPTURED: %U\n :END:\n\n %i")
          ("ie" "Essay or publication" entry
           (file+headline "" "Essays or publications")
           "* STUDY %^{Expound on which thesis}%? :private:\n :PROPERTIES:\n :CAPTURED: %U\n :END:\n\n %i")
          ("iv" "Video blog or screen cast" entry
           (file+headline "" "Screen casts or vlogs")
           "* RECORD %^{Record on what topic}%? :private:\n :PROPERTIES:\n :CAPTURED: %U\n :END:\n\n %i")))
  (setq org-capture-templates-contexts
        '(("r" ((in-mode . "gnus-article-mode")
                (in-mode . "gnus-summary-mode")))))

  (defun contrib/org-capture-no-delete-windows (oldfun args)
    (cl-letf (((symbol-function 'delete-other-windows) 'ignore))
      (apply oldfun args)))

  (with-eval-after-load "org-capture"
    (advice-add 'org-capture-place-template :around 'contrib/org-capture-no-delete-windows))

  :bind ("C-c c" . org-capture))

5.3.4 Org agenda

The org-agenda is not just a single interface. It rather is your conduit to a set of utilities from where you can keep track of all the tasks you have written in the files declared as part of org-agenda-files (see its value in the section that covers the base Org configurations). Calling org-agenda will present you with a list of possible options. Here is a primer (there are many more functions documented in the manual):

  • The a is where you keep track of all the items that have a date assigned to them, be it SCHEDULED or DEADLINE. To assign such a value to a heading use C-c C-s or C-c C-d respectively.
  • The t will list all your tasks, regardless of whether they have a date assigned to them. You can then filter by keyword, regular expression, etc. Check the top of the buffer for information on how to do that.
  • And the n will offer you a combined view of the above.

Now a few words about some of my customisations (remember to use Emacs' documentation facilities over each item, such as C-h v):

  • Always ask for confirmation when hitting C-k from the agenda views. That command removes the entry in the original file.
  • I do not make heavy use of the built-in diary and am generally not interested in meticulous time tracking, the 'special' day of X, etc. It is the kind of busywork that has no place in my simple, as-stress-free-as-possible life. The Org manual suggests that we can use the diary functionality to “keep track of anniversaries, lunar phases, sunrise/set”. Sure…
  • Show all dates in the current view, including those that have no entries. I find it easier that way to assess how far apart the tasks are.
  • Do not produce a message in the echo area showing the outline path. It is too distracting.
  • Any valid time expressions in headlines should be ignored. Otherwise they are used as part of the relevant sorting methods.
  • The default view should just show me a three day span. All the commands for changing views still work as expected (e.g. hit w to show the current week).
  • Disable follow mode by default. This produces a view of the current item's original context in the other window. It can be toggled on by hitting F.
  • Time stamps should always be expressed in 24h format.
  • Make minor tweaks to the time grid view format.
  • Disable the key bindings that cycle through the agenda files: too easy to hit while trying to perform some other action. Besides, I do not need that kind of functionality.
(use-package org-agenda
  :after org
  (setq org-agenda-confirm-kill t)
  (setq org-agenda-dim-blocked-tasks t)
  (setq org-agenda-include-diary nil)
  (setq org-agenda-show-all-dates t)
  (setq org-agenda-show-outline-path nil)

  ;; All the "skip" need to be reviewed
  (setq org-agenda-skip-additional-timestamps-same-entry t)
  (setq org-agenda-skip-deadline-prewarning-if-scheduled t)
  (setq org-agenda-skip-scheduled-delay-if-deadline t)
  (setq org-agenda-skip-scheduled-if-deadline-is-shown t)
  (setq org-agenda-skip-scheduled-if-done t)
  (setq org-agenda-skip-timestamp-if-deadline-is-shown t)
  (setq org-agenda-skip-timestamp-if-done t)

  (setq org-agenda-search-headline-for-time nil)
  (setq org-agenda-span 3)
  (setq org-agenda-start-on-weekday 1)  ; Monday
  (setq org-agenda-start-with-follow-mode nil)
  (setq org-agenda-timegrid-use-ampm nil)
  (setq org-agenda-time-grid
        '((daily today require-timed)
          (0800 1000 1200 1400 1600 1800 2000)
          "      " "················"))
  (setq org-agenda-use-time-grid t)
  (setq org-agenda-window-setup 'current-window)
  (setq org-agenda-todo-list-sublevels t)
  :bind (("C-c a" . org-agenda)
         :map org-mode-map
         ("C-'" . nil)
         ("C-," . nil)))

5.3.5 Org source code blocks

These are just some basic settings that are particularly useful when inserting source code blocks. I do not want Org to mess up with my indentation, while I need to see the native syntax highlighting for that language.

The org-src-window-setup is accessed via the C-c ' key once inside a code block that has a language assigned to it.

(use-package org-src
  :after org
  (setq org-src-window-setup 'current-window)
  (setq org-src-fontify-natively t)
  (setq org-src-preserve-indentation t)
  (setq org-src-tab-acts-natively t)
  (setq org-edit-src-content-indentation 0))

5.3.6 Org export

Org's "export" facility has the power to convert a .org file into a number of common formats, including .pdf. I only ever use it to produce the HTML version of this document or similar tasks along those lines. In the future, I might spend some time leveraging its potential for more demanding workflows.

(use-package ox
  :after org
  (setq org-export-with-toc t)
  (setq org-export-headline-levels 8)
  (setq org-export-backends
        '(ascii html latex md))
  (setq org-export-dispatch-use-expert-ui nil)) TODO review Org's exporting facilities [0/2]   emacs org TODO research derived export backends TODO automate webpage creation for dotemacs

5.3.7 Consistent Org heading IDs (and anchor tags)

Everything in this section is copied directly from this detailed tutorial on Org header IDs. Basically, the problem is that exported HTML does not have reliable anchor tags for the various sections of the document. This fixes the issue (read the article for more).

(use-package org-id
  :after org
  :commands (contrib/org-get-id
  (setq org-id-link-to-org-use-id 'create-if-interactive-and-no-custom-id)

  (defun contrib/org-get-id (&optional pom create prefix)
    "Get the CUSTOM_ID property of the entry at point-or-marker
POM. If POM is nil, refer to the entry at point. If the entry
does not have an CUSTOM_ID, the function returns nil. However,
when CREATE is non nil, create a CUSTOM_ID if none is present
already. PREFIX will be passed through to `org-id-new'. In any
case, the CUSTOM_ID of the entry is returned."
    (org-with-point-at pom
      (let ((id (org-entry-get nil "CUSTOM_ID")))
         ((and id (stringp id) (string-match "\\S-" id))
          (setq id (org-id-new (concat prefix "h")))
          (org-entry-put pom "CUSTOM_ID" id)
          (org-id-add-location id (buffer-file-name (buffer-base-buffer)))

  (defun contrib/org-id-headlines ()
    "Add CUSTOM_ID properties to all headlines in the current
file which do not already have one."
    (org-map-entries (lambda ()
                       (contrib/org-get-id (point) 'create)))))

5.3.8 Simple presentations inside of Emacs (org-tree-slide)

I like the idea of easily converting an .org file into a set of pseudo slides. It is simple and has no external dependencies. My needs are pretty simple and straightforward: just show some text, narrowing it to the current section.

To make presentations happen, I use org-tree-slide to narrow the buffer to each heading. The darkroom package centres the view, while removing the mode line. While org-superstar prettifies the heading bullets, while removing the leading stars. The other tweaks in the code block below are complementary to this "slideshow effect".

For the font specified herein, make sure to understand the overall configurations by reading the section on primary font settings. I opted not to use text-scale-adjust or some variant thereof, because that only operates on the text of the focused window, whereas I want all interfaces to adapt to the new size so that I can, for example, show the minibuffer while doing a presentation.

(use-package darkroom
  (setq darkroom-text-scale-increase 0))

(use-package org-superstar              ; supersedes `org-bullets'
  :after org
  (setq org-superstar-remove-leading-stars t))

(use-package org-tree-slide
  :after (org darkroom)
  (setq org-tree-slide-breadcrumbs nil)
  (setq org-tree-slide-header nil)
  (setq org-tree-slide-slide-in-effect nil)
  (setq org-tree-slide-heading-emphasis nil)
  (setq org-tree-slide-cursor-init t)
  (setq org-tree-slide-modeline-display nil)
  (setq org-tree-slide-skip-done nil)
  (setq org-tree-slide-skip-comments t)
  (setq org-tree-slide-fold-subtrees-skipped t)
  (setq org-tree-slide-skip-outline-level 8)
  (setq org-tree-slide-never-touch-face t)

  (defun prot/org-presentation ()
    "Specifies conditions that should apply locally upon
activation of `org-tree-slide-mode'."
    (if (eq darkroom-tentative-mode nil)
          (darkroom-tentative-mode 1)
          (org-superstar-mode 1)
          (org-indent-mode 1)
          (blink-cursor-mode -1)
          (setq-local cursor-type '(bar . 1))
      (darkroom-tentative-mode -1)
      (org-superstar-mode -1)
      (org-indent-mode -1)
      (blink-cursor-mode 1)
      (setq-local cursor-type 'box)

  :bind (("<f9>" . org-tree-slide-mode)
         :map org-tree-slide-mode-map
         ("<C-right>" . org-tree-slide-move-next-tree)
         ("<C-left>" . org-tree-slide-move-previous-tree))
  :hook (org-tree-slide-mode-hook . prot/org-presentation))

5.4 Email settings

Configuring email can be quite the challenge, largely because we have been used to the likes of Thunderbird, where you log in once and then everything "just works". The toolset for my current setup consists of the following:

  • Gnus (also pronounced as "News" or "Nooz", etc.), which is a powerful newsreader and email client that is built into Emacs.
  • The built-in capabilities to send email.

Previous versions of this document (prior to 2020-01-30) relied on external tools for fetching and reading email, in particular mu4e, the front-end to the mu mail indexer, and offlineimap. Whereas now I let Gnus handle the task of synchronising with the IMAP server and with all the news groups I am subscribed to. No need to worry about some complex mechanism for syncing, storing, indexing email.

Note that I do not work in an office context, do not need HTML in my email, and do not try to come up with some fancy mechanism for updating my sources and notifying me of the results. My approach is to update my sources manually: the idea is that if I have time to check my news, I also am able to act on them. If you have more complex needs, you will definitely need to adapt things accordingly. Still, this section contains lots of useful information to get you started.

5.4.1 Base email settings

Before configuring the email client, we need to establish the absolute essentials: who we are, where our credentials are stored, and whether encryption is supported. This is done in the first two package declarations.

The prot/auth-get-field is a generic tool for finding the values pertaining to our login credentials. You will see this function used elsewhere in this document. For example, to find the username and password for host hostname we do:

(prot/auth-get-field "hostname" :user)   ; login name
(prot/auth-get-field "hostname" :secret) ; password

Then we set up the interface for composing emails:

  • The mail-user-agent and message-mail-user-agent concern the default email composition buffer, called with C-x m or any other facility that falls back to the compose-mail function. The default is message-mode. When Gnus is running, it will insert relevant paraphernalia, the most important of which is the "Gcc" header. The Gcc saves a copy of the outgoing message to a specified group. In my case that is the "Sent" directory of my default IMAP account.
  • Function prot/message-header-add-gcc is directly related to the above. The inserted header points to my public email account, which is declared in user-mail-address. This concerns only the creation of new emails. While replying to a message, the appropriate information is filled in automatically, based on parameters I specify in the section about account settings and essential configurations.
  • The value of message-citation-line-format is expanded into something like "NAME <EMAIL> [2020-02-19 Wed]:". To learn about all the date-related specifiers, it is better for you to read the documentation with C-h v format-time-string.
(use-package auth-source
  (setq auth-sources '("~/.authinfo.gpg" "~/.authinfo"))
  (setq user-full-name "Protesilaos Stavrou")
  (setq user-mail-address "")

  (defun prot/auth-get-field (host prop)
    "Find PROP in `auth-sources' for HOST entry."
    (let* ((source (auth-source-search :host host))
           (field (plist-get
                   (flatten-list source)
      (if source
        (error "No entry in auth sources")))))

(use-package epa-file
  (setq epa-file-cache-passphrase-for-symmetric-encryption nil))

(use-package message
  (setq mail-user-agent 'message-user-agent)
  (setq compose-mail-user-agent-warnings nil)
  (setq message-mail-user-agent nil)    ; default is `gnus'
  (setq mail-signature "Protesilaos Stavrou\\n")
  (setq message-signature "Protesilaos Stavrou\\n")
  (setq message-citation-line-format "%f [%Y-%m-%d, %R %z]:\n")
  (setq message-citation-line-function
  (setq message-confirm-send nil)
  (setq message-kill-buffer-on-exit t)
  (setq message-wide-reply-confirm-recipients t)
  (setq message-default-charset 'utf-8)
  (add-to-list 'mm-body-charset-encoding-alist '(utf-8 . base64))

  (defun prot/message-header-add-gcc ()
    "While `gnus' is running, add pre-populated Gcc header.

The Gcc header places a copy of the outgoing message to the
appropriate directory of the IMAP server, as per the contents of

In the absence of a Gcc header, the outgoing message will not
appear in the appropriate IMAP directory, though it will still be

Add this function to `message-header-setup-hook'."
    (if (gnus-alive-p)
          (when (message-fetch-field "Gcc")
            (message-remove-header "Gcc"))
          (message-add-header "Gcc: nnimap+pub:Sent"))
      (message "Gnus is not running. No GCC field inserted.")))

  :hook ((message-header-setup-hook . prot/message-header-add-gcc)
         (message-setup-hook . message-sort-headers)))

Below is a sample with the contents of my authinfo.gpg. This is read by gnus and smtpmail to be able to both fetch and send messages from the given account. I strongly encourage you to encrypt this file if you add your login credentials there. Do it from inside dired with : e while the point is over the file. Emacs can decrypt all encrypted files automatically.

machine prv port 993 login MAIL password SECRET
machine inf port 993 login MAIL password SECRET
machine pub port 993 login MAIL password SECRET

machine port 465 login MAIL password SECRET
machine port 465 login MAIL password SECRET
machine port 465 login MAIL password SECRET

Refer to your email provider's documentation in order to determine the port number and server address you need to use for sending and receiving messages. The MAIL is either your email address or some username for logging into the account.

Note that the terms I use above for prv, inf, and pub are just arbitrary names for the given MAIL and SECRET combination. This allows us to reference each name in the Gnus configurations, and share those in a public document like this one, without worrying about leaking private data.

5.4.2 Gnus for reading email, mailing lists, and more

The documentation describes Gnus as the "coffee-brewing, all singing, all dancing, kitchen sink newsreader". I chuckled when I first read it, thinking to myself that the developers have an interesting sense of humour. Then I decided to quickly go through the list of user-facing customisation options: M-x customize-apropos-groups RET gnus RET … Not so funny after all!

Simply put, Gnus is massive. This makes it both extremely powerful and incredibly complicated for new users. Do not let that scare you though: start small and gradually tweak things as you go. This is how you approach Emacs itself. Learn the basics and then figure out your needs from then on.

Now some basic information on the abstractions that Gnus relies on:

  1. The default Gnus buffer is called "Group". It will present you with a list of all the news sources you have subscribed to. By default, Gnus only displays messages that have not been read. The same applies for groups. The "Group" buffer will be empty the very first time you log in because you have not subscribed to anything yet. Use g to fetch new messages from the sources. If you only want to refresh the group at point, do it with M-g.
  2. The "Server" buffer contains a list with all the sources you have specified for discovering news. In my case, these are my email accounts and a Usenet server where mailing lists are hosted. To access the "Server" buffer from inside the "Group" buffer, just hit the caret sign ^. To subscribe to an item, place the point over it and hit u. Do that for your email's inbox and for whatever mailing lists you intend to follow.
  3. The "Summary" buffer contains all the messages of a group. Hitting the return key over a message will split the view in two, with the list above and the message below. Use n or p to move to the next or previous unread message (or N and P to just the next/prev). You access the "Summary" buffer both from the "Group" and the "Server" by entering a group.

It is essential to take things slowly (and first test whether your messages are being sent and that you can receive them). Each buffer has some unique functions that are relevant to the current interface. To learn more about them, use C-h m. Do it for all three of the above. Also rely on C-h k to get information about what each key does in the given context (or just start a key sequence and then hit C-h to display possible combinations in a new Help buffer).

Now a couple more things about the "Group" buffer:

  • A group can be assigned a level of importance. This is a grade whose highest score is 1 and the lowest is 6 (customisable though). Each level has a different colour. To assign a new value to the group at point, do it with S l and then give it a number. Once you have graded your groups, you can perform various actions on a per-level basis. For example, to refresh all levels from 1 up to 3 but not higher, pass a numeric argument to the standard g command. So C-3 g (this is the same as C-u 3 g).
  • Groups can be organised by topic. Create a new one with T n and give it a name. Move a group to a topic with T m. To toggle the view of topics use t (I have a hook that does this automatically at startup). The level of indentation tells us whether a topic is a sub-set of another. Use TAB or C-u TAB to adjust it accordingly. As with levels, you can operate on a per-topic basis. For example, to catch up on all the news of a given topic (mark all as read), you place the point over it, hit c and then confirm your choice.

As noted, Gnus will only show you a list of unread items. To view all your groups, hit L. Use the lower case version l to view only the unread ones. To produce a Summary buffer with read items, hit C-u RET over a group and specify the number of messages you want to list (the other option is C-u M-g from inside the Summary). Another useful trick for the Summary buffer is the use of the caret sign (^) to show you the previous message that the current item is a reply to.

Consider watching my Introduction to Gnus (2020-02-02).

Notwithstanding the customisation options and various idiosyncratic design choices, some prior experience with Emacs' various interfaces will definitely come in handy: Gnus uses similar metaphors for navigating and parsing information. It still is important to read the manual though.

Now here comes the nice part of leveraging the integration that Emacs offers: in my Org mode configurations I have a simple template to capture the current buffer's link. This means that we can quickly convert any item into a task/note and always be able to go back to the original message by following the link. Found an interesting suggestion in some mailing list? Capture it. Need to act on an email later? Capture, capture, capture. Same principle applies to the integration with Dired as a means of attaching files to emails (see next section).

The package declarations below are divided into several subsections to make things easier to read and keep track of. Remember to use C-h v VAR to read documentation about each VAR or simply place the point over it and then hit C-h v to pre-populate the results (C-h f is the equivalent for functions, C-h o for other symbols). Whenever you see some formatting customisations concerning time units, it is better refer to the documentation of the function format-time-string to understand the meaning of the various date/time specifiers. Gnus account settings and essential configurations

Here I only furnish the essentials for the basic Gnus functionality. Subsequent sections expand on the particulars.

  • The gnus-select-method sets the default method for fetching news items. As I want to read mail from several accounts in addition to following Usenet sources, I choose to set it to nil.
  • The gnus-secondary-select-methods is where my accounts are specified. Each nnimap list points to a specific line in my authinfo.gpg file (whose format I described in the base email settings). My emails all use the same server so this method allows me to specify the username (email) and password combination for each of them without making this information public. I am not sure whether the nnimap-stream and nnimap-authinfo-file are needed, but I keep them for the sake of completeness.
  • The gnus-parameters are designed to move my outgoing messages to the "Sent" folder of the account that replies to a given email and to use the right email address, depending on the context. While the variable gnus-gcc-mark-as-read ensures that the outgoing messages are marked as read. The prot/auth-get-field is defined in Base email settings. It is used to get the relevant user name.
  • The "agent" is enabled here and configured in the following section.
  • Setting the gnus-novice-user to nil has the effect of reducing prompts for potentially destructive commands, such as deleting an email. Too many confirmations end up being annoying, but you might opt to keep this to t if you are still new to Gnus.
  • The variables concerning the "dribble" file may be reviewed. The idea is to store the state of Gnus in case Emacs crashes. This has never happened and, therefore, I am not putting too much effort into solving a highly unlikely problem.
  • Consider reviewing nnmail-expiry-wait only after you have some experience with Gnus. I set it to a fairly high value.
  • As for the configurations of mm-encode and mml-sec, these are meant to come into effect when encrypting and signing an outgoing message with C-c C-m C-e (mml-secure-message-sign-encrypt). The guided key selection will ask for confirmation on who to encrypt to. It presents a list with the available keys. Items are marked with m and then the mail can be sent with the standard commands (e.g. C-c C-c). I select myself and whomever the other party is. This is an extra step just to make sure that I have everything right with regard to the keys and the correspondent[s] when using encryption. If this becomes a task I use regularly, I will need to streamline things. For the time being, I want the added confirmation.
(use-package gnus
  ;; accounts
  (setq gnus-select-method '(nnnil))
  (setq gnus-secondary-select-methods
        '((nntp "")
          (nnimap "prv"
                  (nnimap-address "")
                  (nnimap-stream ssl)
                  (nnimap-authinfo-file "~/.authinfo.gpg"))
          (nnimap "inf"
                  (nnimap-address "")
                  (nnimap-stream ssl)
                  (nnimap-authinfo-file "~/.authinfo.gpg"))
          (nnimap "pub"
                  (nnimap-address "")
                  (nnimap-stream ssl)
                  (nnimap-authinfo-file "~/.authinfo.gpg"))))
  (setq gnus-parameters
            (gcc "nnimap+prv:Sent")
             (concat user-full-name " " "<"
                     (prot/auth-get-field "prv" :user)
            (gcc "nnimap+inf:Sent")
             (concat user-full-name " " "<"
                     (prot/auth-get-field "inf" :user)
           (posting-style               ; Uses default name+mail
            (gcc "nnimap+pub:Sent")))))
  (setq gnus-gcc-mark-as-read t)
  (setq gnus-agent t)
  (setq gnus-novice-user nil)
  ;; checking sources
  (setq gnus-check-new-newsgroups 'ask-server)
  (setq gnus-read-active-file 'some)
  ;; dribble
  (setq gnus-use-dribble-file t)
  (setq gnus-always-read-dribble-file t)
  :bind ("C-c m" . gnus))

(use-package nnmail
  (setq nnmail-expiry-wait 30))

(use-package mm-encode
  (setq mm-encrypt-option 'guided))

(use-package mml-sec
  (setq mml-secure-openpgp-encrypt-to-self t)
  (setq mml-secure-openpgp-sign-with-sender t)
  (setq mml-secure-smime-encrypt-to-self t)
  (setq mml-secure-smime-sign-with-sender t)) Gnus agent

The "agent" is a technical term described in the Gnus manual which basically represents the bridge between our Gnus and the server to which it connects to. Gnus is said to be "plugged" when a connection is established. Else it is "unplugged".

Technicalities aside, we can use the agent to configure the handling of messages. For example, we can set an expiry date, after which the message is deleted, or we can create a queue of outgoing messages when Gnus is in an unplugged state.

(use-package gnus-agent
  :after gnus
  (setq gnus-agent-article-alist-save-format 1)  ; uncompressed
  (setq gnus-agent-cache t)
  (setq gnus-agent-confirmation-function 'y-or-n-p)
  (setq gnus-agent-consider-all-articles nil)
  (setq gnus-agent-directory "~/News/agent/")
  (setq gnus-agent-enable-expiration 'ENABLE)
  (setq gnus-agent-expire-all nil)
  (setq gnus-agent-expire-days 30)
  (setq gnus-agent-mark-unread-after-downloaded t)
  (setq gnus-agent-queue-mail t)        ; queue if unplugged
  (setq gnus-agent-synchronize-flags nil)) Gnus article (message view)

In Gnus parlance, the "article" is the window that contains the content of the summary's selected item. This has its own major mode, which is great for us: we can define behaviours and key bindings that only apply when the article is in focus.

I have no particular interest in the HTML-related variables, because I practically never have to read such messages. As a general rule, email that can only be read in HTML is likely spam or annoying enough to be treated as such.

With regard to images, I prefer to inhibit any inline items. If I need to see it, I can always call gnus-article-show-images.

Note that gnus-article-sort-functions requires the most important function to be declared last.

With regard to the key bindings, I have redefined some of the existing ones to suit my workflow and better match my intuitions. For example, in the Article view, hitting s takes you to the Summary buffer. I find that to be a waste, since we can already move between buffers with standard keys. Instead, the s can be used to save the attachment at point. Similarly, I want o to behave just like in dired, where it opens the attachment at point (MIME part) in another buffer.

Finally, here is a tip that I do not configure as I always prefer a manual check: when you receive someone's public PGP key, you can mark it and epa-import-keys-region (though I should probably write a function for this task).

(use-package gnus-art
  :after gnus
  (setq gnus-article-browse-delete-temp 'ask)
  (setq gnus-article-over-scroll nil)
  (setq gnus-article-show-cursor t)
  (setq gnus-article-sort-functions
        '((not gnus-article-sort-by-number)
          (not gnus-article-sort-by-date)))
  (setq gnus-article-truncate-lines nil)
  (setq gnus-html-frame-width 80)
  (setq gnus-html-image-automatic-caching t)
  (setq gnus-inhibit-images t)
  (setq gnus-max-image-proportion 0.7)
  (setq gnus-treat-display-smileys nil)
  (setq gnus-article-mode-line-format "%G %S %m")
  (setq gnus-visible-headers
        '("^From:" "^To:" "^Cc:" "^Newsgroups:" "^Subject:" "^Date:"
          "Followup-To:" "Reply-To:" "^Organization:" "^X-Newsreader:"
  (setq gnus-sorted-header-list gnus-visible-headers)
  :bind (:map gnus-article-mode-map
              ("i" . gnus-article-show-images)
              ("s" . gnus-mime-save-part)
              ("o" . gnus-mime-copy-part))) Gnus asynchronous operations

By default, Gnus performs all its actions in a synchronous fashion. This means that Emacs is blocked until Gnus has finished. By enabling this library, we can use certain functions in a non-blocking way. I do this for sending email.

(use-package gnus-async
  :after gnus
  (setq gnus-asynchronous t)
  (setq gnus-use-article-prefetch 15)) Gnus group (main interface)

I already outlined the utility of the group buffer in the introductory section on Gnus for reading email, mailing lists, and more. In short, it is the epicentre of Gnus, where all your subscribed groups are presented and from where you can browse through your updates.

I use groups in tandem with topics, which allows me to quickly follow updates on the theme I am interested in at the moment. It also allows me to perform per-topic actions, such as updating only the groups it contains or "catching up" to them (marking them as read).

I choose to disable the default behaviour of always showing a group that has "ticked" items (the equivalent of starred or marked as important).

Note that gnus-group-sort-functions requires the most important function to be declared last.

(use-package gnus-group
  :after gnus
  (setq gnus-level-subscribed 6)
  (setq gnus-level-unsubscribed 7)
  (setq gnus-level-zombie 8)
  (setq gnus-list-groups-with-ticked-articles nil)
  (setq gnus-group-sort-function
  (setq gnus-group-mode-line-format "%%b")
  :hook ((gnus-group-mode-hook . hl-line-mode)
         (gnus-select-group-hook . gnus-group-set-timestamp))
  :bind (:map gnus-agent-group-mode-map
              ("M-n" . gnus-topic-goto-next-topic)
              ("M-p" . gnus-topic-goto-previous-topic)))

(use-package gnus-topic
  :after (gnus gnus-group)
  (setq gnus-topic-display-empty-topics nil)
  :hook (gnus-group-mode-hook . gnus-topic-mode)) Gnus summary

This section assumes you have already read my introductory remarks on Gnus for reading email, mailing lists, and more.

Note that the various sort functions expect the primary filter method to be declared last, in case more that one function is to be invoked. The sorting is set to reverse chronological order (newest first).

Threads should not be hidden, while messages whose root has been removed should be grouped together in some meaningful way (which may not always be fully accurate). Furthermore, when moving up or down in the list of messages using just n or p, I want to go to the next message, regardless of whether it has been read or not. I can otherwise rely on standard Emacs motions.

The gnus-user-date-format-alist basically adapts the date to whether the message was within the day or the one before, else falls back to a default ISO-style value. It is then called with %&user-date;.

Also notice the standard behaviour of the %f specifier that is used in the gnus-summary-line-format. It has a conditional behaviour, where it will show the contents of the "From" header field, unless these match some exception, defined in gnus-ignored-from-addresses. When the exception is met, the specifier will fetch the contents of the "To" field instead, prepending to them gnus-summary-to-prefix and/or gnus-summary-newsgroup-prefix (I have no use for the latter). This is useful when viewing a summary buffer with, say, all your sent messages.

Remember to check the docstring of the relevant variables (C-h v), in order to get a list with all the available parameters.

The formatting of the threads using Unicode characters was originally taken from the relevant Emacs wiki entry plus some minor tweaks by me. Otherwise the summary's display is tweaked to show output in nice columns, while being a little bit cleaner than default.

(use-package gnus-sum
  :after (gnus gnus-group)
  (setq gnus-auto-select-first nil)
  (setq gnus-summary-ignore-duplicates t)
  (setq gnus-suppress-duplicates t)
  (setq gnus-save-duplicate-list t)
  (setq gnus-summary-goto-unread nil)
  (setq gnus-summary-make-false-root 'adopt)
  (setq gnus-summary-thread-gathering-function
  (setq gnus-thread-sort-functions
        '((not gnus-thread-sort-by-date)
          (not gnus-thread-sort-by-number)))
  (setq gnus-subthread-sort-functions
  (setq gnus-thread-hide-subtree nil)
  (setq gnus-thread-ignore-subject nil)
  (setq gnus-user-date-format-alist
        '(((gnus-seconds-today) . "Today at %R")
          ((+ 86400 (gnus-seconds-today)) . "Yesterday, %R")
          (t . "%Y-%m-%d %R")))

  ;; When the %f specifier in `gnus-summary-line-format' matches my
  ;; name, this will use the contents of the "To:" field, prefixed by
  ;; the string I specify.  Useful when checking your "Sent" summary.
  (setq gnus-ignored-from-addresses "Protesilaos Stavrou")
  (setq gnus-summary-to-prefix "To: ")

  (setq gnus-summary-line-format "%U%R%z %-16,16&user-date;  %4L:%-30,30f  %B%s\n")
  (setq gnus-summary-mode-line-format "%p")
  (setq gnus-sum-thread-tree-false-root "─┬> ")
  (setq gnus-sum-thread-tree-indent " ")
  (setq gnus-sum-thread-tree-leaf-with-other "├─> ")
  (setq gnus-sum-thread-tree-root "")
  (setq gnus-sum-thread-tree-single-leaf "└─> ")
  (setq gnus-sum-thread-tree-vertical "│")

  ;;;; TODO test and develop these concepts
  ;; (defun prot/gnus-summary-save-parts-all ()
  ;;   "Save all MIME parts (attachments) to the ~/Downloads directory."
  ;;   (interactive)
  ;;   (let ((directory "~/Downloads"))
  ;;     (gnus-summary-save-parts ".*" directory)))
  ;; (defun prot/gnus-summary-search ()
  ;;   (interactive)
  ;;   (let ((regexp (read-from-minibuffer "Search regexp in articles: ")))
  ;;     (when major-mode 'gnus-summary-mode
  ;;           (gnus-summary-search-article regexp))))
  :hook (gnus-summary-mode-hook . hl-line-mode)
  :bind (:map gnus-agent-summary-mode-map
              ("<delete>" . gnus-summary-delete-article)
              ("n" . gnus-summary-next-article)
              ("p" . gnus-summary-prev-article)
              ("N" . gnus-summary-next-unread-article)
              ("P" . gnus-summary-prev-unread-article)
              ("M-n" . gnus-summary-next-thread)
              ("M-p" . gnus-summary-prev-thread)
              ("C-M-n" . gnus-summary-next-group)
              ("C-M-p" . gnus-summary-prev-group)
              ("C-M-^" . gnus-summary-refer-thread))) Gnus server

The "server" is where your news sources are listed and from where you can browse items you would like to subscribe to (e.g. your email account's Inbox or some mailing list on Usenet). Make sure to read about these concepts in the introductory section about Gnus.

(use-package gnus-srvr
  :after gnus
  :hook ((gnus-browse-mode-hook gnus-server-mode-hook) . hl-line-mode)) Gnus intersection with Dired

We can use the built-in directory editor (file manager) as a more convenient way of performing certain tasks that relate to emails, such as attaching all the marked items of the dired buffer to an email we are currently composing or wish to initiate the composition of.

Run C-h m inside of a Dired buffer that has gnus-dired-mode enabled and search for "gnus" to see all the relevant key bindings and the functions they call. I only ever use C-c C-m C-a (C-m is the same as RET).

By the way, make sure to check my comprehensive Dired configurations.

(use-package gnus-dired
  :after (gnus dired)
  :hook (dired-mode-hook . gnus-dired-mode)) TODO Extend Gnus [0/2]   emacs gnus TODO add RSS feeds TODO parse Atom feeds as well?

5.4.3 Sending email (SMTP)

These are the base settings for the SMTP functionality. Passwords and other critical information are stored in ~/.authinfo.gpg, as demonstrated in the base email settings. What follows is just a mirroring of the contents of that file.

With regard to the asynchronous functionality, it is meant to improve performance by carrying out the relevant tasks in a non-blocking way.

(use-package smtpmail
  (setq smtpmail-default-smtp-server "")
  (setq smtpmail-smtp-server "")
  (setq smtpmail-stream-type 'ssl)
  (setq smtpmail-smtp-service 465)
  (setq smtpmail-queue-mail nil))

(use-package smtpmail-async
  :after smtpmail
  (setq send-mail-function 'async-smtpmail-send-it)
  (setq message-send-mail-function 'async-smtpmail-send-it))

5.4.4 Contact management (EBDB)

This is a contact manager that integrates nicely with Gnus (see previous sections in email settings). It is an alternative to the more established bbdb package. I use it to store names and addresses from my email correspondence. Nothing fancy here.

ebdb has a comprehensive manual and a broad range of customisation options. My needs are simple though, as I have basically lived without any contact management app for ages (I practised “social distancing” before it became a thing!).

This is how EBDB works inside of the Gnus "Summary" or "Article" buffers:

  • Hit ; ; to display a window on the side of the message with information about the records in the database that correspond to those in the message. If no record exists, the window will not be displayed. Alternatively, set ebdb-mua-pop-up to t to always get the window, whether empty or not.
  • With ; : you can create new database entries with the information harvested from the current message. There are ways to make this automatic, but I did not experiment with them. If I store someone's contact information, it means I have used it before or have a clear intent to do so. No automatic records, please.
  • Type ; followed by C-h to bring up a help buffer with all the available keybindings (this works for every key chord, by the way).

Some useful commands that are not specific to Gnus:

  • In a message composition buffer (using e.g. the compose-mail bound to C-x m), hit TAB while in a header field where emails are appropriate to either complete what you have typed in or get feedback on possible completion candidates.
  • ebdb-cite-records lets you input a contact in the current buffer. If you do this in an org-mode buffer, it will prefix the email address with a mailto: tag.

It is also possible to connect EBDB with the Diary and with Org's Agenda views. I have no use for such features. If, however, my needs ever evolve to something more demanding, I am confident this tool will cope with them. For the time being, this package declaration is enough.

(use-package ebdb
  :pin gnu                              ; Prefer ELPA version
  (require 'ebdb-gnus)
  (require 'ebdb-message)
  (setq ebdb-default-window-size 0.2)
  (setq ebdb-mua-pop-up nil)
  (setq ebdb-mua-auto-update-p 'query)
  (setq ebdb-mua-sender-update-p 'query)
  (setq ebdb-mua-reader-update-p 'query)
  (setq ebdb-add-aka 'query)
  (setq ebdb-add-name 'query)
  (setq ebdb-add-mails 'query))

5.4.5 Bug tracker (debbugs)

This tool is dedicated to the task of browsing the list of reported bugs about Emacs, though one can always subscribe to the relevant Usenet source (mailing list).

Reading bug reports can help discover aspects of Emacs I am not aware of. Plus, this is where I can track any entry that concerns my themes on ELPA (see section on the Modus themes).

Technically, debbugs is not part of my email configurations, in the sense that it does not extend the functionality of Gnus. It does, nonetheless, produce a summary of messages using Gnus as its backend. So it inherits all my Gnus customisations and feels part of the same experience.

Note that in the package declaration below the debbugs-gnu-lars-workflow is a reference to Lars Magne Ingebrigtsen. As far as I can tell, that variable simply wraps two others that remove duplicate entries. Per the source code in debbugs-gnu.el:

  • gnus-suppress-duplicates t
  • gnus-save-duplicate-list t

While I agree with that approach and already configure these, I still set "lars" to nil to guard against any future expansion of its scope.

(use-package debbugs
  (setq debbugs-gnu-lars-workflow nil)  ; sorry Lars, I'm paranoid
  (setq debbugs-gnu-mail-backend 'gnus))

5.5 Version control tools

5.5.1 Generic version control (VC) framework

VC is a generic framework that works with several version control systems, else "backends". In practice though, I only ever use it with git. Compared with magit (see section on Magit configs), vc offers a more abstract, buffer-oriented workflow that, I feel, covers all common version control cases.

With VC we can track and ignore files, commit changes, view diffs and logs, push and pull from a remote… Everything you would expect from a version control system. But I am not sure it is possible or convenient to perform tasks such as staging only a particular part of a diff, handling multiple remotes, interactively rebasing the commit log, etc.

As such, I employ VC as my generalist interface to the most common Git interactions: diffs, commits, logs. While I rely on Magit for expedient access to the more advanced features of Git, all of which are rendered approachable through an interactive/modal interface.

I started using VC with this a common scenario: to produce a diff of my current changes formatted as a ready-to-apply patch. This is done with vc-diff (also see the vc-git-diff-switches variable). A diff hunk can be applied while in the *vc-diff* buffer by hitting C-c C-a. Fairly easy. One can always save the diff buffer using the standard C-x C-w command (write-file). Use these to send patches (e.g. via email), without having to go through proprietary web tools.

Another nice feature is a region-specific commit log with C-x v h (vc-region-history). Highlight a region and run the command to get a clear view of how it took form. Quite an easy way to "git blame". For a more macroscopic view of the file's entire history you can always rely on C-x v g or my alias C-x v a which calls vc-annotate. Commits are colour-coded with those on the red side of the spectrum being the newest ones, while those on the blue end representing the older ones.

C-x v v (vc-next-action) is a hidden gem: a paradigm of minimalism done right. On a new project it will first ask you for a backend. Selecting git will run git init. Invoking the command again will track the file and commit the changes. If the file is already being tracked, it will just proceed to the next step. It always follows a logical sequence which translates into a powerful metaphor of going through the process of committing changes to files.

The commit buffer presents you with a summary section followed by the body of the commit message, separated by an empty line (shown as a border) as per the conventions that govern good commit messages. In the lower part you can see the affected files (just the current file by default). The standard C-c C-c confirms the commit, while C-c C-k cancels it. Furthermore, there is C-c C-f to view the file set concerned (in case it is not already visible below the message area), as well as C-c C-d to produce a diff of what is being committed.

You can review the commit log for the current file with C-x v l (vc-print-log) or for the entirety of the current project with C-x v L (vc-print-root-log). From there you can display any diff with d or find the prior state of the commit at point with f. Use the latter as a starting point for reverting to a prior state. Also note that while inside the root log view, you can use an active region to show diffs in that range of commits.

With vc-dir, which I rebind to the more logical "project" mnemonic of C-x v p (prot/vc-dir-project), you can mark with m several files to add to a commit. Use M to mark all files with the same status. This mode offers easy access to the standard VC actions. To commit the file at point or the marked ones hit v. To push do P. Same principle for diffs, logs, etc.

By the way, rebinding vc-dir has another upside of allowing its original key binding to be assigned to vc-diff; which in turn makes that mnemonically consistent with the C-x v D sequence, else vc-root-diff (make sure to check my key re-bindings or aliases).

Some more common actions (read their docstrings for further help):

Command Key chord VC-dir key
vc-update C-x v + +
vc-push C-x v P P
vc-log-incoming C-x v I I
vc-log-outgoing C-x v O O

Based on these (and there are more), you can already see how VC may be used as your main tool for version control, from committing changes, to pulling/pushing from/to a remote, viewing commit logs, etc. It depends on your needs and preferences. Run C-x v C-h to get a full list of the possible commands. Read the docstring of each command for further information. Then in each of the various VC modes try C-h m for further help. Also consult the manual and make liberal use of C-h f or C-h v.

Changes to all tracked files are highlighted in the fringe thanks to the diff-hl package which is defined elsewhere in this document (as I consider it an "interface" element). Any rules that control the placement of VC-related (and other) buffers are defined in the section on window rules and basic tweaks (specifically, refer to the variable display-buffer-alist).

Also watch my Introduction to the Emacs Version Control framework (2020-03-30).

Careful with the keybindings I define. Many of them do other things by default and I might still review them further. When in doubt, stick with the defaults.

(use-package vc
  (setq vc-find-revision-no-save t)
  (require 'log-view)                   ; needed for the key bindings
  :bind (("C-x v b" . vc-retrieve-tag)  ; "branch" switch
         ("C-x v t" . vc-create-tag)
         ("C-x v f" . vc-log-incoming)  ; the actual git fetch
         ("C-x v F" . vc-update)        ; "F" because "P" is push
         ("C-x v d" . vc-diff)
         (:map log-view-mode-map
               ("<tab>" . log-view-toggle-entry-display)
               ("<return>" . log-view-find-revision)
               ("s" . vc-log-search)
               ("o" . vc-log-outgoing)
               ("f" . vc-log-incoming)
               ("F" . vc-update)
               ("P" . vc-push))))

(use-package vc-dir
  (defun prot/vc-dir-project ()
    "Unconditionally display `vc-diff' for the current project."
    (vc-dir (vc-root-dir)))
  :bind (("C-x v p" . prot/vc-dir-project)
         :map vc-dir-mode-map
         ("b" . vc-retrieve-tag)
         ("t" . vc-create-tag)
         ("o" . vc-log-outgoing)
         ("f" . vc-log-incoming) ; replaces `vc-dir-find-file' (use RET)
         ("F" . vc-update)       ; symmetric with P: `vc-push'
         ("d" . vc-diff)         ; align with D: `vc-root-diff'
         ("k" . vc-dir-clean-files)))

(use-package vc-git
  (setq vc-git-diff-switches "--patch-with-stat")
  (setq vc-git-print-log-follow t))

(use-package vc-annotate
  (setq vc-annotate-display-mode 'scale) ; scale to oldest
  :bind (("C-x v a" . vc-annotate)       ; `vc-update-change-log' is not in git
         :map vc-annotate-mode-map
         ("<tab>" . vc-annotate-toggle-annotation-visibility))) Diff-mode

This covers the standard diff-mode, which I use when interfacing with the built-in Version Control framework (see the section on VC), but also find while browsing various newsgroups via Gnus (see email settings).

Overview of these tweaks:

  • Always start the buffer in a read-only state. A typo will mess things up when trying to apply a patch.
  • After a applying a diff hunk move on to the next one.
  • Update hunk headers automatically following an edit to the diff.
  • Do not show word-wise ("refined") changes upon activation. I prefer to do so manually with C-c C-b for the current diff hunk. All such highlights are removed if you generate the buffer again (with g as expected) and the default is to not show word-wise changes.
  • Do not prettify headers. I like the standard "patch" looks.
  • Do not do any further syntax highlighting. This is not the place to check your pretty code, plus it will most likely render colours highly inaccessible.

Basics for diff-mode buffers (as always you can learn more about the current buffer's major mode with C-h m—read How do you learn Emacs? in the FAQ below):

  • C-c C-c or M-o takes you to the point of the changes in the source file. If you run this of the diff hunk's heading, you go to the beginning of the context. But if you place the point somewhere inside of the diff's added changes or context, you will visit that exact position in the original file (does not work for removed text because technically it does not exist). Awesome feature!
  • When working with patches to source code, which are distributed e.g. via email, you can apply the current hunk with C-c C-a or test for compatibility with C-c C-t. This is a nice way to easily merge contributions of from others, without having to go through the workflow of some proprietary Git/Version-Control forge.
  • By default C-c C-n offers a convenience wrapper for Emacs' narrowing commands (such as C-x n n for narrow-to-region. The standard is to narrow to the current diff hunk. With the C-u prefix it will narrow to the current file instead (obviously this only makes sense when inspecting a multi-file set of changes). If you use the default, you must manually exit narrowing with the standard C-x n w, but for this case I prefer prot/diff-restrict-view-dwim which will widen the view when narrowing is already in effect, else narrow as intended (I bind it to C-c C-n to replace the original).
  • With prot/diff-buffer-with-file I have a simple wrapper around a built-in command that compares a buffer to its underlying file. This compares the changes made since the last save and my wrapper is about picking the current buffer directly. If there are no differences, then the command with run vc-diff instead. Note that in my configurations for the Generic version control (VC) framework, I remap the keys that pertain to diffs.
  • With M-n and M-p you move between hunks. With M-} and M-{ or M-N, M-P do the same between files.
(use-package diff
  (setq diff-default-read-only t)
  (setq diff-advance-after-apply-hunk t)
  (setq diff-update-on-the-fly t)
  ;; The following are from Emacs 27.1
  (setq diff-refine nil)
  (setq diff-font-lock-prettify nil)
  (setq diff-font-lock-syntax nil)

  (defun prot/diff-buffer-with-file (&optional arg)
    "Compare buffer to its file, else run `vc-diff'.
With \\[universal-argument] also enable highlighting of word-wise
changes, local to the current buffer."
    (interactive "P")
    (let ((buf nil))     ; this method will "fail" if multi diff buffers
      (if (buffer-modified-p)
            (diff-buffer-with-file (current-buffer))
            (setq buf "*Diff*"))
        (setq buf "*vc-diff*"))
      (when arg
        (with-current-buffer (get-buffer buf)
          (setq-local diff-refine 'font-lock)))))

  (defun prot/diff-restrict-view-dwim (&optional arg)
    "Use `diff-restrict-view', or widen when already narrowed.
By default the narrowing effect applies to the focused diff hunk.
With \\[universal-argument] do it for the current file instead."
    (interactive "P")
    (when (derived-mode-p 'diff-mode)
      (if (buffer-narrowed-p)
            (message "Widened the view"))
        (if arg
              (diff-restrict-view arg)
              (message "Narrowed to file"))
          (message "Narrowed to diff hunk")))))

  ;; `prot/diff-buffer-with-file' replaces the default for `vc-diff'
  ;; (which I bind to another key---see previous section).
  :bind (("C-x v =" . prot/diff-buffer-with-file)
         :map diff-mode-map
         ("C-c C-n" . prot/diff-restrict-view-dwim))) Commit log editing

The log-edit library defines a set of general commands that we can use to, inter alia, commit changes to Git, using the surprisingly powerful built-in Version Control (VC) framework.

Overview of the following tweaks:

  • While editing a commit message, only prompt for confirmation if the list of files has changed since the beginning of the editing session.
  • Once the commit is done, remove its buffer.
  • Always add a newline.
  • Do not offer to manually write an Author: header. Though this can be useful if someone sends a patch.

Also make sure to read the guide for writing a Git commit message. I always use auto-fill-mode to wrap lines to the value of fill-column (see relevant configs).

(use-package log-edit
  (setq log-edit-confirm 'changed)
  (setq log-edit-keep-buffer nil)
  (setq log-edit-require-final-newline t)
  (setq log-edit-setup-add-author nil))

5.5.2 Interactive git front-end (Magit)

As noted in the section on the built-in Version Control framework I use Magit for easy access to the advanced features of Git. While I rely on the built-in tools for all day-to-day operations.

Magit offers a modal interface where the full power of git is neatly organised in sets of keys that are directly accessible without holding down any modifiers.

While inside the magit-status buffer, hit ? to produce a transient menu with the possible vectors to action. Do it again inside each of the Magit buffers to view the keys that work for their context.

Consider viewing my Introduction to Magit (2020-04-04) for how to stage diffs, commit changes, view logs, create branches, and so on. Base Magit settings

Magit has great defaults. I only found a few things that I would like to customise, which I do in the following package declarations.

(use-package magit
  :bind ("C-c g" . magit-status)) Magit commits

The following package is configured in accordance with the guidelines provided by this article on writing a Git commit message. The gist is to write commits that are clean and easy to read. The fill-column is set elsewhere in this document to 72 characters long.

(use-package git-commit
  :after magit
  (setq git-commit-summary-max-length 50)
  (setq git-commit-known-pseudo-headers
  (setq git-commit-style-convention-checks
          overlong-summary-line))) Magit diffs

The settings below are for the diff screens that Magit produces. I just want to highlight changes within a line, not just the line itself. I enable it only for the focused hunk (there is an option for 'all).

(use-package magit-diff
  :after magit
  (setq magit-diff-refine-hunk t)) Magit repo list

When maintaining a number of projects, it sometimes is necessary to produce a full list of them with their corresponding Magit status. That way you can determine very quickly which repositories need to be examined further.

(use-package magit-repos
  :after magit
  :commands magit-list-repositories
  (setq magit-repository-directories
        '(("~/Git/Projects" . 1))))

5.5.3 Smerge and Ediff

Read this section, because it matters more than the code below it.

Sometimes we face a situation where we have conflicting versions of a file and the version control backend cannot solve them on its own. This can happen fairly often when collaborating with other people or, more generally, when we keep our work spread across multiple feature branches with diverging histories. Whenever such conflicts arise, Emacs will automatically annotate the offending files with special markers that show the conflicting differences. Visiting those files will then activate smerge-mode. At which point we are in control.

Smerge revolves around the concept of dividing the conflicting part into an "upper" (red) and a "lower" section (green), possibly with their common ancestor or last point of convergence in the middle (yellow).

With this in mind we can operate on the marked differences by relying on the functions that Smerge provides, all of which are accessed by default through the common prefix of C-c ^. Start by typing the prefix followed by C-h to see all possible key chords. These are the commands I have used the most:

  • C-c ^ u (smerge-keep-upper)
  • C-c ^ l (smerge-keep-lower)
  • C-c ^ b (smerge-keep-base)
  • C-c ^ a (smerge-keep-all)
  • C-c ^ n (smerge-next)
  • C-c ^ p (smerge-prev)

Also check the defhydra I provide for it to make things simpler.

Proceed to edit the file the way you want until no more conflicts exist. You can also do things through manual editing, with standard commands and motions, but that can be prone to errors (which lead to more conflicts). At any moment in this process, you can switch to ediff, which offers a more powerful way of working with differences. Type the key chord C-c ^ E (smerge-ediff).

Ediff is a powerhouse that is likely to cover all your needs in this area (including those you are not aware of). For our purposes, what matters is to understand the basic concepts.

The way this tool works is that it starts by producing a layout of the two conflicting versions with access to a "control panel" for operating on them. By default, the panel is positioned on a new frame, but I find that rather awkward—my config puts it inside an Emacs window instead. While focusing the control panel, you can move between each diff range with n and p. The focused section will be coloured using red, green, and yellow, while all other diffs will be presented in gray.

On each diff, you have three options: to use the version of buffer A (red), of buffer B (green), or a combination of the two. The keys for each of those are a, b, and + respectively. Your choice will be reflected in buffer C (the yellow one). Use these to resolve all conflicts and then quit the session with q.

Concerning the combination of versions between A and B, Ediff has the behaviour of also inserting as plain text the annotation markers that Smerge relied on. As of this writing (2020-04-10), I am not aware of an automatic or convenient way to omit those prior to confirming our edits. To that end, I tweak the wording of the markers to some unique string (see package below) and then run flush-lines to remove them before saving the resulting buffer (so right after the q). For more on this, check prot/ediff-flush-combination-pattern.

For git users, to actually reference the common ancestor (the point before the branching paths started) we must run this command once in our command-line prompt (writes to your global .gitconfig file):

git config --global merge.conflictStyle diff3

This is optional, but I find that I like it. At any rate, the configurations I have below are straightforward (learn more about this powerful tool by hitting ? inside of its control panel and by consulting its comprehensive manual):

  • Do not keep all the buffers after exiting the Ediff session.
  • Keep buffers in an editable state. Otherwise it is impossible to perform the changes we are interested in.
  • Show the common ancestor in another buffer. This helps provide further context of how things took their form.
  • Show only the conflicting parts. This is not a review of all diffs.
  • Prefer putting windows side-by-side, rather than one below the other.
  • Do not enter the ediff session in a new frame. This also means that the control panel will be inside an Emacs window (at the bottom part) rather than in a tiny frame of its own.

There actually is not much in terms of Smerge-related configurations. The package is small and does one thing well. I do, however, specify a hydra (as explained in the relevant section) to make it easier to navigate conflicts. Since smerge-mode only gets activated when visiting a file with diff conflicts, we can load the hydra upon accessing it. Makes things nice and simple. No need to have a separate key binding.

Also watch my video of Smerge and Ediff for git conflict resolution (2020-04-10).

(use-package smerge-mode
  (defhydra prot/hydra-smerge-mode
    (:color pink :hint nil :post (smerge-auto-leave))
   ^Motions^      ^Actions^
_n_: Next      _b_: Base
_p_: Prev      _u_: Upper
^^             _l_: Lower
^^             _a_: All
    ("n" smerge-next)
    ("p" smerge-prev)
    ("b" smerge-keep-base)
    ("u" smerge-keep-upper)
    ("l" smerge-keep-lower)
    ("a" smerge-keep-all)
    ("q" nil "cancel" :color blue))

  (defun prot/smerge-mode-hydra ()
    "Load `prot/hydra-smerge-mode' in `smerge-mode'."
    (when smerge-mode
  :hook ((vc-find-file-hook . prot/smerge-mode-hydra)
         (magit-diff-visit-file-hook . prot/smerge-mode-hydra)))

(use-package ediff
  (setq ediff-keep-variants nil)
  (setq ediff-make-buffers-readonly-at-startup nil)
  (setq ediff-merge-revisions-with-ancestor t)
  (setq ediff-show-clashes-only t)
  (setq ediff-split-window-function 'split-window-horizontally)
  (setq ediff-window-setup-function 'ediff-setup-windows-plain)

  ;; Tweak those for safer identification and removal
  (setq ediff-combination-pattern
        '("<<<<<<< prot-ediff-combine Variant A" A
          ">>>>>>> prot-ediff-combine Variant B" B
          "####### prot-ediff-combine Ancestor" Ancestor
          "======= prot-ediff-combine End"))

  ;; TODO automate process in a robust way, or at least offer a good key
  ;; binding.
  (defun prot/ediff-flush-combination-pattern ()
    "Remove my custom `ediff-combination-pattern' markers.

This is a quick-and-dirty way to get rid of the markers that are
left behind by `smerge-ediff' when combining the output of two
diffs.  While this could be automated via a hook, I am not yet
sure this is a good approach."
    (flush-lines ".*prot-ediff.*" (point-min) (point-max) nil)))

5.6 Shells and terminal emulators

It should come to no surprise that Emacs can operate as both a terminal emulator for command line shells and toolkit for terminal emulators. The present section covers both cases.

5.6.1 Command-line shells Shell (M-x shell)

This is a shell (Bash, in my case) that runs inside of Emacs. Unlike the terminal emulators defined below, this one can use standard Emacs keys and behaves much like an ordinary buffer. It also integrates nicely with the built-in completion tools, which makes it particularly nice to work with (see Minibuffer essentials and Icomplete).

The one area where this Shell differs substantially from ordinary buffers is with regard to the command prompt: you can re-run a command on the scroll-back buffer by just hitting RET while point is on its line (no need to go back to the end and cycle the command history with M-p or M-n).

Another peculiarity relative to the standard commands in the terminal is to search backward through your history with M-r (whereas in a terminal emulator you use C-r).

Run C-h m inside of a shell buffer to learn about all the key bindings and corresponding functions.

(use-package shell
  :commands shell-command
  (setq ansi-color-for-comint-mode t)
  (setq shell-command-prompt-show-cwd t) ; Emacs 27.1

  (defun prot/shell-multi ()             ; TODO review
    "Spawn a new instance of `shell' and give it a unique name
based on the directory of the current buffer."
    (let* ((parent (if (buffer-file-name)
                       (file-name-directory (buffer-file-name))
           (name (car (last (split-string parent "/" t)))))
      (with-current-buffer (shell)
         (generate-new-buffer-name (concat "*shell: " name "*"))))))
  :bind (("<s-return>" . shell)
         ("<s-S-return>" . prot/shell-multi))) Eshell (like M-x shell but in elisp)

While I did read the eshell manual fairly early in my introduction to Emacs (July 2019), I failed to appreciate its added value. Sure, you could run Elisp commands as well as those of the standard shells. But that did not seem like an obvious step forward. More of a lateral movement at best, for it was asking for a whole new way of thinking about the shell that could potentially complicate things in other areas.

After gaining some experience with Emacs, I now (April 2020) have developed a newfound appreciation of Eshell's flexibility. The fact that it can understand elisp means that you can develop entirely new modes of interaction that involve the command line and extend into other Emacs major modes. For example, you can cd into a directory and then dired to access the full power of Emacs' superb file manager (also check the section on Dired).

The core value of Eshell is that it behaves like an ordinary buffer. This means that we can build our own functions for moving around and operating on text. prot/eshell-put-last-output-to-buffer puts that in concrete terms. The function will take the output of the last command and put it in a new buffer. You can then select that buffer, edit it as you will and, say, C-x C-w to write it permanently to a file.

Here is another case where integration with standard Emacs commands truly shines. In Eshell, cd followed by the equals sign will produce a numbered list of the directories you have accessed. With cd -NUM you can switch to the one you want. So far so good, how about we also switch to a dired buffer from there? Do it with cd -NUM && dired.

Speaking of directories, eshell-cd-on-directory allows you to omit cd and just type in the path to a directory. Awesome! And while on the topic of Dired, you can use dired-jump to always switch to a dired buffer of the present working directory. To that I add my own little utility (prot/eshell-file-parent-dir) for jumping to the directory of the file at point: very useful in cases where you get files from different sources, such as with find (though do also check the various options for running find with dired, like find-name-dired).

Extensibility is seamless. prot/eshell-complete-history offers a completion interface to the input history. No need for awkard back and forth or for searching for a regexp without any immediate feedback. In a similar fashion prot/eshell-complete-recent-dir provides completion for switching directly to a recent directory. While the more general prot/eshell-find-subdirectory-recursive lets you switch to a subdirectory somewhere inside the path you are.

Instead of outlining the remaining features of Eshell, I strongly encourage you to read the manual. It is fairly short. What I have here is just an early sample of what can be achieved with fairly few little tweaks.

Also check these valuable resources:

(use-package eshell
  :bind ("<s-return>" . eshell))

(use-package esh-mode
  ;; used by other functions below
  (declare-function ffap-file-at-point "ffap.el")

  (defun prot/eshell-insert-file-at-point ()
    "Insert (cat) contents of file at point."
    (let ((file (ffap-file-at-point)))
      (if file
            (insert (concat "cat " file))
        (user-error "No file at point"))))

  (defun prot/eshell-kill-save-file-at-point ()
    "Add to kill-ring the absolute path of file at point."
    (let ((file (ffap-file-at-point)))
      (if file
          (kill-new (concat (eshell/pwd) "/" file))
        (user-error "No file at point"))))

  (defun prot/eshell-find-file-at-point ()
    "Run `find-file' for file at point (ordinary file or dir).
Recall that this will produce a `dired' buffer if the file is a
    (let ((file (ffap-file-at-point)))
      (if file
          (find-file file)
        (user-error "No file at point"))))

  (defun prot/eshell-file-parent-dir ()
    "Open `dired' with the parent directory of file at point."
    (let ((file (ffap-file-at-point)))
      (if file
          (dired (file-name-directory file))
        (user-error "No parent dir for file to jump to"))))

  (defun prot/eshell-mkcd (dir)         ; TODO define alias
    "Make a directory, or path, and switch to it."
    (eshell/mkdir "-p" dir)
    (eshell/cd dir))

  (defun prot/eshell-put-last-output-to-buffer ()
    "Produce a buffer with output of last `eshell' command."
    (let ((eshell-output (kill-region (eshell-beginning-of-output)
      (with-current-buffer (get-buffer-create  "*last-eshell-output*")
        (yank)           ; TODO do it with `insert' and `delete-region'?
        (switch-to-buffer-other-window (current-buffer)))))

  (defun prot/eshell-complete-redirect-to-buffer ()
    "Complete the syntax for appending to a buffer via `eshell'."
     (concat " >>> #<"
             (format "%s"
                     (read-buffer-to-switch "Switch to buffer: "))

  (defun prot/eshell-narrow-output-highlight-regexp ()
    (let ((regexp (read-regexp "Regexp to highlight")))
      (narrow-to-region (eshell-beginning-of-output)
      (goto-char (point-min))
      (highlight-regexp regexp 'hi-yellow)))
  :bind (:map eshell-mode-map
              ("M-k" . eshell-kill-input)
              ("C-c M-w" . prot/eshell-kill-save-file-at-point)
              ("C-c i" . prot/eshell-insert-file-at-point)
              ("C-c f" . prot/eshell-find-file-at-point)
              ("C-c C-f" . prot/eshell-find-file-at-point)
              ;; The C-c o shadows `prot/olivetti-mode', but that is
              ;; fine in this map
              ("C-c o" . prot/eshell-put-last-output-to-buffer)
              ("C-c >" . prot/eshell-complete-redirect-to-buffer)
              ("C-c C-j" . prot/eshell-file-parent-dir)
              ("C-c h" . prot/eshell-narrow-output-highlight-regexp)))

(use-package esh-module
  (setq eshell-modules-list             ; Needs review

(use-package em-dirs
  :after esh-mode
  (setq eshell-cd-on-directory t))

(use-package em-tramp
  :after esh-mode
  (setq password-cache t)
  (setq password-cache-expiry 600))

(use-package em-hist
  :after esh-mode
  (setq eshell-hist-ignoredups t)
  (setq eshell-save-history-on-exit t)

  (defun prot/eshell-complete-history ()
    "Insert element from `eshell' history using completion."
    (let ((hist (ring-elements eshell-history-ring)))
       (completing-read "Input history: " hist nil t))))

  (defun prot/eshell-complete-recent-dir (&optional arg)
    "Switch to a recent `eshell' directory using completion.
With \\[universal-argument] also open the directory in a `dired'
    (interactive "P")
    (let* ((dirs (ring-elements eshell-last-dir-ring))
           (dir (icomplete-vertical-do ()
                  (completing-read "Switch to recent dir: " dirs nil t))))
      (insert dir)                      ; Not good enough
      (eshell-send-input)               ; Should cd directly…
      (when arg
        (dired dir))))

  ;; `cl-remove-if' is used right below
  (declare-function cl-remove-if "cl-seq")

  (defun prot/eshell-find-subdirectory-recursive ()
    "Recursive `eshell/cd' to subdirectory.
This command has the potential for infinite recursion: use it
wisely or prepare to use `eshell-interrupt-process'."
    (let* ((dir (eshell/pwd))
           (contents (directory-files-recursively dir ".*" t nil nil))
           ;; (contents (directory-files dir t))
           (find-directories (mapcar (lambda (x)
                                       (when (file-directory-p x)
                                         (abbreviate-file-name x)))
           (subdirs (delete nil find-directories))
           (cands (cl-remove-if (lambda (x) (string-match-p "\\.git" x)) subdirs))
           (selection (icomplete-vertical-do ()
                        (completing-read "Find sub-directory: " cands nil t))))
      (insert selection)

  :bind (:map eshell-hist-mode-map
              ("M-s" . nil) ; I use this for lots of more useful commands
              ("M-r" . prot/eshell-complete-history) ; use this to find input history
              ("C-c d" . prot/eshell-find-subdirectory-recursive)
              ("C-c =" . prot/eshell-complete-recent-dir)))

5.6.2 Terminals inside of Emacs Vterm terminal emulator

The built-in terminal emulators are not up to par with the likes of Xterm and its peers. Perhaps they were a good compromise in yester years, but we have come to expect better from our system. Thankfully there is a package that might eventually makes its way into Emacs proper, provided all legal requirements are met. This is Vterm, which is an implementation of the external libvterm library (and which requires cmake in order to be compiled).

Vterm is a fully fledged terminal emulator (not mere shell) inside of Emacs. Its main differences with Shell (M-x shell) can be summarised thus:

  • Vterm can handle graphics and ANSI escape sequences. Shell cannot.
  • Shell behaves more like an ordinary Emacs buffer. Vterm is like an external application that has been embedded in the Emacs frame.
  • Vterm does tab-completion like a standard terminal. Shell can use the Emacs completion framework. To make this concrete, with Vterm if you type cd G <tab> it expands the Git subdirectory. Whereas in Shell with my completion framework (and tweaks) I can do cd G/P/m-t <tab> to expand to cd Git/Projects/modus-themes/.

My workflow is to keep the Shell as my main conduit to the command line, such as for when I need to call one of my scripts, and only use Vterm when I really need a CLI tool that is likely to produce graphical artefacts.

Pro tip: if you always can put the current buffer in another frame with the M-x tear-off-window command. Works nicely when you need to keep Vterm in sight but out of your main editing space (e.g. to put it on a secondary monitor).

(use-package vterm
  :commands vterm
  (setq vterm-disable-bold-font nil)
  (setq vterm-disable-inverse-video nil)
  (setq vterm-disable-underline nil)
  (setq vterm-kill-buffer-on-exit nil)
  (setq vterm-max-scrollback 9999)
  (setq vterm-shell "/bin/bash")
  (setq vterm-term-environment-variable "xterm-256color")) Built-in terminals (fallback option)

term and ansi-term are built-in terminal emulators like Vterm. Not to be confused with the command line shells. They run inside of Emacs but are basically alien to the rest of the Emacs milieu: they do not reuse standard key bindings like C-n.

Only call those if you absolutely need them AND you have no access to either Vterm or a standalone, fully fledged terminal emulator.

As far as I can tell, based on reading the comments in term.el and elsewhere in the docs, the major difference between term and ansi-term is the ability of the latter to run one or multiple buffers simultaneously. Better check the documentation for this point. It does not seem to be a strong point, since this is also possible with the other options in the Emacs space (e.g. with C-u M-x shell).

(use-package term
  :commands (term ansi-term)
  (setq term-buffer-maximum-size 9999)
  (setq term-completion-autolist t)
  (setq term-completion-recexact t)
  (setq term-scroll-to-bottom-on-output nil))

5.7 Tools for manual pages (manpages)

Emacs offers a couple of commands for reading manual pages: man and woman. The former relies on the standard Unix tools, while the latter is an elisp implementation of the same idea. As I only ever run a GNU/Linux system, I am okay with just man.

Why bother?

  • All the goodies of consistency: fonts, themes, operating on text with your familiar Emacs functionality, handling buffers…
  • Each manpage provides direct links to other items it references.

What you can do inside such a buffer (with minor tweaks by me):

  • Hit i to go to the information node you want using completion (same principle as with the Info pages of C-h i and the like).
  • g will generate the buffer anew. Do it to reformat the text manually, though this should also happen automatically when adjusting a window's size.
  • n and p move between section headings.
  • Hit RET while over a referenced manpage to produce a new buffer with its contents.
  • s takes you directly to the familiar "See Also" section.
  • Use m to search for another manpage using your completion framework. If you invoke this command while point is over a referenced manpage, it becomes the default choice (same concept as with common help commands, C-h f, C-h v, and with many others like find-library).

Need to filter out those man buffers? Check my Ibuffer entry.

While there are customisation options for this tool, I find the defaults to work as expected. Note that the capitalisation of those symbols is canonical.

(use-package man
  :bind (:map Man-mode-map
              ("i" . Man-goto-section)
              ("g" . Man-update-manpage)))

5.8 Proced (process monitor, similar to `top')

This is a built-in tool that allows you to monitor running processes and act on them accordingly. These are the basic settings I have right now. Would need to experiment with it a bit more. It works fine though.

(use-package proced
  :commands proced
  (setq proced-auto-update-flag t)
  (setq proced-auto-update-interval 1)
  (setq proced-descend t)
  (setq proced-filter 'user))

And with this nimble tool we get live narrowing of the list, based on the terms of our search.

(use-package proced-narrow
  :after proced
  :bind (:map proced-mode-map
              ("/" . proced-narrow)))

5.9 Pass interface (password-store)

The external pass program, aka "password-store", is a password manager that uses GPG and standard UNIX tools to handle passwords. Encrypted files are stored in a plain directory structure. Very simple, very nice: now all data is available with a variety of interfaces, such as standard CLI, a dmenu interface, a graphical front-end like qtpass, etc.

The package below provides an Emacs interface to some of the most common actions, in the form of a list of candidates that can be narrowed down (such as with icomplete). I use it to quickly store a password to the kill ring.

(use-package password-store
  :commands (password-store-copy
  (setq password-store-time-before-clipboard-restore 30))

And this one adds a major mode for browsing the pass keychain. Call it with M-x pass. There is a helpful section at the top with key bindings and their functions.

(use-package pass
  :commands pass)

5.10 Elfeed (feed reader for RSS/Atom)

Settings for the feed reader package. I mostly care about the unique buffers tweak. It allows me to open a feed entry and keep it around while I go on browsing the feed list.

(use-package elfeed
  :commands elfeed
  (setq elfeed-use-curl t)
  (setq elfeed-curl-max-connections 10)
  (setq elfeed-db-directory "~/.emacs.d/elfeed")
  (setq elfeed-enclosure-default-dir "~/Downloads")
  (setq elfeed-search-clipboard-type 'CLIPBOARD)
  (setq elfeed-search-title-max-width 100)
  (setq elfeed-search-title-min-width 30)
  (setq elfeed-search-trailing-width 16)
  (setq elfeed-show-truncate-long-urls t)
  (setq elfeed-show-unique-buffers t)

  (defun prot/elfeed-feeds ()
    "Loads a file with RSS/Atom feeds.  This file contains valid
syntax for use by the `elfeed' package."
    (let ((feeds "~/.emacs.d/feeds.el.gpg"))
      (when (file-exists-p feeds)
        (load-file feeds))))

  (defun prot/elfeed-eww-other-window ()
    "Browse `elfeed' link at point in `eww' other window."
    (let* ((entry (if (eq major-mode 'elfeed-show-mode)
                      elfeed-show-entry (elfeed-search-selected :single)))
           (link (elfeed-entry-link entry)))
      (eww link)))

  :hook (after-init-hook . prot/elfeed-feeds)  ; Hook should be elfeed-specific
  :bind (:map elfeed-search-mode-map
              ("w" . elfeed-search-yank)
              ("g" . elfeed-update)
              ("G" . elfeed-search-update--force)
              ("o" . prot/elfeed-eww-other-window)
              :map elfeed-show-mode-map
              ("w" . elfeed-show-yank)))

5.11 Emacs HTML parser


As far as I can tell, the following shr-* variables concern an HTML parser that is used by a variety of tools, including Elfeed (defined right above). I guess we could scope them by using hooks, but I see no need for different settings.

What these do:

  • Open links in a new Emacs window, instead of the system's browser. This Emacs web browser is called eww.
  • Use monospaced fonts, since that is what I want to have everywhere in Emacs.
  • Do not preserve colours from websites, as they may be inaccessible (see my Modus theme).
  • Keep images to 70% of the window. This number is arbitrary. It just feels like a good upper limit (not a fan of decorative images inside of blog posts).
  • Line length at same number of characters as fill-column (defined elsewhere in this doc at 72).
(use-package shr
  (setq shr-use-fonts nil)
  (setq shr-use-colors nil)
  (setq shr-max-image-proportion 0.7)
  (setq shr-width (current-fill-column)))

Support the HTML pre tag with proper syntax highlighting. Got this snippet directly from its GitHub project page.

(use-package shr-tag-pre-highlight
  :after shr
  (add-to-list 'shr-external-rendering-functions
               '(pre . shr-tag-pre-highlight))
  (when (version< emacs-version "26")
    (with-eval-after-load 'eww
      (advice-add 'eww-display-html :around

5.12 Emacs Web Wowser (EWW)


(use-package eww
  :commands (eww
  (setq eww-restore-desktop t)
  (setq eww-header-line-format "%u")

  (defun prot/eww-visit-history ()
    "Revisit `eww' history using completion."
    (let ((history eww-prompt-history))
      (icomplete-vertical-do ()
         (completing-read "Visit website from history: " history nil t)))))

  ;; (declare-function cl-remove-if "cl-seq")
  ;; (defun prot/eww-list-buffers ()
  ;;   (interactive)
  ;;   (let ((bufs (cl-remove-if (lambda (x)
  ;;                               (with-current-buffer x
  ;;                                 (if (derived-mode-p 'eww-mode)
  ;;                                     x
  ;;                                   nil)))
  ;;                             (buffer-list))))
  ;;      (completing-read "Visit EWW buffer: " bufs nil t)))

  (defvar prot/eww-global-map
    (let ((map (make-sparse-keymap)))
      (define-key map "s" 'eww-search-words)
      (define-key map "o" 'eww-open-in-new-buffer)
      (define-key map "f" 'eww-open-file)
      (define-key map "w" 'prot/eww-visit-history)
    "Key map to scope `eww' bindings for global usage.
The idea is to bind this to a prefix key, so that its defined
keys follow the pattern of <PREFIX> <KEY>.")
  :bind-keymap ("C-c w" . prot/eww-global-map))

(use-package browse-url
  :after eww
  (setq browse-url-browser-function 'eww-browse-url))

5.13 Emacs IRC client


My .authinfo.gpg entry for IRC looks like this:

machine port 6667 login NAME password PASS

The prot/auth-get-field is defined in Base email settings (may be better to have it elsewhere, but I keep it now together with the email configurations).

(use-package erc
  :disabled                             ; Needs testing and refinements
  (setq erc-prompt-for-password nil)

  (defun prot/erc-dwim ()
    "Switch to latest `erc' buffer or log in."
    (let* ((irc "")
           (nick (prot/auth-get-field irc :user))
           ;; Gets the password, but does not read(?) it
           (pass (prot/auth-get-field irc :secret))
           (bufs (erc-buffer-list)))
      (if bufs
          (erc-track-switch-buffer 1)
        (erc :server irc
             :nick nick
             :password pass             ; Does not work
             :full-name nick))))
  :bind ("C-c e" . prot/erc-dwim))

5.14 Bongo (Music player)

I already tried EMMS and various other options but did not stick with them. I felt I was missing something or maybe I just tested them too early into my Emacs journey. Now using Bongo and am quite happy with it.

Concerning the customisations below, these can be summarised thus:

  • Hide icons.
  • No mode line indicators.
  • Do not ask for directory tree insertion.
  • With Dired, the "Music" directory doubles as a Bongo library (see prot/bongo-dired-library and the relevant hook).
  • Because of the above, prefer playlist buffers (pro tip: you can use dired-jump inside of a playlist buffer to switch to that directory—see my Dired section for the relevant configs).
  • While contrib/bongo-add-dired-file integrates Dired mark command with Bongo. The function is provided in this Emacs Wiki entry (minor tweaks by me).

The way I play music is very simple. I load up a directory tree with a bunch of audio files. Then I C-u C-c C-r from inside a Bongo buffer to play the tracks in random order. Done! I rarely switch tracks manually and change playlists in regular intervals (a directory tree typically contains hundreds of music files).

This sequence is conveniently mapped to C-RET inside of the Bongo Library buffer (so the Dired buffer of ~/Music and its sub-directories). The command will operate on the directory at point or on the marked items, if they exist. Note that I used to bind that action to just SPC but I realised it would interfere with tasks in wdired (and probably elsewhere).

I have a couple of videos about my workflow with Bongo and Dired:

(use-package bongo
  :commands bongo
  (setq bongo-default-directory "~/Music")
  (setq bongo-prefer-library-buffers nil)
  (setq bongo-insert-whole-directory-trees t)
  (setq bongo-logo nil)
  (setq bongo-action-track-icon nil)
  (setq bongo-display-track-icons nil)
  (setq bongo-display-track-lengths nil)
  (setq bongo-display-header-icons nil)
  (setq bongo-display-playback-mode-indicator t)
  (setq bongo-display-inline-playback-progress nil)
  (setq bongo-mark-played-tracks nil)
  (setq bongo-header-line-mode nil)
  (setq bongo-header-line-function nil)
  (setq bongo-mode-line-indicator-mode nil)
  (setq bongo-enabled-backends '(vlc mpv))
  (setq bongo-vlc-program-name "cvlc")

  (defun contrib/bongo-add-dired-files ()
    "Add marked files inside of a Dired buffer to the Bongo library"
    (let (file-point file (files nil))
       (setq file-point (dired-move-to-filename)
             file (dired-get-filename)
             files (append files (list file)))
       nil t)
        (set-buffer bongo-default-playlist-buffer-name)
        (mapc 'bongo-insert-file files))))

  (defun prot/bongo-dired-library ()
    "Set `bongo-dired-library-mode' when accessing ~/Music.

This is meant to be hooked to `dired-mode'.  Upon activation, the
directory and all its sub-directories become a valid library
buffer for Bongo, from where we can, among others, add tracks to
playlists.  The added benefit is that Dired will continue to
behave as normal, making this a superior alternative to a
purpose-specific library buffer."
    (when (string-match-p "\\`~/Music/" default-directory)
      (set (make-local-variable 'bongo-dired-library-mode) 't)))

  (defun prot/bongo-clear-playlist-and-stop ()
    "Stop playback and clear the entire `bongo' playlist buffer.

Contrary to the standard `bongo-erase-buffer', this also removes
the currently-playing track."
    (when (bongo-playlist-buffer-p)

  (defun prot/bongo-play-random ()
    "Play a random track with `bongo' and set random playback."
    (when (or (bongo-playlist-buffer-p)
      (bongo-random-playback-mode 1)))

  (defun prot/bongo-library-insert-and-play-random ()
    "Add directory tree or marked items to the `bongo' playlist.
Create the playlist buffer if necessary.

This is meant to work while inside a `dired' buffer that doubles
as a library buffer (see `prot/bongo-dired-library')."
    (when (bongo-library-buffer-p)
      (unless (bongo-playlist-buffer-p)

  (defun prot/bongo-insert-playlist ()
    "Insert the contents of a `bongo' playlist.

The files are stored in a predetermined path inside my Music

Upon insertion, playback starts immediately, in accordance with
    (let* ((path "~/Music/playlists/")
           (dotless directory-files-no-dot-files-regexp)
           (playlists (mapcar
                       (directory-files path t dotless))))
      (when (bongo-playlist-buffer-p)
         (completing-read "Select playlist: " playlists nil t))

  :hook ((dired-mode-hook . prot/bongo-dired-library)
         (bongo-playlist-mode-hook . hl-line-mode))
  :bind (("<C-XF86AudioPlay>" . bongo-pause/resume)
         ("<C-XF86AudioNext>" . bongo-next)
         ("<C-XF86AudioPrev>" . bongo-previous)
         ("<M-XF86AudioPlay>" . bongo-show)
         :map bongo-playlist-mode-map
         ("C-d" . prot/bongo-clear-playlist-and-stop)
         ("I" . prot/bongo-insert-playlist)
         :map bongo-dired-library-mode-map
         ("<C-return>" . prot/bongo-library-insert-and-play-random)))

6 General interface and interactions

This section contains configurations for all aspects of the Emacs user interface, as well lots of small or self-contained tweaks that cover a wide range of built-in libraries.

6.1 Disable GUI components

Overview of these settings:

  • I normally use the GTK (GUI) variant of Emacs. I prefer not to have any of the elements that come with it. This keeps the window clean. The only "interface" component that remains in place is the mode line, which is not part of the GUI toolkit anyway…
  • The start-up screen that offers an overview of GNU Emacs is also disabled. It is useful for beginners, but is rendered obsolete once you familiarise yourself with the essentials.
  • The pair of key bindings that involve z minimise the Emacs frame. This is technically an interface action, in that it assumes my window manager has a minimise function or that I want to have such a motion inside of Emacs. Disable them.
  • Also disable the "hello file" function, because it crashes Emacs. I assume this has to do with font rendering and missing font files, as I experienced similar issues on various terminal emulators.
(use-package emacs
  (menu-bar-mode -1)
  (tool-bar-mode -1)
  (scroll-bar-mode -1)
  (setq use-file-dialog nil)
  (setq use-dialog-box t)               ; only for mouse events
  (setq inhibit-splash-screen t)
  (global-unset-key (kbd "C-z"))
  (global-unset-key (kbd "C-x C-z"))
  (global-unset-key (kbd "C-h h")))

6.2 Modus themes (my very accessible themes) and other visuals

I am using my own themes. They are designed to conform with the highest accessibility standard for colour contrast between foreground and background values. This stands for a minimum contrast ratio of 7:1, also known as the WCAG AAA standard. In simpler terms, they are good for readability.

I call this project "Modus themes". It consists of "Modus Operandi" (light theme) and "Modus Vivendi" (dark). The source code and installation instructions are available on their GitLab page. The list of supported packages is comprehensive. There also are lots of customisation options to tweak the looks of the themes.

The two themes are distributed as standalone packages under the terms of the GNU General Public License. You can find them in the official ELPA repository (latest tagged release), MELPA (latest commit to the master branch), as well as MELPA Stable (latest tag). Each theme is its own package because I know that people tend to use one or the other. And also due to the fact that one is not a prerequisite for the other. I personally use both, switching between on-the-fly by calling prot/modus-themes-toggle.

Note though that because I am using these themes locally as part of their development process, I am not configuring here the ELPA/MELPA versions. So what you see in this use-package declaration is a bit different than what others will be using.

Also note that the values I set for these variables are not indicative of my preferences: I try different combinations to test things across a range of scenaria. The scale for the fonts I specify here is called "minor second", representing a 1.125 rate.

The hook contrib/after-load-theme-hook and its accompaniments are taken from the Centaur Emacs project.

Lastly, if you are curious about the underlying methodology, read my essay on the on the design of the Modus themes (2020-03-17).

(use-package emacs
  (setq custom-safe-themes t)           ; Due to my dev needs

  ;; The variables do not reveal my preferences.  I am always testing
  ;; things
  (defun prot/modus-operandi ()
    "Enable some Modus Operandi variables and load the theme.
This is used internally by `prot/modus-themes-toggle'."
    (setq modus-operandi-theme-slanted-constructs t
          modus-operandi-theme-bold-constructs nil
          modus-operandi-theme-visible-fringes nil
          modus-operandi-theme-3d-modeline t
          modus-operandi-theme-subtle-diffs nil
          modus-operandi-theme-intense-standard-completions t
          modus-operandi-theme-distinct-org-blocks nil
          modus-operandi-theme-proportional-fonts t
          modus-operandi-theme-rainbow-headings nil
          modus-operandi-theme-section-headings t
          modus-operandi-theme-scale-headings nil
          modus-operandi-theme-scale-1 1.125
          modus-operandi-theme-scale-2 1.266
          modus-operandi-theme-scale-3 1.424
          modus-operandi-theme-scale-4 1.602
          modus-operandi-theme-scale-5 1.802)
    ;;;; NOTE check the project's README for the following one.  Only
    ;;;; for advanced usage.
    ;; modus-operandi-theme-override-colors-alist
    (load-theme 'modus-operandi t))

  (defun prot/modus-vivendi ()
    "Enable some Modus Vivendi variables and load the theme.
This is used internally by `prot/modus-themes-toggle'."
    (setq modus-vivendi-theme-slanted-constructs t
          modus-vivendi-theme-bold-constructs nil
          modus-vivendi-theme-visible-fringes nil
          modus-vivendi-theme-3d-modeline t
          modus-vivendi-theme-subtle-diffs nil
          modus-vivendi-theme-intense-standard-completions t
          modus-vivendi-theme-distinct-org-blocks nil
          modus-vivendi-theme-proportional-fonts t
          modus-vivendi-theme-rainbow-headings nil
          modus-vivendi-theme-section-headings t
          modus-vivendi-theme-scale-headings nil
          modus-vivendi-theme-scale-1 1.125
          modus-vivendi-theme-scale-2 1.266
          modus-vivendi-theme-scale-3 1.424
          modus-vivendi-theme-scale-4 1.602
          modus-vivendi-theme-scale-5 1.802)
    ;;;; NOTE check the project's README for the following one.  Only
    ;;;; for advanced usage.
    ;; modus-vivendi-theme-override-colors-alist
    (load-theme 'modus-vivendi t))

  (defun prot/modus-themes-custom ()
    "Override faces by accessing the theme colour variables.
Run this via `contrib/after-load-theme-hook'."
     ((eq (car custom-enabled-themes) 'modus-operandi)
      (modus-operandi-theme-with-color-variables ; NOTE check the project's README for this one
         `(cursor ((,class (:background ,blue-alt)))))))
     ((eq (car custom-enabled-themes) 'modus-vivendi)
      (modus-vivendi-theme-with-color-variables ; NOTE check the project's README for this one
         `(cursor ((,class (:background ,red-alt)))))))))

  (defvar contrib/after-load-theme-hook nil
    "Hook run after a color theme is loaded using `load-theme'.")

  (defun contrib/run-after-load-theme-hook (&rest _)
    "Run `contrib/after-load-theme-hook'."
    (run-hooks 'contrib/after-load-theme-hook))

  (advice-add #'load-theme :after #'contrib/run-after-load-theme-hook)

  (defvar prot/modus-themes-toggle-hook nil
    "Hook that runs after `prot/modus-themes-toggle' is invoked.")

  (defun prot/modus-themes-toggle ()
    "Toggle between `prot/modus-operandi' and `prot/modus-vivendi'.
Also run `prot/modus-themes-toggle-hook'."
    (if (eq (car custom-enabled-themes) 'modus-operandi)
          (disable-theme 'modus-operandi))
      (disable-theme 'modus-vivendi))
    (run-hooks 'prot/modus-themes-toggle-hook))

  :bind ("<f5>" . prot/modus-themes-toggle)
  :hook ((after-init-hook . prot/modus-operandi)
         (contrib/after-load-theme-hook . prot/modus-themes-custom)))

6.2.1 Measuring relative colour luminance

The following is an Elisp implementation of the formula that determines whether a combination of two colours is accessible or not. It was offered to me by Omar Antolín Camarena. For the purposes of my themes, the contrast ratio needs to be 7:1 or higher, which conforms with the highest accessibility standard (WCAG AAA).

To see it in action, you can read the org-mode report on the minor review of Modus Operandi palette (2020-05-10). Which also goes to show how tricky the selection of colours can be.

(defun wcag (hex)
  (apply #'+
          (lambda (k x)
            (* k (if (<= x 0.03928)
                     (/ x 12.92)
                   (expt (/ (+ x 0.055) 1.055) 2.4))))
          '(0.2126 0.7152 0.0722)
          (color-name-to-rgb hex))))

(defun clr (c1 c2)
  (let ((ct (/ (+ (wcag c1) 0.05)
               (+ (wcag c2) 0.05))))
    (max ct (/ ct))))

6.2.2 Mode line

The mode line is an integral part of the Emacs interface. While there are lots of third party packages that style it in a variety of ways, I find the default to be particularly good.

I just configure it to reshuffle some of the indicators. Nothing too fancy.

Meanwhile, the mode-line-defining-kbd-macro is tweaked to use a more appropriate string for its indicator and colours that are designed specifically for the mode line (the default uses the generic font-lock warning face).

(use-package emacs
  (setq mode-line-percent-position '(-3 "%p"))
  (setq mode-line-defining-kbd-macro
        (propertize " Macro" 'face 'mode-line-emphasis))
  (setq-default mode-line-format
                  "  "
                  (vc-mode vc-mode)
                  " "
                  " "

In the following sub-sections I provide customisations for some tools that place information on the mode line. Again, nothing flamboyant. Battery status

Emacs offers a built-in library for presenting information about the status of the laptop's battery. Using it allows me to eliminate my dependence on the system panel and thus keep Emacs in full screen view without any interruptions.

The default update interval is set to a single minute (in seconds), which is generally fine though I find that a slightly higher value works just as well. As for the format, it is designed to show a context-dependent, single character indicator about the current status, as well as the battery's overall percentage.

Variable battery-mode-line-limit will hide the indicator if the value is above the declared threshold. 99 basically means "full". I use that instead of a 100 because sometimes the battery only ever fills up to 99.99, meaning that the indicator remains present at all times.

(use-package battery
  (setq battery-mode-line-format " [%b%p%%]")
  (setq battery-mode-line-limit 99)
  (setq battery-update-interval 180)
  (setq battery-load-low 20)
  (setq battery-load-critical 10)
  :hook (after-init-hook . display-battery-mode)) Display current time

I normally use Emacs in fullscreen view. No system panels, no window decorations, no icons and blinking indicators. Nothing to distract me. While I really like this environment, sometimes I need to take a look at the time… Thankfully Emacs offers a convenient, built-in way of displaying such information in the mode line.

The display-time-format can be configured to show the current date and time in all the various formats we would expect, using a string of specifiers (find the docs with C-h v format-time-string). Setting its value to nil means that the information on display will be the combined result of display-time-24hr-format and display-time-day-and-date. I prefer to just write a string directly, keeping those two inactive.

The display-time-mode can output more than just the current time. It also shows the load average and an email indicator. I only need the time and date. The rest is noise.

(use-package time
  (setq display-time-format "%H:%M  %Y-%m-%d")
  ;;;; Covered by `display-time-format'
  ;; (setq display-time-24hr-format t)
  ;; (setq display-time-day-and-date t)
  (setq display-time-interval 60)
  (setq display-time-mail-directory nil)
  (setq display-time-default-load-average nil)
  :hook (after-init-hook . display-time-mode)) Keycast mode

Once enabled, this package uses the mode line to show the keys being pressed and the command they call. It is quite useful for screen casting.

The placement of the indicator is controlled by keycast-window-predicate which I set to the current window. The moody library offers that specific piece of functionality. Moody can also be used to customise the looks of the mode line, though I do not want that.

The tweaks to the keycast-substitute-alist prevent the display of self-inserting characters and some other commands that are not particularly useful while screen casting. Now the indicator will only show commands, which looks cleaner. I got the idea and original piece of Elisp from the dotfiles of André Alexandre Gomes and then added a few tweaks of my own.

(use-package moody :ensure)

(use-package keycast
  :after moody
  :commands keycast-mode
  (setq keycast-window-predicate 'moody-window-active-p)
  (setq keycast-separator-width 1)
  (setq keycast-insert-after 'mode-line-buffer-identification)
  (setq keycast-remove-tail-elements nil)

  ;; TODO pending review
  (add-to-list 'keycast-substitute-alist '(mouse-event-p nil))
  (add-to-list 'keycast-substitute-alist '(mouse-movement-p nil))
  (add-to-list 'keycast-substitute-alist '(mwheel-scroll nil))
  ;; (add-to-list 'keycast-substitute-alist '(mouse-set-point nil))
  ;; (add-to-list 'keycast-substitute-alist '(mouse-set-region nil))
  ;; (add-to-list 'keycast-substitute-alist '(mouse-drag-secondary nil))
  ;; (add-to-list 'keycast-substitute-alist '(mouse-drag-line nil))
  ;; (add-to-list 'keycast-substitute-alist '(mouse-drag-drag nil))
  ;; (add-to-list 'keycast-substitute-alist '(mouse-start-end nil))
  ;; (add-to-list 'keycast-substitute-alist '(mouse-drag-region nil nil))
  ;; (add-to-list 'keycast-substitute-alist '(mouse-drag-track nil nil))
  ;; (add-to-list 'keycast-substitute-alist '(mouse-drag-region-rectangle nil))
  ;; (add-to-list 'keycast-substitute-alist '(mouse-drag-and-drop-region nil))
  ;; (add-to-list 'keycast-substitute-alist '(mwheel-event-button nil))
  ;; (add-to-list 'keycast-substitute-alist '(dframe-mouse-event-p nil))
  ;; (add-to-list 'keycast-substitute-alist '(mouse-drag-events-are-point-events-p nil))
  (add-to-list 'keycast-substitute-alist '(self-insert-command "." "Typing…"))
  (add-to-list 'keycast-substitute-alist '(org-self-insert-command "." "Typing…")))

6.2.3 Window divider mode

This is a built-in mode that draws vertical window borders in a slightly different way than the default, which I find more consistent. Only using it because of that, though it can also adjust the side of the borders as well as their placement.

(use-package emacs
  (setq window-divider-default-right-width 1)
  (setq window-divider-default-bottom-width 1)
  (setq window-divider-default-places 'right-only)
  :hook (after-init-hook . window-divider-mode))

6.2.4 Fringe mode

The fringes are areas to the right and left side of an Emacs frame. They can be used to show status-related or contextual feedback such as line truncation indicators, continuation lines, code linting markers, etc.

The default fringe width (nil) is 8 pixels on either side, which I approve of. It is possible to set the value of the fringe-mode to something like '(10 . 5) which applies the varied width to the left and right side respectively. Otherwise, we can use a single integer that controls both sides.

The use of setq-default is necessary, otherwise these values become buffer-local.

(use-package fringe
  (fringe-mode nil)
  (setq-default fringes-outside-margins nil)
  (setq-default indicate-buffer-boundaries nil)
  (setq-default indicate-empty-lines nil)
  (setq-default overflow-newline-into-fringe t))

6.2.5 Diff highlights in the fringe

The diff-hl package uses the left or right fringe to highlight changes in the current buffer. The indicators are colour-coded to denote whether a change is an addition, removal, or change that includes a bit of both.

The package offers some more features, such as the ability to move between diff hunks while editing the buffer. I still need to experiment with those before customising them to my liking.

At any rate, this package is meant a general tool for version control systems, rather than a git-specific one. Much like the built-in VC (see section on the generic version control (VC) framework).

(use-package diff-hl
  :after vc
  (setq diff-hl-draw-borders nil)
  (setq diff-hl-side 'left)
  :hook (after-init-hook . global-diff-hl-mode))

6.2.6 Optional visual indicators or layout elements

This is a collection of modes or interfaces I seldom use or, rather, I use under special circumstances. They are useful, but there is not need for them to be available at all times. Current line highlight (hl-line-mode)

This is a mode that I only activate via hooks for certain buffers where the current line itself is more important that the actual column (e.g. in Dired buffers). Here I configure it so that the highlight applies only to the current window. There is also a "global" variant, for when the equivalent mode is used (I have no plan to use that).

(use-package hl-line
  (setq hl-line-sticky-flag nil)) Rainbow mode for colour testing

The following package reads a colour value, such as hexadecimal RGB, and sets the background for the value in that colour. Quite useful when reviewing my themes (rainbow-mode is activated manually).

(use-package rainbow-mode
  :commands rainbow-mode
  (setq rainbow-ansi-colors nil)
  (setq rainbow-x-colors nil)) Toggles for line numbers and whitespace indicators
Display line numbers (buffer-local)
I seldom use line numbers, but here it is. This toggles the setting for the local buffer. A global option is also available, but I prefer the buffer-specific variant because there are contexts where global display is not useful (such as occur).
Display invisible characters (whitespace)
Viewing invisible characters (whitespace) can be very helpful under certain circumstances. Generally though, I do not keep it active.

As for delete-trailing-whitespace, I prefer to call it manually because sometimes it causes problems, such as with diffs.

(use-package whitespace
  (defun prot/toggle-invisibles ()
    "Toggles the display of indentation and space characters."
    (if (bound-and-true-p whitespace-mode)
        (whitespace-mode -1)
  :bind (("<f6>" . prot/toggle-invisibles)
         ("C-c z" . delete-trailing-whitespace)))

(use-package display-line-numbers
  (defun prot/toggle-line-numbers ()
    "Toggles the display of line numbers.  Applies to all buffers."
    (if (bound-and-true-p display-line-numbers-mode)
        (display-line-numbers-mode -1)
  :bind ("<f7>" . prot/toggle-line-numbers)) “Focus mode” for writing (olivetti-mode)

I spend much of my time in Emacs reading and writing long form texts. It is nice to be able to easily toggle a mode that centres the buffer, allowing for greater comfort.

Olivetti fulfils that niche very nicely. It is not aggressive in its interface requirements, respects my existing line settings and with my preference for auto-filling text, while it does not introduce any kind of functionality beyond the scope of bringing the current window's buffer to the centre of the view. This is exactly what I need. Any other enhancement, such as a larger font size can be handled by another function.

The prot/olivetti-mode offers a toggle for activating this mode, passing additional font-related parameters (see base typeface configurations) and removing the fringe on both sides of the selected window (check my fringe-mode settings).

Note that for my simple presentations inside of Emacs (org-tree-slide) I use darkroom instead of olivetti. They offer similar features, except that the former hides the mode line, giving the faux slide-show effect that is so pertinent to presentations.

(use-package olivetti
  (setq olivetti-body-width 100)
  (setq olivetti-minimum-body-width 80)
  (setq olivetti-recall-visual-line-mode-entry-state t)

  (defun prot/olivetti-mode ()
    "Toggle `olivetti-mode' with additional parameters.
Fringes are disabled for the current window.  For the
font-related changes see `prot/variable-pitch-mode'."
    (if (bound-and-true-p olivetti-mode)
          (olivetti-mode -1)
          (set-window-fringes (selected-window) nil) ; Use default width
      (olivetti-mode 1)
      (set-window-fringes (selected-window) 0 0)
  :bind ("C-c o" . prot/olivetti-mode)) Rainbow blocks

This package is quite useful when debugging highly structured code that you are not familiar with. It will highlight an entire code block in a single colour, making it easier to understand the overall structure (my Modus themes support it, of course).

Also note that there is another package that applies a rainbow effect only to the delimiters. Between the two, I prefer this one. At any rate, I activate rainbow-blocks-mode manually, when I feel that I am missing something that I cannot spot right away.

A less intrusive, built-in alternative is to set the variable show-parent-style 'expression (see my configs for parentheses).

(use-package rainbow-blocks
  :commands rainbow-blocks-mode
  (setq rainbow-blocks-highlight-braces-p t)
  (setq rainbow-blocks-highlight-brackets-p t)
  (setq rainbow-blocks-highlight-parens-p t))

6.3 Language settings for prose and code

This section is all about configurations and packages that deal with natural or programming language enhancements.

6.3.1 Line length (column count)

The column count is set to 72. The standard line length is 80 characters, so having it at something less allows for such things as quoting plain text, indenting, etc. git commit messages also make good use of this method. The column count is used by auto-fill-mode and similar tools (or when manually invoking text formatting with M-q).

(use-package emacs
  (setq-default fill-column 72)
  (setq sentence-end-double-space t)
  (setq sentence-end-without-period nil)
  (setq colon-double-space nil)
  :hook (after-init-hook . column-number-mode))

6.3.2 Recognise subwords

It is better you do C-h f subword-mode. Basically, this alters the way Emacs understands word boundaries. So, camelCaseWords are exposed as their constituents rather than one long word, meaning that motions will behave accordingly.

(use-package subword
  :hook (prog-mode-hook . subword-mode))

6.3.3 Auto-fill plain text and comments

Make sure we run the minor mode that keeps paragraphs within the column limit I prefer: you can always do it manually for the current paragraph with M-q. By targeting text-mode we also affect every major mode derived from it, which means that we correctly do not get this utility in programming-related modes (in principle, those come from prog-mode). The adaptive mode improves the handling of things like bulleted and numbered lists, where it recognises the text's prefix and tries to align to it rather than the absolute beginning of the line.

(use-package emacs
  :diminish auto-fill-function
  (setq adaptive-fill-mode t)
  :hook (text-mode-hook . (lambda ()

6.3.4 Comment lines, regions, boxes, etc.

Just some basic configurations for commenting structured text. This is mostly a placeholder for potentially more targeted and detailed settings that would involve per-mode hooks.

The purpose of my reviewed key bindings is to make them more consistent. Helps with mnemonics. They also are more ergonomic. To this end, I have the following:

  • The standard commenting function is now bound to the simple C-;. This runs a "do what I meant" function I have defined, whose detailed documentation can be read below.
  • C-: (C-S-;) will kill the comment on the current line. This is particularly helpful when the comment follows text you would like to keep. The operation can be performed regardless of where the point is on the line. Some modes disable this behaviour (e.g. trying it on source code inside of org-mode—for those cases, focus the block with C-c ').
  • The M-; will just append a comment to the line, rather than the default comment-dwim.

Note that C-; is occupied by some flyspell command that I have no use for (disabled in the relevant package declaration).

Lastly, use M-j (alias C-M-j) when you want to continue an existing comment on a new line with respect for the current indentation. If you are not inside of a comment, this will just create an indentation-aware new line.

(use-package newcomment
  (setq comment-empty-lines t)
  (setq comment-fill-column nil)
  (setq comment-multi-line t)
  (setq comment-style 'multi-line)

  (defun prot/comment-dwim (&optional arg)
    "Alternative to `comment-dwim': offers a simple wrapper
around `comment-line' and `comment-dwim'.

If the region is active, then toggle the comment status of the
region or, if the major mode defines as much, of all the lines
implied by the region boundaries.

Else toggle the comment status of the line at point."
    (interactive "*P")
    (if (use-region-p)
        (comment-dwim arg)
        (comment-line arg))))

  :bind (("C-;" . prot/comment-dwim)
         ("C-:" . comment-kill)
         ("M-;" . comment-indent)
         ("C-x C-;" . comment-box)))

6.3.5 Flyspell (spell check)

I need spell checking for both English and Greek. In previous versions of this section I had configurations that would automate spell checking for both languages at once. It worked but was rather slow.

Upon further inspection, I have realised that I seldom need to work in mixed language circumstances. Moreover, I now understand that I do not need to have spell checking always on. It can be activated manually, with the flyspell functions defined in the :commands segment below. It is thus possible to keep things simple and to switch dictionaries only when necessary.

Also bear in mind that the key binding C-; is disabled because I re-purpose that for a faster version of C-x C-; (much more useful for my work—see the section on comments).

The two functions I define are meant to ease my workflow, while superseding the default binding for ispell-word. What I normally use is my M-$, which will spell-check the active region or toggle the dictionaries I use. The C-M-$ is defined as a fallback option in case I have already highlighted a region and then realise I need the other dictionary.

Note that aspell and its dictionaries are not part of Emacs. I install the relevant packages from my distro's archive.

(use-package flyspell
  :commands (ispell-change-dictionary
  (setq flyspell-issue-message-flag nil)
  (setq flyspell-issue-welcome-flag nil)
  (setq ispell-program-name "aspell")
  (setq ispell-dictionary "en_GB")

  (defun prot/ispell-toggle-dictionaries ()
    "Toggle between English and Greek dictionaries."
    (if (string= ispell-current-dictionary "en")
        (ispell-change-dictionary "el")
      (ispell-change-dictionary "en")))

  (defun prot/flyspell-dwim (&optional beg end)
    "Run `flyspell-region' on the active region, else toggle the
ispell dictionaries with `prot/ispell-toggle-dictionaries'."
    (interactive "r")
      (if (use-region-p)
          (flyspell-region beg end)

  :bind (("M-$" . prot/flyspell-dwim)
         ("C-M-$" . prot/ispell-toggle-dictionaries)
         :map flyspell-mode-map
         ("C-;" . nil)))

6.3.6 Code linters

There are several code linting tools in the Emacs milieu. The two I configure here are front-ends to external programs for code analysis.

While it is possible to run the linter globally or through hooks, I prefer to only do so manually on a per-buffer basis. There are many cases where I am merely browsing code, where whatever diagnostics are of no import to the task at hand.

With regard to the defhydra calls in this section, remember to refer to my fundamentals for hydras. Flymake

This is a built-in linter interface that can be extended to support any programming language. I do, however, feel that the alternative Flycheck (see below) offers a more refined experience, though I still am hesitant to reach any conclusions.

(use-package flymake
  :commands flymake-mode
  (setq flymake-fringe-indicator-position 'left-fringe)
  (setq flymake-suppress-zero-counters t)
  (setq flymake-start-on-flymake-mode t)
  (setq flymake-no-changes-timeout nil)
  (setq flymake-start-on-save-buffer t)
  (setq flymake-proc-compilation-prevents-syntax-check t)
  (setq flymake-wrap-around nil)

  (defhydra prot/hydra-flymake (:color pink :hint nil)
_n_: Next error
_p_: Previous error
_d_: Diagnostics' buffer
    ("d" flymake-show-diagnostics-buffer)
    ("n" flymake-goto-next-error)
    ("p" flymake-goto-prev-error)
    ("q" nil "cancel" :color blue))
  :bind (:map flymake-mode-map
              ("C-c h l" . prot/hydra-flymake/body))) Flycheck

This linting interface has excellent coverage for a broad range of languages, while there are lots of extensions to it to further broaden its scope.

Flycheck has some subtle visual and functional advantages over the built-in Flymake (see above), though the most obvious one is the overall activity around its development.

(use-package flycheck
  :commands flycheck-mode
  (setq flycheck-check-syntax-automatically
        '(save mode-enabled))

  (defun prot/flycheck-list-errors-toggle ()
    "Toggle the display of `flycheck-mode' diagnostics' buffer."
    (let ((diagnostics (get-buffer-window flycheck-error-list-buffer)))
      (unless flycheck-mode
        (user-error "Flycheck mode not enabled"))
      (if diagnostics
          (delete-window diagnostics)

  (defhydra prot/hydra-flycheck (:color pink :hint nil)
   ^Actions^             ^Helpers^
_n_: Next error       _c_: Check buffer
_p_: Previous error   _l_: List diagnostics
_e_: Explain error    _x_: Disable checker
_d_: Display error
    ("l" prot/flycheck-list-errors-toggle)
    ("c" flycheck-buffer)
    ("n" flycheck-next-error)
    ("p" flycheck-previous-error)
    ("e" flycheck-explain-error-at-point)
    ("d" flycheck-display-error-at-point)
    ("x" flycheck-disable-checker :color blue)
    ("q" nil "cancel" :color blue))
  :bind (:map flycheck-mode-map
              ("C-c h l" . prot/hydra-flycheck/body))) Flycheck mode line indicator

While Flymake has a good mode line status area, Flycheck's leaves something to be desired. This package bridges that gap. It offers a visual experience that involves both text and purpose-specific faces (set of styles).

By default, the status feedback in flycheck-indicator-status-icons employs descriptive terms. It can, nonetheless, be configured to show characters instead of long strings (which I guess opens the sluicegate for the flow of iconography, emojis, unicode symbols…). I prefer to reduce verbosity using common text characters, at the potential cost of obscuring the meaning of some of the less common types of feedback (what does "suspicious" mean in this context, anyway?).

I also tweak the characters that accompany all three classes of report (notes, warnings, errors) in an attempt to distinguish between them.

If you want to be creative with your choice of symbol, note that while trying to complete a description through C-x 8 RET, you can press ? to list all candidates in the "Completions" buffer with a preview next to each entry. Neat!

(use-package flycheck-indicator
  :after flycheck
  (setq flycheck-indicator-icon-error (string-to-char "!"))
  (setq flycheck-indicator-icon-info (string-to-char "·"))
  (setq flycheck-indicator-icon-warning (string-to-char "*"))
  (setq flycheck-indicator-status-icons
        '((not-checked "%")
          (no-checker "-")
          (running "&")
          (errored "!")
          (finished "=")
          (interrupted "#")
          (suspicious "?")))
  :hook (flycheck-mode-hook . flycheck-indicator-mode)) Elisp packaging requirements

With these in place we can perform checks that pertain to Emacs lisp packaging. I use it for my themes but also for any elisp library I may want to send patches to.

The two packages here provide the necessary functionality for both of the linter front-ends configured above.

(use-package flycheck-package
  :after flycheck

(use-package package-lint-flymake
  :after flymake

6.3.7 Eldoc (elisp live documentation feedback)

When editing elisp, this mode will display useful information about the construct at point in the echo area. For functions it will display the list of arguments they accept. While it will show the the first sentence of a variable's documentation string.

At first, I dismissed this package. Upon closer inspection, it does offer a lightweight complementary facility to that of the standard help commands: C-h f FUNCTION, C-h v VARIABLE.

(use-package eldoc
  (global-eldoc-mode 1))

6.3.8 Support for various major modes

These provide syntax highlighting and additional features for environments that are not already supported by Emacs. Arch Linux package recipes

If you are on Arch or derivatives and check PKGBUILD files, this is for you.

(use-package pkgbuild-mode
  :mode ("PKGBUILD" . pkgbuild-mode)) Markdown

I edit lots of Markdown files. This makes things easier.

(use-package markdown-mode
  (setq markdown-fontify-code-blocks-natively t)
  :mode ("\\.md$" . markdown-mode)) YAML

This adds support for YAML files.

(use-package yaml-mode
  :mode (("\\.yml$" . yaml-mode)
         ("\\.yaml$" . yaml-mode))) CSS

This is the built-in mode for working with CSS and SCSS. I just want it to not apply previews to colour references. If I ever need that, there is rainbow-mode (see relevant section).

(use-package css-mode
  (setq css-fontify-colors nil))

6.3.9 Configure 'electric' behaviour

Emacs labels as "electric" any behaviour that involves contextual auto-insertion of characters. This is a summary of my settings:

  • Indent automatically.
  • If electric-pair-mode is enabled (which I might do manually), insert quotes and brackets in pairs. Only do so if there is no alphabetic character after the cursor.
  • The cryptic numbers in the pairs set, correspond to curly single and double quotes and these «». The contents of this set are always inserted in pairs, regardless of major mode.
    • To get those numbers, evaluate (string-to-char CHAR) where CHAR is the one you are interested in. For example, get the literal tab's character with (string-to-char "\t").
  • While inputting a pair, inserting the closing character will just skip over the existing one, rather than add a new one. So typing ( will insert () and then typing ) will just be the same as moving forward one character C-f.
  • Do not skip over whitespace when operating on pairs. Combined with the above point, this means that a new character will be inserted, rather than be skipped over. I find this better, because it prevents the point from jumping forward, plus it allows for more natural editing.
  • The whitespace characters are space (\s), tab (\t), and newline (\n).
  • The rest concern the conditions for transforming quotes into their curly equivalents. I keep this disabled, because curly quotes are distinct characters. It is difficult to search for them. Just note that on GNU/Linux you can type them directly by hitting the "compose" key and then an angled bracket (< or >) followed by a quote mark.
(use-package electric
  (setq electric-pair-inhibit-predicate'electric-pair-conservative-inhibit)
  (setq electric-pair-preserve-balance t)
  (setq electric-pair-pairs
        '((8216 . 8217)
          (8220 . 8221)
          (171 . 187)))
  (setq electric-pair-skip-self 'electric-pair-default-skip-self)
  (setq electric-pair-skip-whitespace nil)
  (setq electric-pair-skip-whitespace-chars
  (setq electric-quote-context-sensitive t)
  (setq electric-quote-paragraph t)
  (setq electric-quote-string nil)
  (setq electric-quote-replace-double t)
  :hook (after-init-hook . (lambda ()
                             (electric-indent-mode 1)
                             (electric-pair-mode -1)
                             (electric-quote-mode -1))))

6.3.10 Parentheses

Configure the mode that highlights matching delimiters or parentheses. I consider this of utmost importance when working with languages such as elisp.

Summary of what these do:

  • Activate the mode.
  • Show the matching delimiter/parenthesis if on screen, else show nothing. It is possible to highlight the expression enclosed by the delimiters, by using either mixed or expression. The latter always highlights the entire balanced expression, while the former will only do so if the matching delimiter is off screen.
  • Highlight parentheses even if the point is in their vicinity. This means the beginning or end of the line, with space in between.
  • Do not highlight a match when the point is on the inside of the parenthesis.
(use-package paren
  (setq show-paren-style 'parenthesis)
  (setq show-paren-when-point-in-periphery t)
  (setq show-paren-when-point-inside-paren nil)
  :hook (after-init-hook . show-paren-mode))

6.3.11 Tabs, indentation, and the TAB key

I believe tabs, in the sense of inserting the tab character, are best suited for indentation. While spaces are superior at precisely aligning text. However, I understand that elisp uses its own approach, which I do not want to interfere with. Also, Emacs tends to perform alignments by mixing tabs with spaces, which can actually lead to misalignments depending on certain variables such as the size of the tab. As such, I am disabling tabs by default.

If there ever is a need to use different settings in other modes, we can customise them via hooks. This is not an issue I have encountered yet and am therefore refraining from solving a problem that does not affect me.

Note that tab-always-indent will first do indentation and then try to complete whatever you have typed in. I control how completion works for that particular function in my in-buffer completions section.

(use-package emacs
  (setq-default tab-always-indent 'complete)
  (setq-default tab-width 4)
  (setq-default indent-tabs-mode nil))

6.4 Registers

Registers are compartments that hold data of various sorts. They offer the means for advanced, highly-efficient workflows, especially when combined with keyboard macros.

Registers are called by a single character, which can be a letter (case-sensitive), number, or symbol. Each character can only contain a single register at a time.

To define a register, you call the appropriate command (see table below) and then specify the character you want to store that data at.

Key chord Command
C-x r n number-to-register
C-x r s copy-to-register (think "save string")
C-x r r copy-rectangle-to-register
C-x r SPC point-to-register
C-x r w window-configuration-to-register
C-x r f frameset-to-register (frames and their windows)
C-x r + increment-register (better used with numbers)
C-x r i insert-register (text, number, rectangle)
C-x r j jump-to-register (to point or window/frameset config)

Notes about some of the above:

  • Using point-to-register allows you to revisit a specific location in a buffer, but also reopen the file visited by that buffer in case the buffer is deleted.
  • Calling number-to-register without an argument will just store the number 0 to the register you specify. Whereas C-u 100 C-x r n will store 100. In practice, you often want to use the latter method.
  • Use increment-register to increment a number by one. Pass a numeric argument to increment by that amount instead. For example, to increment by five do C-u 5 C-x r + and then select the register you want to operate on. This only affects the value stored in the register. It does not also insert it in the buffer.
  • Number registers are particularly useful when you want to increment several numbers through a keyboard macro. You can record the motions you need, run increment-register as noted above followed by the standard insert-register. If, however, you just want to increment a single number through a keyboard macro, then just use the counter provided by that facility (refer to my video on easier kmacro counter from 2019-10-14).
  • The registers that store text as a string or a rectangle rely on the active region to capture the data. They are also great for keyboard macros where you cannot rely on yanking from the head of the kill ring (because, say, you are killing other things which push that value further down). Note though that increment-register has a different behaviour when applied to them, where it will append to the register instead (with an option to override the previous value).
  • While on the topic of appending to registers, I define several functions that change how the accumulation of text is supposed to happen. They introduce a space or line separator between the entries you {ap,pre}-pend to the register.
  • In my testing, I could never make window-configuration-to-register persist between sessions (see section on Emacs server and desktop). Whereas all other registers retain their values. So use this command to store window configurations that are otherwise transient in nature. For more permanent setups, rely on frameset-to-register which will produce a new frame (or more if you had) with all the windows in place. Remember that you can always go back to your previous window configuration without using registers, such as with the built-in winner-mode (see section on window history and directional motions).
  • While on the topics of storing registers across sessions, I find that sometimes I will collect too many registers that I do not really need any more. So prot/clear-registers just resets the list.
(use-package register
  :commands (prot/clear-registers
  ;;;;;; Use this if you need it
  ;;;; Define f5 as an alias for C-x r
  ;; (global-set-key (kbd "<f5>") (lookup-key global-map (kbd "C-x r")))

  (defun prot/clear-registers ()
    "Remove all saved registers."
    (setq register-alist nil))

  (defun prot/append-register-space-separator (start end)
    "Append region to register with space in between entries.

A prompt will ask for the register to operate on."
    (interactive "r")
    (let* ((prompt (register-read-with-preview
                    "Append to register with space separator: ")))
      (setq register-separator ?+)
      (set-register register-separator " ")
      (append-to-register prompt start end nil)))

  (defun prot/append-register-line-separator (start end)
    "Append region to register with an empty line in between entries.

A prompt will ask for the register to operate on."
    (interactive "r")
    (let* ((prompt (register-read-with-preview
                    "Append to register with line separator: ")))
      (setq register-separator ?+)
      (set-register register-separator "\n\n")
      (append-to-register prompt start end nil)))

  (defun prot/prepend-register-space-separator (start end)
    "Prepend region to register with space in between entries.

A prompt will ask for the register to operate on."
    (interactive "r")
    (let* ((prompt (register-read-with-preview
                    "Prepend to register with space separator: ")))
      (setq register-separator ?+)
      (set-register register-separator " ")
      (prepend-to-register prompt start end nil)))

  (defun prot/prepend-register-line-separator (start end)
    "Prepend region to register with an empty line in between entries.

A prompt will ask for the register to operate on."
    (interactive "r")
    (let* ((prompt (register-read-with-preview
                    "Prepend to register with line separator: ")))
      (setq register-separator ?+)
      (set-register register-separator "\n\n")
      (prepend-to-register prompt start end nil))))

6.5 Visual bookmarks and extras (bm.el)

This is a deceptively simple-yet-effective package for marking a point in a buffer for future review. I learnt about it from Manuel Uberti's blog entry on the matter (2020-03-19). As Manuel puts it:

I find bm.el really useful when studying source code from others, or when I want to quickly set jumping points in a log file cluttered with stacktraces. It’s quicker then moving around with Isearch or helm-occur, and unlike avy it’s not limited to what is currently visible on the screen.

I use visual bookmarks in much the same way.

(use-package bm
  (setq bm-annotate-on-create nil)
  (setq bm-buffer-persistence t)
  (setq bm-cycle-all-buffers t)
  (setq bm-goto-position nil)
  (setq bm-highlight-style 'bm-highlight-line-and-fringe)
  (setq bm-in-lifo-order t)
  (setq bm-recenter t)
  (setq bm-repository-file "~/.emacs.d/bm-bookmarks")
  (setq bm-repository-size 100)
  (setq bm-show-annotations t)
  (setq bm-wrap-immediately t)
  (setq bm-wrap-search t)
  :bind (("<C-f8>" . bm-next)
         ("<C-S-f8>" . bm-previous)
         ("<s-f8>" . bm-toggle-buffer-persistence)
         ("<f8>" . bm-toggle)))

6.6 Outline minor mode

NOTE 2020-05-26: This section will need to be expanded to complement my configurations for Imenu.

For the defhydra, refer to my fundamentals for hydras.

(use-package outline
  :diminish outline-minor-mode
  (use-package bicycle :ensure)

  (defhydra prot/hydra-outline (:idle nil :color pink
                                      :hint nil :post (deactivate-mark))
   ^Motions^                            ^Show^
_n_: Next heading                    _a_: All
_p_: Prev heading                    _e_: Entry
_f_: Fore same level               _TAB_: Cycle
_b_: Back same level   _<S-iso-lefttab>_: Cycle global
    ("n" outline-next-visible-heading)
    ("p" outline-previous-visible-heading)
    ("f" outline-forward-same-level)
    ("b" outline-backward-same-level)
    ("a" outline-show-all)
    ("e" outline-show-entry)
    ("TAB" bicycle-cycle)
    ("<S-iso-lefttab>" bicycle-cycle-global)
    ("q" nil "cancel" :color blue))
  :bind (:map outline-minor-mode-map
              ("C-c h o" . prot/hydra-outline/body)))

(use-package outline-minor-faces
  :after outline
  (outline-minor-mode-hook . outline-minor-faces-add-font-lock-keywords))

6.7 Custom movements and motions

I generally rely on the default keys to move around (plus my Super-KEY additions to economise on some repetitive tasks). There are, however, some motions that are rather cumbersome or too specialised. While there are some commands that are not available at all. For those cases, I use custom functions.

6.7.1 Go to actionable beginning or end of buffer

This package by Damien Cassou offers the ability to move to the first or last actionable point in a buffer rather than the absolute maximum or minimum point. It does so by wrapping M-< and M-> around a "do what I mean" behaviour where the initial command will take you to the actionable part, while another call will go to the absolute position. Nice and simple!

Check the package upstream for information on the supported modes and on how to contribute your own extensions (I provided support for rg.el, which I define elsewhere in this document).

Here the dolist combined with the :diminish keyword are meant to remove all lighters that this package produces: one for every minor mode it provides. I got this from issue 43 on the beginend repo.

(use-package beginend
  :diminish beginend-global-mode
  (dolist (mode beginend-modes) (diminish (cdr mode)))
  (beginend-global-mode 1))

6.7.2 Mark by semantic unit (expand-region)

The expand-region library offers the means to mark syntactic elements by semantic unit: word, symbol, string, balanced expression… The expansion is incremental, from the smallest to the largest match, with each increment performed by = following the initial invocation of the command (decrement with - or reset with 0) or by just repeatedly hitting C-=.

There are lost of specialised functions bundled up together with this package. Rather than bind each one to a key, I prefer to keep things simple: er/expand-region only ever needs a couple extra calls to match compound units. The few exceptions are for commands I use all the time.

Note though that this library is complementary to existing commands (which can be ignored or altogether superseded). Here is a quick reference:

Function Key binding
mark-paragraph M-h
mark-sexp C-M-SPC
mark-whole-buffer C-x h
mark-word M-@
(use-package expand-region
  :pin gnu                              ; Prefer ELPA version
  (setq expand-region-smart-cursor t)
  :bind (("C-=" . er/expand-region)
         ("C-M-=" . er/mark-outside-pairs)
         ("C-+" . er/mark-symbol)))

6.7.3 Collection of unpackaged commands or tweaks

These are small functions that facilitate my day-to-day work with Emacs. They are written by me, unless otherwise noted.

Backward kill symbolic expression
I just bind the built-in backward-kill-sexp to C-M-backspace. This makes it consistent with kill-whole-line (C-S-backspace).
Change case "do what I mean"
Emacs provides *-dwim versions of its commands for changing the casing of a word or a region. With an active region, they operate on it, else they operate on the word. I bind them in the keys where you would find the originals (which, in turn, render the region-specific variants redundant).
Copy the current line
Just place the entire line in the kill ring. Do not be smart about it.
Count words in buffer
By default, the binding M-= will count the lines, words, characters in the region. I never use that. What I do use is the equivalent for the whole buffer.
Insert pairs
These are several functions that allow for manual insertion of various common delimiters I use. The idea is to offer the means of quickly wrapping a word/symbol/region in those characters without having to always run a minor mode that "intelligently" decides when to insert pairs. The basic pattern for these keybindings is derived from the default M-( which runs insert-parentheses: the basis of all these commands. Note also that M-` shadows the default keys for tmm-menubar, which is a command I am 100% sure I will never use.
Faster multiline move
Move up or down by 15 lines at a time. This is the equivalent of C-u 15 C-n or C-u 15 C-p.
Kill this buffer
A faster way of killing the focused buffer, without the need for confirming which buffer to kill. Unsaved files will still ask for confirmation though, which is exactly as it should be. Bound to s-k.
Kill to the begging of line
Just remove everything from the point till the beginning of the line. Do not try anything fancy like adjusting indentation or removing the line break.
New line above or below
Create a new line above or below the current one and place the point at its beginning. Does not matter where the point is on the current line. Does not try to account for any indentation rules.
Rename current buffer
Change the file name being visited by the current buffer. This updates both the name of the buffer and the file system path. I extracted this command from the Crux package.
Shrink whitespace
By default, M-SPC will reduce multiple spaces into a single one, while M-\ will remove all space at once. Then there is C-x C-o which is the inverse of C-o. For the first two of the aforementioned, I use cycle-spacing: a single invocation will leave one space; a second removes all space; while a third consecutive call restores the original state. Also, since I have no use for the default function bound to M-o I make it an alias for C-x C-o.
Transpose characters
Tweak the way the original command works, so that it always transposes the two characters before point. This will no longer have the effect of moving the character forward following repeated invocations of the command. Just a toggle to quickly fix the last typo.
Transpose or swap lines
Behave as the original transpose-lines, but also be aware of the active region in which case swap the line at point (region end) with the line at mark (region beginning).
Transpose or swap paragraphs
A wrapper around the original transpose-paragraphs that fulfils my expectations: if region is active, swap paragraph at mark (region beginning) with one at point (region end), otherwise transpose current paragraph with the one after it (drag paragraph downward/forward).
Transpose or swap sentences
Same principle as above. When region is active it exchanges the sentences at either end of it, otherwise is drags the sentence forward. The added effect here is that if point is at the end of the paragraph, where no further 'dragging' can occur, it transposes the sentence at point with the one before it.
Transpose or swap words
Same as with the other "swap" commands. In addition, this alters transpose-words while at the end of line where it will only transpose the last two words before the point. It thus avoids transposing words across lines or paragraphs.
Unfill text DWIM
With M-q we can break a long line into a block that ends each line at the fill-column (I configure Emacs to do this automatically, where appropriate, but often do it manually). With M-^ we can join the line at point with the one above. But there seems to be no way of doing this for a block of text. So I assign the following function to M-Q (must also hold down Shift).
Yank over current line or region
Replace the contents of the line at point (or the active region) with the most recent item from the kill ring.
(use-package emacs
  (defun prot/copy-line ()
    "Copies the entirety of the current line."
    (copy-region-as-kill (point-at-bol) (point-at-eol))
    (message "Current line copied"))

  (defun prot/insert-double-quotes (&optional arg)
    "Insert a pair of double quotes or wrap ARG with them."
    (interactive "P")
    (insert-pair arg ?\" ?\"))

  (defun prot/insert-double-smart-quotes (&optional arg)
    "Insert a pair of double smart quotes or wrap ARG with them."
    (interactive "P")
    (insert-pair arg ?\“ ?\”))

  (defun prot/insert-single-smart-quotes (&optional arg)
    "Insert a pair of single smart quotes or wrap ARG with them."
    (interactive "P")
    (insert-pair arg ?\‘ ?\’))

  (defun prot/insert-elisp-quotes (&optional arg)
    "Insert a pair of elisp symbol quotes or wrap ARG with them."
    (interactive "P")
    (insert-pair arg ?\` ?\'))

  (defun prot/multi-line-next ()
    "Moves point 15 lines down."
    (forward-line 15))

  (defun prot/multi-line-prev ()
    "Moves point 15 lines up."
    (forward-line -15))

  (defun prot/kill-line-backward ()
    "Kill from point to the beginning of the line."
    (kill-line 0))

  (defun prot/new-line-below ()
    "Create a new line below the current one.  Move the point to
the absolute beginning.  Also see `prot/new-line-above'."

  (defun prot/new-line-above ()
    "Create a new line above the current one.  Move the point to
the absolute beginning.  Also see `prot/new-line-below'."
    (forward-line -1))

  (defun contrib/rename-file-and-buffer ()
    "Rename current buffer and if the buffer is visiting a file, rename it too."
    (let ((filename (buffer-file-name)))
      (if (not (and filename (file-exists-p filename)))
          (rename-buffer (read-from-minibuffer "New name: " (buffer-name)))
        (let* ((new-name (read-from-minibuffer "New name: " filename))
               (containing-dir (file-name-directory new-name)))
          (make-directory containing-dir t)
           ((vc-backend filename) (vc-rename-file filename new-name))
            (rename-file filename new-name t)
            (set-visited-file-name new-name t t)))))))

  (defun prot/transpose-chars ()
    "Always transposes the two characters before point.  There is
no 'dragging' the character forward.  This is the behaviour of
`transpose-chars' when point is at end-of-line."
    (transpose-chars -1)

  (defun prot/transpose-or-swap-lines (arg)
    "If region is active, swap the line at mark (region
beginning) with the one at point (region end).  This leverages a
facet of the built-in `transpose-lines'.  Otherwise transpose the
current line with the one before it ('drag' line downward)."
    (interactive "p")
    (if (use-region-p)
        (transpose-lines 0)
      (transpose-lines arg)))

  (defun prot/transpose-or-swap-paragraphs (arg)
    "If region is active, swap the paragraph at mark (region
beginning) with the one at point (region end).  This leverages a
facet of the built-in `transpose-paragraphs'.  Otherwise
transpose the current paragraph with the one after it ('drag'
paragraph downward)."
    (interactive "p")
    (if (use-region-p)
        (transpose-paragraphs 0)
      (transpose-paragraphs arg)))

  (defun prot/transpose-or-swap-sentences (arg)
    "If region is active, swap the sentence at mark (region
beginning) with the one at point (region end).  This leverages a
facet of the built-in `transpose-sentences'.  Otherwise transpose
the sentence before point with the one after it ('drag' sentence
forward/downward).  Also `fill-paragraph' afterwards.

Note that, by default, sentences are demarcated by two spaces."
    (interactive "p")
    (if (use-region-p)
        (transpose-sentences 0)
      (transpose-sentences arg))

  (defun prot/transpose-or-swap-words (arg)
    "If region is active, swap the word at mark (region
beginning) with the one at point (region end).

Otherwise, and while inside a sentence, this behaves as the
built-in `transpose-words', dragging forward the word behind the
point.  The difference lies in its behaviour at the end of a
line, where it will always transpose the word at point with the
one behind it (effectively the last two words).

This addresses two patterns of behaviour I dislike in the
original command:

1. When a line follows, `M-t' will transpose the last word of the
line at point with the first word of the line below.

2. While at the end of the line, `M-t' will not transpose the
last two words, but will instead move point one word backward.
To actually transpose the last two words, you need to invoke the
command twice."
    (interactive "p")
    (if (use-region-p)
        (transpose-words 0)
      (if (eq (point) (point-at-eol))
            (backward-word 1)
            (transpose-words 1)
            (forward-char 1))
        (transpose-words arg))))

  (defun prot/unfill-region-or-paragraph (&optional region)
    "Join all lines in a region, if active, while respecting any
empty lines (so multiple paragraphs are not joined, just
unfilled).  If no region is active, operate on the paragraph.
The idea is to produce the opposite effect of both
`fill-paragraph' and `fill-region'."
    (let ((fill-column most-positive-fixnum))
      (if (use-region-p)
          (fill-region (region-beginning) (region-end))
        (fill-paragraph nil region))))

  (defun prot/yank-replace-line-or-region ()
    "Replace the line at point with the contents of the last
stretch of killed text.  If the region is active, operate over it
instead.  This command can then be followed by the standard
`yank-pop' (default is bound to M-y)."
    (if (use-region-p)
          (delete-region (region-beginning) (region-end))
      (delete-region (point-at-bol) (point-at-eol))

  :bind (("<C-M-backspace>" . backward-kill-sexp)
         ("M-c" . capitalize-dwim)
         ("M-l" . downcase-dwim)        ; "lower" case
         ("M-u" . upcase-dwim)
         ("<C-f2>" . contrib/rename-file-and-buffer)
         ("C-S-w" . prot/copy-line)
         ("M-=" . count-words)
         ("M-\"" . prot/insert-double-quotes)
         ("C-M-\"" . prot/insert-double-smart-quotes)
         ("C-M-'" . prot/insert-single-smart-quotes)
         ("M-`" . prot/insert-elisp-quotes)
         ("s-k" . kill-this-buffer)
         ("M-k" . prot/kill-line-backward)
         ("C-S-n" . prot/multi-line-next)
         ("C-S-p" . prot/multi-line-prev)
         ("<C-return>" . prot/new-line-below)
         ("<C-S-return>" . prot/new-line-above)
         ("M-SPC" . cycle-spacing)
         ("M-o" . delete-blank-lines)
         ("C-t" . prot/transpose-chars)
         ("C-x C-t" . prot/transpose-or-swap-lines)
         ("C-S-t" . prot/transpose-or-swap-paragraphs)
         ("C-x M-t" . prot/transpose-or-swap-sentences)
         ("M-t" . prot/transpose-or-swap-words)
         ("M-Q" . prot/unfill-region-or-paragraph)
         ("C-S-y" . prot/yank-replace-line-or-region)))

6.7.4 Go to last change

I could not find any built-in method of reliably moving back to the last change. Using the mark ring is always an option, but does not fill the exact same niche.

The C-z binding is disabled in the opening sections of this document. It minimises the Emacs GUI. A complete waste of an extremely valuable key binding!

(use-package goto-last-change
  :bind ("C-z" . goto-last-change))

6.8 Cursor and mouse settings

6.8.1 Cursor appearance and tweaks

My cursor for the current window is a box character that blinks. Other windows use a vertical bar that is 2 pixels wide.

The default blink settings are close to my expectations, though I do apply some small tweaks to the interval between blinks and the delay for the initial blinking. Where I differ substantially from the defaults is the number of blinks before switching to a non-blinking state. The original value of blink-cursor-blinks is just 10 blinks, which can be fairly short in a number of scenaria.

(use-package emacs
  (setq-default cursor-type 'box)
  (setq-default cursor-in-non-selected-windows '(bar . 2))
  (setq-default blink-cursor-blinks 50)
  (setq-default blink-cursor-interval 0.75)
  (setq-default blink-cursor-delay 0.2)
  :hook (after-init-hook . blink-cursor-mode))

6.8.2 Beacon mode (highlight cursor position)

This package will light a beam where the cursor is. A nice way of quickly discovering your current position in the current window.

There is a minor mode with several customisation options, in case you want to have this functionality to be readily available. I generally do not need it to always be on, so I just bind a key for calling the command manually and only tweak the visuals of the beam effect.

For the way the beacon-color is determined through a custom hook, refer to the section on the Modus themes (my very accessible themes).

(use-package beacon
  (setq beacon-blink-delay 0.1)
  (setq beacon-blink-duration 0.25)
  (setq beacon-size 25)

  (defun prot/beacon-color ()
    "Set the value of `beacon-color' depending on theme.

This is meant to be called from `prot/modus-themes-toggle-hook'."
    (cond ((eq (car custom-enabled-themes) 'modus-operandi)
           (setq beacon-color "#b60000"))
          ((eq (car custom-enabled-themes) 'modus-vivendi)
           (setq beacon-color "#fb6859"))))

  (beacon-mode -1)
  :bind ("<s-escape>" . beacon-blink)
  :hook (prot/modus-themes-toggle-hook . prot/beacon-color))

6.8.3 Mouse wheel behaviour

The value of mouse-wheel-scroll-amount means the following:

  • By default scroll by one line.
  • Hold down Shift to do so by five lines.
  • Hold down Meta to scroll half a screen.
  • Hold down Control to adjust the size of the text. This is added in Emacs 27.

By enabling mouse-drag-copy-region we automatically place the mouse selection to the kill ring. This is the same behaviour as terminal emulators that place the selection to the clipboard (or the primary selection).

The other options in short:

  • Hide mouse pointer while typing.
  • Enable mouse scroll.
  • Faster wheel movement means faster scroll.
  • Scroll window under mouse pointer regardless of whether it is the current one or not.
(use-package mouse
  ;; In Emacs 27, use Control + mouse wheel to scale text.
  (setq mouse-wheel-scroll-amount
          ((shift) . 5)
          ((meta) . 0.5)
          ((control) . text-scale)))
  (setq mouse-drag-copy-region t)
  (setq make-pointer-invisible t)
  (setq mouse-wheel-progressive-speed t)
  (setq mouse-wheel-follow-mouse t)
  :hook (after-init-hook . mouse-wheel-mode))

6.8.4 Scrolling behaviour

Page scrolling should keep the point at the same visual position, rather than force it to the top or bottom of the viewport. This eliminates the friction of guessing where the point has warped to.

As for per-line scrolling, I dislike the default behaviour of visually re-centring the point. With the following, it will stay at the top/bottom of the screen while moving in that direction (use C-l to reposition it). This does not result in more manual interventions to recenter text, because of the above.

(use-package emacs
  (setq scroll-preserve-screen-position t)
  (setq scroll-conservatively 1)        ; affects `scroll-step'
  (setq scroll-margin 0))

6.8.5 Delete selection

This is a very helpful mode. It kills the marked region when inserting directly to it. It also has checks to ensure that yanking over a selected region will not insert itself when mouse-drag-copy-region is in effect (see the section on the mouse wheel behaviour).

(use-package delsel
  :hook (after-init-hook . delete-selection-mode))

Pro tip: On Emacs 27.1 you can create a rectangular region by holding down Ctrl and Meta while dragging the mouse with the left click pressed.

6.8.6 Tool tips

These settings control how tool tips are to be handled when hovering the mouse over an actionable item:

  • I just want to make sure that the GTK theme is not used for those: I prefer the generic display which follows my current theme's styles.
  • The delay is slightly reduced for the initial pop-up, while it has been increased for immediate pop-ups thereafter.
(use-package tooltip
  (setq tooltip-delay 0.5)
  (setq tooltip-short-delay 0.5)
  (setq x-gtk-use-system-tooltips nil)
  :hook (after-init-hook . tooltip-mode))

6.9 Conveniences and minor extras

6.9.1 Auto revert mode

This mode ensures that the buffer is updated whenever the file changes. A change can happen externally or by some other tool inside of Emacs (e.g. kill a Magit diff).

(use-package autorevert
  (setq auto-revert-verbose t)
  :hook (after-init-hook . global-auto-revert-mode))

6.9.2 Preserve contents of system clipboard

Say you copied a link from your web browser, then switched to Emacs to paste it somewhere. Before you do that, you notice something you want to kill. Doing that will place the last kill to the clipboard, thus overriding the thing you copied earlier. We can have a kill ring solution to this with the following:

(use-package emacs
  (setq save-interprogram-paste-before-kill t))

Now the contents of the clipboard are stored in the kill ring and can be retrieved from there (e.g. with M-y).

6.9.3 Generic feedback

The common thread of these options is the feedback they provide us with or simplify common tasks so that their feedback does not cause friction:

  • Show the current buffer's name as the frame's title. This only affects window managers that have window decorations. If you do not know what that means, then you are most likely using an environment where frame titles are already visible.
  • Faster feedback for key chords (keys appear in the echo area).
  • Allow inputting Greek while preserving Emacs keys. Toggle with C-\.
  • Ignore visual or audible bells. Emacs has more appropriate ways of providing error/warning messages or hints that something is not producing the desired results (e.g. a failed isearch will return no results, while the failed match will be styled accordingly in the echo area). By the way, use C-h e to bring up the log with the echo area's messages.
  • Answer with just the initials when dealing with "yes/no" questions.
  • Enable actions for narrowing the buffer, region {up,down}casing (all caps or no caps), dired single-buffer navigation (bound to a). Disable overwrite-mode.
(use-package emacs
  (setq frame-title-format '("%b"))
  (setq echo-keystrokes 0.25)
  (setq default-input-method "greek")
  (setq ring-bell-function 'ignore)

  (defalias 'yes-or-no-p 'y-or-n-p)
  (put 'narrow-to-region 'disabled nil)
  (put 'upcase-region 'disabled nil)
  (put 'downcase-region 'disabled nil)
  (put 'dired-find-alternate-file 'disabled nil)
  (put 'overwrite-mode 'disabled t))

6.9.4 Newline characters for file ending

For some major modes, such as diff-mode, a final newline is of paramount importance. Without it you cannot, for instance, apply a patch cleanly. As such, the mode-require-final-newline will add a newline character when visiting or saving a buffer of relevance.

(use-package emacs
  (setq mode-require-final-newline 'visit-save))

6.9.5 Altered zap and easier repeats

I seldom use the functionality related to this section, but when I do I prefer it to work the way I expect. zap-up-to-char will delete everything from point up to the character you provide it with. Think of how you may want to delete a file name but keep its file type extension.

The repeat command is bound by default to C-x z. I make it so that subsequent repetitions require only hitting another z. In practice though, you should not bother with this. Let keyboard macros handle that task.

Pro tip: to make a keyboard macro out of your most recent commands, use C-x C-k l which calls kmacro-edit-lossage. The list is editable, so remove anything that is not required and then save what is left. The result is stored as the latest keyboard macro (and you also have the power to cycle through kmacros, store them in specific keys, etc.).

Moving on to the mark, practically every Emacs motion that operates on a portion of text will set the mark automatically. You can also do it manually with C-SPC (hit it twice if you do not wish to activate the region). It is then possible to cycle through the marks in reverse order by passing a prefix argument C-u C-SPC. With the evaluation of set-mark-command-repeat-pop as t we can continue cycling by repeated presses of C-SPC. Again though, this is not the type of functionality I rely on: for more deliberate actions of this sort, consider Emacs' notion of "registers".

(use-package emacs
  (setq repeat-on-final-keystroke t)
  (setq set-mark-command-repeat-pop t)
  :bind ("M-z" . zap-up-to-char))

6.9.6 Package lists

With this I just want to enable line highlighting when browsing the list of packages. I generally use hl-line-mode on all interfaces where the current line is more important than the exact column of the point.

(use-package package
  :commands (list-packages
  :hook (package-menu-mode-hook . hl-line-mode))

7 History and state

This section contains configurations for packages that are dedicated to the task of recording the state of various Emacs tools, such as the history of the minibuffer or the list of recently-visited files.

In practice, these are some of the most useful configurations one can make, as lots of functions depend on them. For example, a record of the minibuffer's history of inputs allows the completion framework to guess the most likely course of action. Typing M-x g gives me gnus as the first possible option, which is exactly what I want.

7.1 Emacs server and desktop

The following uses the first running process of Emacs as the one others may connect to. This means that calling emacsclient (with or without --create-frame), will share the same buffer list and data as the original running process, aka "the server". The server persists for as long as there is an Emacs frame attached to it.

(use-package server
  :hook (after-init-hook . server-start))

With some exceptions aside, I only ever use Emacs in a single frame. What I find more useful is the ability to save the state I was in: the name the of buffers, the cursor's position in each of them, the recent file list, the minibuffer history, my stored registers.

The state of the available buffers and the values of each register are called the "desktop" (for the other items see the following sections on recording various types of history).

Preserving the "desktop" saves me from any possible crash or when I need to close Emacs and re-launch it later (my hardware is limited, so I do not keep it running while I am away).

Overview of my settings:

  • Enable the mode that saves the "desktop", instructing it to load a small number of buffers at launch (desktop-restore-eager). The remainder of the buffer list will be loaded lazily.
  • Now we must tell it where to store the files it generates and how often it should save. Concerning the latter, the default is to store the state every time it changes. I find that a bit too much, so I set a timeout of five minutes of idleness.
  • Note the desktop-load-locked-desktop. By default, Emacs locks the desktop file while it runs. The lock is removed upon exiting. This is a safety mechanism. There are two cases where the lock can create issues:
    • Emacs has crashed, meaning that it exited abruptly and was not able to unlock the desktop. Upon re-launch Emacs will prompt you whether to load the locked file. You normally want to answer affirmatively.
    • Emacs runs in daemon mode, where it does not ask questions upon loading. In this case the lock is ignored.
    • Because I am only affected by the former, I choose to disable the prompt and just load the thing directly. Otherwise, I would set it to nil.
  • Do not restore frame configurations. If I need to store one of those, I use registers, specifically C-x r f (see section on Registers).
  • Ask what to do in case the session has a newer file that the one it initially started out with (e.g. when a new frame runs in parallel to the older one).
(use-package desktop
  (setq desktop-auto-save-timeout 300)
  (setq desktop-dirname "~/.emacs.d/")
  (setq desktop-base-file-name "desktop")
  (setq desktop-files-not-to-save nil)
  (setq desktop-globals-to-clear nil)
  (setq desktop-load-locked-desktop t)
  (setq desktop-missing-file-warning t)
  (setq desktop-restore-eager 3)
  (setq desktop-restore-frames nil)
  (setq desktop-save 'ask-if-new)
  (desktop-save-mode 1))

7.2 Record various types of history

7.2.1 Recentf (recent files and directories)

This is a built-in minor mode that keeps track of the files you have opened, allowing you revisit them faster. Its true power consists in the fact that its data, maintained in recentf-list, is a simple variable. This means that we can access it through any relevant piece of Elisp functionality.

To that end, the functions I define herein are meant to either control the contents of the list or allow me to access them through my completion framework or a dedicated file listing (see, in particular, Minibuffer essentials and Icomplete and configurations for Dired).

Note that there exists a built-in recentf-open-files function for accessing the recent files through a bespoke buffer. I find that I have no use for it.

(use-package recentf
  (setq recentf-save-file "~/.emacs.d/recentf")
  (setq recentf-max-saved-items 200)
  (setq recentf-exclude '(".gz" ".xz" ".zip" "/elpa/" "/ssh:" "/sudo:"))

  (defun prot/recentf-keep-predicate (file)
    "Additional conditions for saving in `recentf-list'.
Add this function to `recentf-keep'.

     ((file-directory-p file) (file-readable-p file))))
  (add-to-list 'recentf-keep 'prot/recentf-keep-default-predicate)

  (defun prot/recentf ()
    "Select item from `recentf-list' using completion.
The user's $HOME directory is abbreviated as a tilde."
    (icomplete-vertical-do ()
      (let ((files (mapcar 'abbreviate-file-name recentf-list)))
         (completing-read "Open recentf entry: " files nil t)))))

  (defun prot/recentf-dirs (&optional arg)
    "Select directory from `recentf-list' using completion.
With \\[universal-argument] present the list in a `dired' buffer.
This buffer is meant to be reused by subsequent invocations of
this command (otherwise you need to remove the `when' expression.

Without \\[universal-argument], the user's $HOME directory is
abbreviated as a tilde.  In the Dired buffer paths are absolute."
    (interactive "P")
    (let* ((list (mapcar 'abbreviate-file-name recentf-list))
           (dirs (delete-dups
                  (mapcar (lambda (file)
                            (if (file-directory-p file)
                                (directory-file-name file)
                              (substring (file-name-directory file) 0 -1)))
           (buf "*Recentf Dired*")
           (default-directory "~"))
      (when (get-buffer buf)
        (kill-buffer buf))
      (if arg
          (dired (cons (generate-new-buffer-name buf) dirs))
        (icomplete-vertical-do ()
           (completing-read "Recent dirs: " dirs nil t))))))

  :hook (after-init-hook . recentf-mode)
  :bind (("s-r" . prot/recentf)
         ("C-x C-r" . prot/recentf-dirs)))

7.2.2 Minibuffer history

Keeps a record of actions involving the minibuffer. This is of paramount importance to a fast and efficient workflow involving any completion framework that leverages the built-in mechanisms.

Emacs will remember your input and choices and will surface the desired results towards the top as the most likely candidates. Make sure to also check my configurations for the minibuffer and icomplete.

I set the length to a fairly high number, while I make sure that duplicate entries remain in tact. The assumption is that duplicate entries increase the likelihood of returning the candidate I am searching or.

(use-package savehist
  (setq savehist-file "~/.emacs.d/savehist")
  (setq history-length 30000)
  (setq history-delete-duplicates nil)
  (setq savehist-save-minibuffer-history t)
  (savehist-mode 1))

7.2.3 Record cursor position

Just remember where the point is in any given file. This can often be a subtle reminder of what you were doing the last time you visited that file, allowing you to pick up from there.

(use-package saveplace
  (setq save-place-file "~/.emacs.d/saveplace")
  (setq save-place-forget-unreadable-files t)
  (save-place-mode 1))

7.2.4 Backups

And here are some settings pertaining to backups. I rarely need those, but I prefer to be safe in the knowledge that if something goes awry there is something to fall back to.

(use-package emacs
  (setq backup-directory-alist
        '(("." . "~/.emacs.d/backup/")))
  (setq backup-by-copying t)
  (setq version-control t)
  (setq delete-old-versions t)
  (setq kept-new-versions 6)
  (setq kept-old-versions 2)
  (setq create-lockfiles nil))

8 Frequently Asked Questions about this document

There are some persistent questions that pop up in my email exchange, so I thought I would cover them all in this section.

8.1 How do you learn Emacs?

There is no one-size-fits-all approach to learning. What one finds satisfactory, another may consider insufficient. In my opinion, the best way to learn Emacs is to start small, be patient, and brace yourself for a lot of reading and experimentation.

The best skill you can master, the one that will always help you in your Emacs journey, is the built-in help system. Learn to ask Emacs about things you do not know about. This section documents the essentials of Emacs' introspectable nature.

Know that C-h is the universal key for help commands (broadly understood). It works both as a prefix and as a suffix. Some common help commands:

  • C-h f allows you to search for documentation about functions.
  • C-h v is the same for variables.
  • C-h o is a wrapper of the above two, so you are searching for functions or variables. The proper name for any of these items is called a "symbol" (think of name-of-thing as symbolising a definition in the source code).
  • C-h k will prompt you for a key binding. Once you type it in you will get help about what command corresponds to that key (if any). Note that this depends on the major mode you are in. For example C-c C-c does something different in Org and Eshell buffers. Try C-h k C-c C-c to find about the different functions these will invoke in their respective major mode.

In the above scenaria we see how C-h is used as a prefix. You are starting a key sequence with it. So here are some cases of using it as a suffix:

  • For every incomplete key sequence if you follow it up with C-h you will get a help buffer about all possible key combination that complete that sequence. For example, if you type C-c C-h inside of an Org buffer you will get all possible commands for Org mode and for all other minor modes you have active.
  • The C-h suffix can be appended to longer key sequences. Indeed the length is irrelevant. Suppose you want to learn more about some of the advanced features of registers. C-x r is the common prefix for those commands, so you just do C-x r C-h and you get a buffer with more information.
  • And, as you may imagine, you can even append the C-h suffix to the C-h prefix. This is a fancy way of saying that C-h C-h will show you help about help commands themselves. But because this is a special case, it comes with some extras. Try it!

All help buffers include links to other parts of Emacs, from where you can learn ever more information. For example, the help for C-c C-h includes links to the commands that correspond to each key chord. Follow the link to get documentation about that symbol.

More generally, you will find that a symbol is linked to its source. Look carefully at the top of the buffer that display symbol documentation and you will find a link to the source code (library) from where the function/variable (symbol) comes from.

Also know that the source code can be accessed at any time by means of M-x find-library.

Help commands that ask you for a symbol to input can also be aware of the context of the point (the cursor). So if you are over the name of a function and you type C-h f, that function will be the default match. Hitting RET (Enter) will take you to its documentation. This is a great way to study source code, because it will guide you to other libraries or other parts of the same library from where you can understand how things are pieced together. And it also works with the find-library command.

Finally, you should practice C-h m as much as possible. This is the help command for getting information about the major mode you are in and all the minor modes you have active. It will show you some valuable documentation as well as the main key bindings and their commands. Try it whenever you use something you have not mastered yet. For example, do it in a dired buffer to see the main operations you can perform inside of it.

8.2 What is the purpose of "prot/" and "contrib/"?

The prot/ prefix in symbols works as a namespace that captures all my additions. These can be functions or variables. Its utility or purpose is two-fold:

  1. It informs others that this symbol is not part of core Emacs or some other package. Otherwise it can be difficult to understand why something you copied did not work. Say, for example, I have a function that accepts an argument: (prot/function prot/variable). If none of these had the namespace you could be misled into thinking that your Emacs setup already includes those symbols (and then you will get an error message).
  2. It makes it easier for me (and others) to quickly discover what additions I have made, for whatever reason that may. For instance, M-x prot/ will give me matches for all my interactive functions (depending on the completion framework, one can access those with just M-x p/ as in my Minibuffer essentials and Icomplete). This also means that I can do M-s o prot/ (or M-x occur prot/) to produce an Occur buffer with all my symbols (pass a number argument for displaying N lines of context C-u 5 M-s o). From there I can, say, browse them all easily or even edit them using all of Emacs' relevant powers (Occur is made editable with e, but you should be able to find that by using the information in How do you learn Emacs?).

The same principles apply for the contrib/ namespace. Whenever I copy something from another user, I use that namespace to tell others that this is not part of Emacs or any other package (and I always link to the source).

Adding contrib/ also has another longer-term benefit. It tells my future self that some bespoke configuration was needed to solve a particular problem and, maybe, this has since been solved by a good package or even a newer version of Emacs. A case in point is the hook I use for my themes: contrib/after-load-theme-hook. As of 2020-05-03 there is no built-in way of accomplishing that bit of functionality (as far as I know), but there may be one at some point and it might be an improvement overall.

Again, occur or similar tools will filter these out when necessary. Imagine having to do that without any namespaces… You would need to check each symbol one by one to determine its origin.

Last but not least, the convention of separating namespaces with a forward slash is not particularly important (and the linter for Elisp packaging will complain about it, if you ever go down that path). It could be something like prot-- or my- or whatever. What matters more is to keep things consistent and fairly easy to identify.

8.3 Why do you use so many `setq'?

To be clear about the question, these are equivalent:

;; Style A
(setq var-1 value-1)
(setq var-2 value-2)

;; Style B
(setq var-1 value-1
      var-2 value-2)

You will notice that most of my configurations follow style A. The reason is due to convenience. It is easier for me to share code with users who may not be familiar with how elisp works (I was in that group not so long ago and am still learning my way around).

Style B used to give me more problems with copy-pasting, while it did not solve any real issue (besides, I finalise style A using a keyboard macro, so there is no real difference in typing).

Style A consists of balanced expressions that are easier to keep consistent. This is especially true when you have a mixture of values, some boolean types, property lists, association lists… Of course, experience is key to understanding that either style is valid.

8.4 Why do you explicitly set variables the same as default?

You may have realised that many of my configurations will use a setq that declares a value that is the same as its original in the source code. I do this for two reasons:

  1. To raise awareness of its existence. If someone does not like how the defaults work, they know where to look.
  2. I guard against future versions that could be changing the defaults.

Obviously point 2 is not particularly strong for built-in libraries that are already very stable, though I find that, on balance, nothing bad comes out of it.

At any rate, one must always read the NEWS (C-h n for view-emacs-news) whenever they upgrade to a new version of Emacs. Though there is no equivalent mechanism for individual packages… So here we are.

8.5 What is you distro and/or window manager?

Given that I mostly live inside of Emacs, the rest of the environment is no longer important to me (provided it does not impede my usage of Emacs).

Prior to my Emacs days, I used to have a highly customised session centred around BSPWM. Now I can happily live inside of any desktop environment or window manager. What matters is that my Emacs works as intended, which practically means that the environment should not bind any keys system-wide (with the exception of some standard ones like those for accessing TTYs).

I have been on GNU/Linux since the summer of 2016. For the most time I have used Debian and Arch Linux. As of 2020-05-03, I am on Arch. My criterion for picking a distro is that it is community-driven and has a strong following that ensures its longer-term continuity and overall stability. This is consistent with my focus on Emacs (and thoroughgoing customisations of the computing setup more generally): I need things to work so that I may not be bothered by too much admin work (and yes, Arch is super-stable if you know what you are doing—and, well, Debian is designed for that).

My Emacs is built from source, using the emacs-27 branch (check the AUR if you are on Arch or derivatives, though always read the PKGBUILD). Normally I would be using the packages provided by my distro's standard repos but am opting for this method instead in order to make sure my Modus themes are compatible with changes upstream.

I do not optimise for portability across different versions or operating systems. I do, nonetheless, provide inline comments when a certain option or configuration is specific to a yet-to-be-released version of Emacs, to the extent that I am aware that such is the case.

8.6 Did you know XYZ package is better than the defaults?

As a rule of thumb, I choose external packages only after I give a fair chance to the defaults. The idea is to take things slowly and learn along the way, while consulting the official manual and relevant documentation (I strongly encourage you to master the information I provide in How do you learn Emacs?).

Without exposure to the built-in tools it is impossible to make an informed decision about what is actually missing and what could be improved further. Additionally, it is difficult to appreciate the underlying rationale that led to a given element of design without actually trying to put yourself in that mindset or workflow.

Put differently, keep an open mind about the alternatives before deciding to reproduce the thing you had before, else you are assuming your prior knowledge to be true in advance of any possible evidence to the contrary (a sign of dogmatism).

The process of learning the internals of Emacs means that I write my own Elisp functions when I feel that a standard tool could be tweaked on the margin of its core utility in order to do what I want (see my answer to What is the purpose of "prot/" and "contrib/"?). External packages that I do use are either a clear upgrade over the defaults or otherwise extend the functionality of what is already available.

You will not find any superficialities herein: no rainbow-coloured mode lines, no icons, no tool bars with bells and spinning wheels, nothing. I respect the fact that they exist, but find that they do not contribute to my productivity.

Though a former Vim user for ~3 years, I decided not to use evil-mode or any kind of Vi emulation (remember the point about keeping an open mind?). I wanted to do things differently in order to ultimately set on the best approach going forward. I have eventually settled on a system that builds on top of the "Emacs way" to key bindings, which I discuss in my hour-long presentation about Switching to Emacs (2019-12-20). I believe that a mnemonics-based set of keys is easier to get started with. It expands organically as you familiarise yourself with multitude of Emacs' functionalities: there is an entire world of applications outside the narrow confines of editing code.

8.7 What is the meaning of the `s-KEY' bindings?

Some sections of this document assign functions to key combinations that involve the "Super" key (also known as the "Windows key" on some commercial keyboards). This is represented as a lower case s.

In most cases, those key bindings are alternative ways of invoking common commands that are by default bound to longer key chords. The original keys will continue to function as intended (for example, C-x o is also s-o). Otherwise they bind my own custom functions.

To find all my keybindings of this sort in the source code version of this document from inside of Emacs, do M-s o (or M-x occur) followed by the pattern "[a-zA-Z<]?-?s-.+?" (please contact me if you know how to improve this).

Note that your choice of desktop environment (or window manager) might render some of these useless. The DE/WM will simply intercept the key chord before it is ever sent to Emacs. For example, GNOME has a hidden key mapping to s-p, which does something with monitors (last time I checked on GNOME 3.30). Such bindings are scattered throughout the config database that is normally accessed with gsettings on the command line or the graphical dconf-editor (not pretty either way).

8.8 What hardware do you use?

I am using a Lenovo Thinkpad X220 that I got in 2018 for ~80 EUR. This is the first computer I ever bought: before that I had a Macbook that was offered to me as part of an office job—but do not ask me about it because at the time I only knew how to copy/paste using right click and that sort of thing, while I never bothered with the hardware side of things once I got better at using the computer (my switch to GNU/Linux was about freedom and consumer sovereignty, i.e. politics, not tech-only considerations).

The laptop is mostly deployed as part of a desktop setup, attached to an external monitor, mouse, and keyboard. The monitor is 1080p and I got it for free from a clearance. As for the mouse and keyboard, I bought those from a toy store for 7 EUR combined. The keyboard's layout is US Qwerty.

For my videos I use the built-in camera and microphone (sorry if the production quality is sub-standard!). Since we are here: the recording software is OBS Studio, while I do no video editing.

8.9 Why keep many completion UIs? (typology of my completion interfaces)

In Completion framework and extras I define lots of completion-related functions and configure their accoutrements to deliver on a consistent experience (some completion-related functions are defined in a few other places as well, always with the appropriate references, though they extend the principles documented herein).

To understand why I am maintaining seemingly overlapping settings for icomplete, icomplete-vertical, and embark you need to have a sense of the full picture of how I conceive of completion workflows as part of my general Emacs usage:

One-off commands

These are the regular means one would use to transition from state A to state B, such as to switch buffers or call an interactive function like M-x eval-buffer. The plan is to get on with the task with as little friction as possible. There is no real need for extended interactions with the minibuffer. Things should happen quickly and seamlessly. Preferably with few or no extra visual effects and as little fanfare as possible.

This explains why I use the standard horizontal layout of Icomplete by default. It does not interfere with my windows because that kind of feedback is not needed for the given interaction.

However, the horizontal view fails when the list consists of naturally long candidates, such as those returned by imenu (see configuration for Imenu (dynamic completion-based buffer navigation)). In those cases we still want a transient interface, but must rely on verticality to make sense of the nuances between the completion candidates.

With icomplete-vertical we can cope with that requirement through a simple toggle that is available on demand. Though we can enable it presciently on a per-function basis, which is what I normally do for a whole range of commands (search for the icomplete-vertical-do macro call, with common cases other than Imenu being to select an entry from the kill-ring or choose a filesystem path among the items in the recentf-list—also see section on recentf).

Involved interactions

These are scenaria where one expects to interact with the list of candidates itself. There likely is a need to have a view of all the matches as well as a means of running additional commands that are specific to the task and contribute towards a more complex workflow.

In this light, the standard "Completions" buffer is essential. Same for the bespoke buffers produced by embark-occur (refer to the section on Embark (actions for completion candidates)). It feels natural to have a sense of all the available matches for those cases where you need to search, say, for some documentation and are probably interested in reading more than one entry (e.g. when you want to configure a new package), as well as potentially perform some other action as part of a longer process of interaction (with or without recursive minibuffers).

In more concrete terms: search for help with C-h f, C-h v, C-h o. The Completions buffer will show you what matches your input. Switch to that buffer (e.g. prot/focus-minibuffer-or-completions), place point over an item you are interested in and run a bespoke command for describe-symbol for the thing at point (this is, for instance, what my prot/describe-symbol-at-point does, and I bind it to convenient keys—again make sure to check Embark for all the actions it offers). The help buffer will pop up, allowing you to, say, switch to it in order to copy something, then go to a file that is being edited in the other window to iterate on it, and then get back to the Completions buffer for more of the same. This is also why recursive minibuffers are valuable, as they allow you to keep the current session active while still accessing things like M-x.

As for the "Completions" buffer itself (and "Embark Occur"), I must stress that this is a buffer. It allows you to, for example, write its contents to a regular file (C-x C-w). And with dedicated key maps such as those provided by embark or the completion-list-mode-map, you can always define context-specific bindings for your bespoke commands (e.g. I use h for help, w for writing to the kill-ring). Plus you can always control the placement of a buffer with display-buffer-alist (see the section on Window rules and basic tweaks).

The following table summarises the above:

contents transient persistent
short icomplete completions / embark-occur (grid)
long icomplete-vertical completions / embark-occur (list)

Finally, I also have commands that connect the completion-specific interface to more general tools, like ibuffer and dired (embark does that as well). So when some involved type of interaction demands extra attention, I can always leverage the full power of the wider Emacs ecosystem.

9 Important configurations that I stopped using

As my understanding of Emacs improves, I critically assess the utility of tools I once used. What is not needed is promptly removed. However, there are some cases where the deprecation of a package is basely solely on subjective criteria: the tools are still good, it just happens that I discovered something I like more.

Furthermore, there may be packages I once demonstrated in a video of mine, where I added my own extensions or whatnot and which people found useful.

Those are all stored in this section.

9.1 DEPRECATED Ivy/Counsel/Swiper plus filtering and scoring

WARNING 2020-02-10: I have switched to a simpler, custom solution that uses the built-in icomplete tool. All deprecated packages contain the special :disabled keyword. For the new system, refer to the previous sections:

Below is the latest state of these settings prior to the switch.

This is a suite of tools that enhance several aspects of the Emacs experience. Basically we have:

  • ivy is the mechanism that handles all selection lists, narrowing them down using a variety of possible builders (regular expressions of flexible matching). It also provides a base interface for any function that needs to receive input based on a list of candidates.
  • counsel provides a superset of functions for navigating the file system, switching buffers, etc. that expand on the basic features supported by Ivy. For instance, switching buffers with Counsel offers a preview of their contents in the window, whereas regular Ivy does not.
  • swiper is a tool for performing searches, powered by Ivy, all while presenting a preview of the results.

9.1.1 DEPRECATED Configurations for Ivy

WARNING 2020-02-10: I have switched to a simpler, custom solution that uses the built-in icomplete tool. All deprecated packages contain the special :disabled keyword. For the new system, refer to the previous sections:

Below is the latest state of these settings prior to the switch.

A few highlights of my configurations in the subsequent code block:

  • ivy-height-alist governs the maximum width of the Ivy window to 1/4 of the viewport. I prefer this over an absolute number as I work on monitors with varying dimensions (though note that ivy-posframe will override it, if enabled).
  • ivy-virtual-buffers populates buffer-switching lists with items from the recentf utility. In practice, a recently killed buffer can still be accessed from {ivy,counsel}-switch-buffer as if the kill had never occurred.
  • ivy-re-builders-list allows us to specify the algorithm for matching candidates. Unless declared otherwise, I am using regexp matching by default.
  • ivy-initial-inputs-alist adds some initial input to the commands we specify. I define it to prepend a caret sign (^) by default, unless explicitly stated not to include anything.
  • ivy-use-selectable-prompt solves the problem of trying to create a new file system path that shares a common name with an existing item. Press C-p and proceed without further conflicts.
  • The various ivy-set-occur are meant to specify a function for producing an appropriate buffer when running ivy-occur (see table right below).

And here are some 'hidden' key bindings for making the most out of Ivy (find more in the official manual).

Key Function Description
M-o ivy-dispatching-done Show actions for current match.
C-c C-o ivy-occur Place the list in a standalone buffer.
C-M-m ivy-call Run command, keep minibuffer open.
M-i ivy-insert-current Insert match in the prompt.
M-j ivy-yank-word Put word at point in the minibuffer prompt.
S-SPC ivy-restrict-to-matches Restrict list to prompt (and search anew).
C-SPC ivy-restrict-to-matches My alias for the above.

Adding to the table above, you can always use the universal M-n and M-p to cycle through the history of entries.

With those granted, make sure to inspect the entirety of my dotemacs' section on Selection candidates and search methods, as the following package declaration is but a piece of a greater whole.

(use-package ivy
  :disabled                             ; switched to `icomplete'
  (setq ivy-count-format "(%d/%d) ")
  (setq ivy-height-alist '((t lambda (_caller) (/ (window-height) 4))))
  (setq ivy-use-virtual-buffers t)
  (setq ivy-wrap nil)
  (setq ivy-re-builders-alist
        '((counsel-M-x . ivy--regex-fuzzy)
          (ivy-switch-buffer . ivy--regex-fuzzy)
          (ivy-switch-buffer-other-window . ivy--regex-fuzzy)
          (counsel-rg . ivy--regex-or-literal)
          (t . ivy--regex-plus)))
  (setq ivy-display-style 'fancy)
  (setq ivy-use-selectable-prompt t)
  (setq ivy-fixed-height-minibuffer nil)
  (setq ivy-initial-inputs-alist
        '((counsel-M-x . "^")
          (ivy-switch-buffer . "^")
          (ivy-switch-buffer-other-window . "^")
          (counsel-describe-function . "^")
          (counsel-describe-variable . "^")
          (t . "")))

  (ivy-set-occur 'counsel-fzf 'counsel-fzf-occur)
  (ivy-set-occur 'counsel-rg 'counsel-ag-occur)
  (ivy-set-occur 'ivy-switch-buffer 'ivy-switch-buffer-occur)
  (ivy-set-occur 'swiper 'swiper-occur)
  (ivy-set-occur 'swiper-isearch 'swiper-occur)
  (ivy-set-occur 'swiper-multi 'counsel-ag-occur)
  :hook ((after-init-hook . ivy-mode)
         (ivy-occur-mode-hook . hl-line-mode))
  :bind (("<s-up>" . ivy-push-view)
         ("<s-down>" . ivy-switch-view)
         ("C-S-r" . ivy-resume)
         :map ivy-occur-mode-map
         ("f" . forward-char)
         ("b" . backward-char)
         ("n" . ivy-occur-next-line)
         ("p" . ivy-occur-previous-line)
         ("<C-return>" . ivy-occur-press)))

9.1.2 DEPRECATED Prescient (sort, filter results)

WARNING 2020-02-10: I have switched to a simpler, custom solution that uses the built-in icomplete tool. All deprecated packages contain the special :disabled keyword. For the new system, refer to the previous sections:

Below is the latest state of these settings prior to the switch.

This tool provides a filtering and scoring system that can interface with Ivy. It is a replacement for amx.

Filtering concerns the way matches are determined. It is possible to select candidates by applying the search terms in a number of ways, such as a literal interpretation of the character string, a regular expression, a set of ordered wildcards (fuzzy match), or an initialism. The filters can be applied on a per function basis, though I that breaks the way Ivy highlights things.

The scoring system is based on the frequency and recency of commands. This is extremely valuable, as it will always surface to the top the commands you most likely need. Running M-x is now akin to starting a key chord chain (for example, M-x b will give me bongo as my first match, which is exactly what I need). It eliminates the need for increasingly arcane key bindings—use keys only for the most frequently-used commands.

The value of ivy-prescient-sort-commands defines functions that are exempt from all the sorting operations performed by this tool. I think that, at the very least, swiper should always be excluded from any kind of sorting mechanism, because it already orders its results based on the line number. We should not break that expectation. All other commands are just a matter of preference, where I simply find the generic sorting to offer a better first impression (does not really matter, since I will type something anyhow).

(use-package prescient
  :disabled                             ; switched to `icomplete'
  (setq prescient-history-length 200)
  (setq prescient-save-file "~/.emacs.d/prescient-items")
  (setq prescient-filter-method '(literal regexp))
  (prescient-persist-mode 1))

(use-package ivy-prescient
  :disabled                             ; switched to `icomplete'
  :after (prescient ivy)
  (setq ivy-prescient-sort-commands
        '(:not counsel-grep
  (setq ivy-prescient-retain-classic-highlighting t)
  (setq ivy-prescient-enable-filtering nil)
  (setq ivy-prescient-enable-sorting t)
  (ivy-prescient-mode 1))

9.1.3 DEPRECATED Counsel configurations and key bindings

WARNING 2020-02-10: I have switched to a simpler, custom solution that uses the built-in icomplete tool. All deprecated packages contain the special :disabled keyword. For the new system, refer to the previous sections:

Below is the latest state of these settings prior to the switch.

A few things to consider about the settings in this sub-section:

  • With regard to key bindings, notice that Counsel's implementation for switching buffers will preview the currently matched item. This is particularly distracting when running it for the current window. For that case I use the generic Ivy method. I am okay with Counsel's approach when operating on the other window.
  • As for counsel-yank-pop-separator, its value is a series of em dashes with a newline character at either end. This creates a nice separator line when browsing the kill ring (counsel-yank-pop).
  • The function counsel-rg provides an interface to an external program called ripgrep. This is a great alternative to grep. For me the main selling point is its improved speed. The key chord for it is similar to the built-in occur.
  • The key chord for counsel-git-grep is also inspired by occur. This one will perform a search in the current git repository.
  • I have a few custom functions for finding files using a fluid workflow from fzf to ripgrep (and vice versa). Better check my video on Fuzzy search with “Ivy actions” for FZF and RIPGREP (2019-12-15).

Also make sure to study all the other package declarations in the Selection candidates and search methods section, to appreciate their interplay and the full extent of my customisations.

(use-package counsel
  :disabled                             ; switched to `icomplete'
  :after ivy
  (setq counsel-yank-pop-preselect-last t)
  (setq counsel-yank-pop-separator "\n—————————\n")
  (setq counsel-rg-base-command
        "rg -SHn --no-heading --color never --no-follow --hidden %s")
  (setq counsel-find-file-occur-cmd; TODO Simplify this
        "ls -a | grep -i -E '%s' | tr '\\n' '\\0' | xargs -0 ls -d --group-directories-first")

  (defun prot/counsel-fzf-rg-files (&optional input dir)
    "Run `fzf' in tandem with `ripgrep' to find files in the
present directory.  If invoked from inside a version-controlled
repository, then the corresponding root is used instead."
    (let* ((process-environment
            (cons (concat "FZF_DEFAULT_COMMAND=rg -Sn --color never --files --no-follow --hidden")
           (vc (vc-root-dir)))
      (if dir
          (counsel-fzf input dir)
        (if (eq vc nil)
            (counsel-fzf input default-directory)
          (counsel-fzf input vc)))))

  (defun prot/counsel-fzf-dir (arg)
    "Specify root directory for `counsel-fzf'."
    (prot/counsel-fzf-rg-files ivy-text
                                (concat (car (split-string counsel-fzf-cmd))
                                        " in directory: "))))

  (defun prot/counsel-rg-dir (arg)
    "Specify root directory for `counsel-rg'."
    (let ((current-prefix-arg '(4)))
      (counsel-rg ivy-text nil "")))

  ;; TODO generalise for all relevant file/buffer counsel-*?
  (defun prot/counsel-fzf-ace-window (arg)
    "Use `ace-window' on `prot/counsel-fzf-rg-files' candidate."
    (ace-window t)
    (let ((default-directory (if (eq (vc-root-dir) nil)
      (if (> (length (aw-window-list)) 1)
          (find-file arg)
        (find-file-other-window arg))
      (balance-windows (current-buffer))))

  ;; Pass functions as appropriate Ivy actions (accessed via M-o)
   '(("r" prot/counsel-fzf-dir "change root directory")
     ("g" prot/counsel-rg-dir "use ripgrep in root directory")
     ("a" prot/counsel-fzf-ace-window "ace-window switch")))

   '(("r" prot/counsel-rg-dir "change root directory")
     ("z" prot/counsel-fzf-dir "find file with fzf in root directory")))

   '(("g" prot/counsel-rg-dir "use ripgrep in root directory")
     ("z" prot/counsel-fzf-dir "find file with fzf in root directory")))

  ;; Remove commands that only work with key bindings
  (put 'counsel-find-symbol 'no-counsel-M-x t)
  :bind (("M-x" . counsel-M-x)
         ("C-x C-f" . counsel-find-file)
         ("s-f" . counsel-find-file)
         ("s-F" . find-file-other-window)
         ("C-x b" . ivy-switch-buffer)
         ("s-b" . ivy-switch-buffer)
         ("C-x B" . counsel-switch-buffer-other-window)
         ("s-B" . counsel-switch-buffer-other-window)
         ("C-x d" . counsel-dired)
         ("s-d" . counsel-dired)
         ("s-D" . dired-other-window)
         ("C-x C-r" . counsel-recentf)
         ("s-m" . counsel-mark-ring)
         ("s-r" . counsel-recentf)
         ("s-y" . counsel-yank-pop)
         ("C-h f" . counsel-describe-function)
         ("C-h v" . counsel-describe-variable)
         ("M-s r" . counsel-rg)
         ("M-s g" . counsel-git-grep)
         ("M-s l" . counsel-find-library)
         ("M-s z" . prot/counsel-fzf-rg-files)
         :map ivy-minibuffer-map
         ("C-r" . counsel-minibuffer-history)
         ("s-y" . ivy-next-line)        ; Avoid 2× `counsel-yank-pop'
         ("C-SPC" . ivy-restrict-to-matches))) DEPRECATED Counsel and projectile

WARNING 2020-02-10: I have switched to a simpler, custom solution that uses the built-in icomplete tool. All deprecated packages contain the special :disabled keyword. For the new system, refer to the previous sections:

Below is the latest state of these settings prior to the switch.

I only really need a handful of commands from this setup, namely the ability to switch between buffers or directories in a project and to switch between projects. For the rest, I am already covered by what I have defined in the Counsel section above.

In case you want to use counsel-projectile to its full extent, then enable and configure :bind-keymap. That key binding is just the common prefix to a list of key chords. You can learn about them with M-s p C-h (so append C-h). Note though that the keys I define below do not require such a prefix. They call the commands I need directly.

The counsel-projectile-switch-project is configured to run the fourth action of those available, which configures it to switch to a project and open it inside of dired, rather than ask for a file to open.

While the ivy-initial-inputs-alist extends the variable that is specified in the Ivy section.

(use-package counsel-projectile
  :disabled                             ; switched to `icomplete'
  (add-to-list 'ivy-initial-inputs-alist '(counsel-projectile-switch-project . ""))
  :hook (after-init-hook . counsel-projectile-mode)
  ;; :bind-keymap ("M-s p" . projectile-command-map)
  :bind (("M-s b" . counsel-projectile-switch-to-buffer)
         ("M-s d" . counsel-projectile-find-dir)
         ("M-s p" . (lambda ()
                      (counsel-projectile-switch-project 4)))))

9.1.4 DEPRECATED Swiper commands and settings

WARNING 2020-02-10: I have switched to a simpler, custom solution that uses the built-in icomplete tool. All deprecated packages contain the special :disabled keyword. For the new system, refer to the previous sections:

Below is the latest state of these settings prior to the switch.

This is the search tool that is powered by Ivy. I use it to get an overview of the matching candidates when performing a more complex search. It is not intended as a drop-in replacement for isearch (see section on Isearch), especially since the latter is better for recording macros and jumping to visible sections in the buffer.

Given that Swiper is related to Ivy, do not forget to review the entire section on Selection candidates and search methods.

(use-package swiper
  :disabled                             ; switched to `icomplete'
  :after ivy
  (setq swiper-action-recenter t)
  (setq swiper-goto-start-of-match t)
  (setq swiper-include-line-number-in-search t)
  :bind (("C-S-s" . swiper)
         ("M-s s" . swiper-multi)
         ("M-s w" . swiper-thing-at-point)
         :map swiper-map
         ("M-%" . swiper-query-replace)))

9.1.5 DEPRECATED Ivy extensions

These tools build on the foundation of Ivy and friends. DEPRECATED Ivy rich

WARNING 2020-02-10: I have switched to a simpler, custom solution that uses the built-in icomplete tool. All deprecated packages contain the special :disabled keyword. For the new system, refer to the previous sections:

Below is the latest state of these settings prior to the switch.

With this package we can make good use of the plenty of empty space left by Ivy's default presentation of its items. It enhances several commands, providing each of them with additional information that is pertinent to the task at hand. For example, M-x contains function descriptions, while the buffer list includes information about the major mode and file system path of the items.

The ivy-rich-path-style offers an abbreviation of the file system path that an item is referring to. So far, the only noticeable difference over an absolute value is the use of the tilde (~) instead of /home/USER/ in the buffer list. Whereas recentf continues to display absolute paths. Will need to test this further…

(use-package ivy-rich
  :disabled                             ; switched to `icomplete'
  :after ivy
  (setq ivy-rich-path-style 'abbreviate)

  (setcdr (assq t ivy-format-functions-alist)
  (ivy-rich-mode 1)) DEPRECATED Ivy posframe

WARNING 2020-02-10: I have switched to a simpler, custom solution that uses the built-in icomplete tool. All deprecated packages contain the special :disabled keyword. For the new system, refer to the previous sections:

Below is the latest state of these settings prior to the switch.

This package allows us to reposition Ivy's window anywhere inside of the Emacs frame by placing it inside of a so-called "child frame", i.e. the "posframe". Furthermore, it is possible to use this feature on a per-command basis, all while assigning a different height to each function.

The ivy-posframe-parameters can be found in the source code of the main library. Do M-x find-library RET posframe RET where RET means to hit the Return key in order to proceed. In one of the parameters I show how to define a font that is different from what would normally be used (see this document's section on fonts).

(use-package ivy-posframe
  :disabled                             ; switched to `icomplete'
  :after ivy
  (setq ivy-posframe-parameters
        '((left-fringe . 2)
          (right-fringe . 2)
          (internal-border-width . 2)
          ;; (font . "DejaVu Sans Mono-10.75:hintstyle=hintfull")
  (setq ivy-posframe-height-alist
        '((swiper . 15)
          (swiper-isearch . 15)
          (t . 10)))
  (setq ivy-posframe-display-functions-alist
        '((complete-symbol . ivy-posframe-display-at-point)
          (swiper . nil)
          (swiper-isearch . nil)
          (t . ivy-posframe-display-at-frame-center)))
  (ivy-posframe-mode 1))

9.2 DEPRECATED Reading email with MU4E

WARNING (2020-01-30): I am no longer using MU4E as I switched to Gnus. This section is only kept for reference.

The following configure mu4e, the Mail User Agent. An overview:

  • Include mu4e in the load path. This is necessary because we are using the GNU/Linux distro's package.
  • Use mu4e as the default MUA in Emacs. This concerns actions such as C-x m (compose-mail).
  • I prefer to run the "get mail" command manually rather than rely on a timer. The idea is that when I have time to check my email, I can also refresh its index.
  • Do not provide verbose output about indexing operations.
  • Update manually, because I anyway interact with email only when I have time to check it.
  • Use my selected completion framework where relevant.
  • Define my signature and include it in new messages.
  • Specify the directory where mail is stored. This is where offlineimap is configured to place its findings. Each email account has its own subdirectory therein.
  • Careful with this: Store sent messages in their appropriate place (defined in the "contexts" file—see further below). The docs suggest that IMAP accounts should opt for either moving messages to the trash directory or outright deleting them. The idea is that IMAP is supposed to handle this stuff automatically—my initial tests with my configs do not confirm this, which is why I just tell it to place them in the "sent" directory.
  • Do not kill message buffer upon exit from it. It can always be useful to quickly check something.
  • Always show email addresses (the default is to display just the name).
  • The variables that concern mu4e contexts are relevant because of prot/mu4e-contexts. It loads the mu4e-contexts with all the information about my account setup. I set this in a private and encrypted file. Do check the example in the official docs. I based my work off of it.
  • Message citation is just an improved format for quoting a message in a reply. The format looks like "On 2019-12-09, 16:50 (CET), PERSON <EMAIL> wrote:".
(use-package mu4e
  :disabled                             ; DEPRECATED in favour of `gnus'
  :load-path "/usr/share/emacs/site-lisp/mu4e"
  :after (smtpmail smtpmail-async)
  :commands (mu4e mu4e-update-mail-and-index)
  (setq mail-user-agent 'mu4e-user-agent)
  (setq mu4e-get-mail-command "offlineimap")
  (setq mu4e-hide-index-messages t)
  (setq mu4e-update-interval nil)
  (setq mu4e-completing-read-function 'completing-read)
  (setq mu4e-compose-signature "Protesilaos Stavrou\\n")
  (setq mu4e-compose-signature-auto-include t)
  (setq mu4e-maildir "~/Maildir")
  (setq mu4e-sent-messages-behavior 'sent)
  (setq message-kill-buffer-on-exit nil)
  (setq mu4e-view-show-addresses t)
  (setq mu4e-context-policy 'pick-first)
  (setq mu4e-compose-context-policy 'ask)
  (setq message-citation-line-format "On %Y-%m-%d, %R (%Z), %f wrote:\n")
  (setq message-citation-line-function 'message-insert-formatted-citation-line)

  (defun prot/mu4e-contexts ()
    "Loads a file with the specifics of my email account info."
    (let ((mails "~/.emacs.d/mu4e-contexts.el.gpg"))
      (when (file-exists-p mails)
        (load-file mails))))
  :hook (after-init-hook . prot/mu4e-contexts)
  :bind (:map mu4e-main-mode-map
              ("g" . mu4e-update-mail-and-index)))

To set up offlineimap I used the Arch Wiki entry.

9.2.1 DEPRECATED mu4e extension for org-capture

WARNING (2020-01-30): I am no longer using MU4E as I switched to Gnus. This section is only kept for reference.

With this little snippet, we allow org-capture convert any email into a note, to-do item or whatever. The killer feature is that we get a direct link back to the original email. This way, we can avoid the problem of searching through a pile of messages until we find the one we really need. Nice!

(use-package org-mu4e                   ; no need for `:ensure'
  :disabled                             ; DEPRECATED in favour of `gnus'
  :after (org mu4e)
  (setq org-mu4e-link-query-in-headers-mode nil))