GNU bug report logs - #78545
31.0.50; project-mode-line is slow because it tries to read files on each update

Previous Next

Package: emacs;

Reported by: Yikai Zhao <yikai <at> z1k.dev>

Date: Thu, 22 May 2025 06:17:05 UTC

Severity: normal

Found in version 31.0.50

Full log


View this message in rfc822 format

From: Daniel Colascione <dancol <at> dancol.org>
To: Yikai Zhao <yikai <at> z1k.dev>
Cc: 78545 <at> debbugs.gnu.org, shipmints <at> gmail.com
Subject: bug#78545: 31.0.50; project-mode-line is slow because it tries to read files on each update
Date: Thu, 22 May 2025 13:21:18 -0400

On May 22, 2025 1:02:02 PM EDT, Yikai Zhao <yikai <at> z1k.dev> wrote:
>On Thu, May 22, 2025 at 23:54 Daniel Colascione <dancol <at> dancol.org> wrote:
>
>>
>>
>> On May 22, 2025 7:47:18 AM EDT, Ship Mints <shipmints <at> gmail.com> wrote:
>> >On Thu, May 22, 2025 at 2:17 AM Yikai Zhao <yikai <at> z1k.dev> wrote:
>> >
>> >> After enabling `project-mode-line`, I find my emacs slightly less
>> >> responsive. I think it's because project-mode-line is slow due to
>> >> it trying to read files on each update. Here's how I reproduce it:
>> >>
>> >> - Run `emacs -Q`
>> >> - Evaluate `(setopt project-mode-line t)`
>> >> - Open some random file in a non-project directory, e.g. `/tmp/test`,
>> >> start typing
>> >>
>> >> Here's part of the profiling result:
>> >>
>> >>          114  62% - redisplay_internal (C function)
>> >>           42  23%  - eval
>> >>           41  22%   - project-mode-line-format
>> >>           41  22%    - project-current
>> >>           41  22%     - project--find-in-directory
>> >>           41  22%      - run-hook-with-args-until-success
>> >>           41  22%       - project-try-vc
>> >>           41  22%        - project-try-vc--search
>> >>           25  13%         - project--value-in-dir
>> >>           22  12%          - hack-dir-local-variables
>> >>           22  12%           - #<byte-code-function 24E>
>> >>           22  12%            - hack-dir-local--get-variables
>> >>           22  12%             - dir-locals-find-file
>> >>           22  12%              - locate-dominating-file
>> >>           12   6%               + abbreviate-file-name
>> >>            3   1%          + #<byte-code-function 98F>
>> >>           13   7%         - locate-dominating-file
>> >>            6   3%          + #<byte-code-function 4D9>
>> >>            3   1%            abbreviate-file-name
>> >>            3   1%
>> >>
>> >> It shows that `project-mode-line-format` can take a significant portion
>> >> of the redisplay time. It increases even more when the directory is deep
>> >> or the disk is slow.
>> >>
>> >> IMO the performance of this function should be improved.
>> >>
>> >> I have two questions:
>> >>
>> >> 1. Is it possible to simply cache the result as buffer-local?
>> >> 2. IIUC some of the file reading comes from `project--value-in-dir`,
>> >> which opens a temporary buffer to load dir local variables. Why doesn't
>> >> it simply read the variable value in the current buffer?
>> >>
>> >
>> >I've advised project functions to cache the "invariants" current project
>> >and project root to incur these costs once per buffer.  I also cache the
>> >nuance of a "non project" to avoid repeated project probing on buffers
>> that
>> >don't have a project.
>> >
>> >I thought maybe I was the only person silly enough to invoke costly
>> project
>> >functions for every tab in `tab-bar` and every mode-line update.
>> >
>> >I could contribute optional caching to project.el if there is sufficient
>> >demand for it.  Unless someone else wants to.
>>
>> We could also just have the IO functions signal error if called during a
>> mode line update.
>
>
>Personally I think this is a great idea. Aside from mode line updates, IO
>functions can cause undesired delay in many other functions (like tab bar
>or maybe post command hook). If we have a general way to mark certain hooks
>as “should not block by IO” and be able to capture those violations, it
>would be very helpful.

I've been meaning to write a slow hook detector for a few weeks now. Feel free to beat me to it. I figure we can start by just calling each entry in the major hooks with a new run hooks (with args) variant that records times of each and reports anything over a threshold as a warning.


>
>
>Yikai
>
>
>>




This bug report was last modified 24 days ago.

Previous Next


GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.