search

mawww / kakoune

mawww's experiment for a better code editor
http://kakoune.org
The Unlicense
10.01k stars 714 forks source link

Prune hidden files when loading autoload #4556

Open alexherbo2 opened 2 years ago

alexherbo2 commented 2 years ago

The find command used in the runtime kakrc does not prune hidden files, such as .git.

The startup time can be slower if cloning repositories in the autoload.

Example:

git clone https://github.com/mawww/notmuch.kak.git ~/.config/kak/autoload/notmuch.kak

It can also happen when backup files are created, since .kak files keep the same extension.

Screwtapello commented 2 years ago

Some empirical measurements:

With a warmed cache:

hyperfine --export-markdown /dev/stdout \
    --warmup=2 \
    "find -L ~/.config/kak/autoload -type f -name '*\.kak'" \
    "find -L ~/.config/kak/autoload -name .git -prune -o -type f -name '*\.kak'"
Command Mean [ms] Min [ms] Max [ms] Relative
find -L ~/.config/kak/autoload -type f -name '*\.kak' 4.8 ± 0.4 4.3 6.8 3.46 ± 0.40
find -L ~/.config/kak/autoload -name .git -prune -o -type f -name '*\.kak' 1.4 ± 0.1 1.2 2.3 1.00

Explicitly flushing the disk cache before each run:

hyperfine --export-markdown /dev/stdout \
    --prepare "echo 1 | sudo tee /proc/sys/vm/drop_caches" \
    "find -L ~/.config/kak/autoload -type f -name '*\.kak'" \
    "find -L ~/.config/kak/autoload -name .git -prune -o -type f -name '*\.kak'"
Command Mean [ms] Min [ms] Max [ms] Relative
find -L ~/.config/kak/autoload -type f -name '*\.kak' 68.3 ± 39.5 32.5 143.4 6.11 ± 9.23
find -L ~/.config/kak/autoload -name .git -prune -o -type f -name '*\.kak' 11.2 ± 15.6 4.5 75.9 1.00

So, skipping .git directories is a measurable improvement, but probably not much of one.

alexherbo2 commented 2 years ago

How many Git repositories were in the autoload?

Does a long Git history affect the startup time?

Screwtapello commented 2 years ago

How many Git repositories were in the autoload?

 printf '%s\n' ~/.config/kak/autoload/*/.git | wc -l
14

Does a long Git history affect the startup time?

It depends. A git repo with a lot of loose objects (in .git/objects/??/) will probably take longer to scan than a repo whose objects are all in a single pack (in .git/objects/pack/). A Git repo with a long history will have more objects, but it's also more likely that git gc will have been run to collect loose objects into a pack.