This is a fuzzy Emacs completion style similar to the built-in flex
style,
but with a better scoring algorithm.
Specifically, it is non-greedy and ranks completions that match at
word; path component; or camelCase boundaries higher.
To use hotfuzz, add it to the completion-styles
list:
(setq completion-styles '(hotfuzz))
If using Fido, its insistence on using flex
needs to be thwarted:
(add-hook 'icomplete-minibuffer-setup-hook
(lambda () (kill-local-variable 'completion-styles)))
The following ordinary Emacs completion options are adhered to:
completion-ignore-case
specifies whether matching is case-insignificant.- The
completions-common-part
face is used to highlight what characters of a candidate the search string matched.
Unless the completion UI supports completion-lazy-hilit
, as i.a.
Vertico and Corfu do, only the first
hotfuzz-max-highlighted-completions
completions will be
highlighted out of performance concerns. The default value is large
enough that generally the list of completions will need to be
scrolled beyond the second page to reach non-highlighted
completions, but this optimization may be disabled with:
(setq hotfuzz-max-highlighted-completions most-positive-fixnum)
Optionally, you may compile the bundled dynamic module for improved performance. Ensure GCC, CMake and GNU Make or similar are present, and run
cmake -B build -DCMAKE_BUILD_TYPE=Release -DCMAKE_C_FLAGS=-march=native &&
cmake --build build
and place the resulting shared library somewhere in load-path
.
Add -DEMACS_ROOT=/path/to/emacs-root
(the directory containing bin/
, include/
, etc.)
if it is not found automatically by locating emacs
in PATH
.
The module will be picked up when hotfuzz is loaded;
use (featurep 'hotfuzz-module)
to check for its presence.
Note
Dynamic modules are unable to access invalid Unicode strings.
Consult appends invisible tofus, characters outside the Unicode range (unlikely to match search strings), to attach line numbers and disambiguate completions. Using, e.g., the Supplementary Private Use Area-B instead circumvents encoding issues:
(setq consult--tofu-char #x100000
consult--tofu-range #x00fffe)
The flex
completion style included with Emacs
does the same matching as hotfuzz, and scores candidates based on gap sizes.
It does not, however, attempt to find the optimal score.
For example, given the search string "foo"
,
the matched characters in a candidate could look like
xfxxxoxxxoxfoox
which would score low despite the later contiguous match.
The flx package - which out-of-the-box only supports Ido - has scoring criteria similar to those used by hotfuzz, but works a little differently. Its bountiful use of caching means it can be faster at scoring long candidates. Since the ordering of completions differs between flx and hotfuzz you are encouraged to try both.
The fussy completion style is generic over different fuzzy scoring backends, flx and the Emacs Lisp implementation of hotfuzz being two of them. fussy also implements caching of filtered candidates. Although some of the scoring backends are implemented as dynamic modules, hotfuzz has the advantage of passing all completion candidates to its dynamic module en masse, allowing sorting and filtering to be parallelized.
The orderless completion style allows
every component of a space-delimited (by default) pattern
to match according to any other completion style.
It is very customizable,
but does no sorting and allows the individual sub-patterns to overlap
("foo foo"
filters no additional items compared to "foo"
).
Hotfuzz on the other hand tries to be more clever about sorting,
and so users who dislike that may prefer orderless.