Slowness relative to regular magit

[Macavirus]

I use magit extensively and wondering if this makes any subjective performance difference vs. regular magit behaviour. If so, that would be great to see in the readme!

I think the slowdown I see regularly in magit might have something to do with the emacs fontifying/long lines issues that have been reported elswehere and nothing to do with magit itself, but unsure.

[dandavison]

Hi @Macavirus, sorry to be slow here. I don't think it does: the overhead of calling delta and converting the ANSI escape sequences to emacs text properties is, I believe, small/negligible relative to the multiple external process calls that magit makes.

My personal emacs setup does have some annoying performance issues involving magit which I haven't tracked down: in particular, I find that up/down arrow (magit-prev-line /magit-next-line) in magit diff buffers become very slow sometimes but I still haven't worked out why. (I do need to double-check that still occurs without magit-delta.)

[dandavison]

I find that up/down arrow (magit-prev-line /magit-next-line) in magit diff buffers become very slow sometimes but I still haven't worked out why.

(I do need to double-check that still occurs without magit-delta.)

It sounds like you might have had something specific in mind yourself though?

[mpereira]

@dandavison: in particular, I find that up/down arrow (magit-prev-line /magit-next-line) in magit diff buffers become very slow sometimes but I still haven't worked out why.

What does the profiler show?

[dandavison]

What does the profiler show?

Good question. I think I looked at this before, but I'll try to do it again when it next happens.

[dandavison]

OK it happened again. It seems to be saying that the time is spent in line-move-visual.

   - magit-next-line                                               35   7%
    - line-move                                                    35   7%
       line-move-visual                                            33   6%

(The percentages are low because it captured completing-read from me doing M-x profiler-stop.)

[braineo]

The performance issue I saw was staging and unstaging hunks when there is a huge diff in the repo.

For example, famous package-lock.json can typically have 10,000+ lines 😈 , even though not manipulating this huge json file and it is collapsed in diff, staging other file caused 100% CPU usage and cannot recover sometimes.

[dandavison]

@braineo how is performance in the situation you describe when using magit without magit-delta?

[braineo]

@dandavison

emacs 27.1.5 build from source, CPU usage is quite different only by showing magit-status without touching emacs with 2000 line difference in the repo

 package-lock.json | 2171 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++-----------------------------------------------------------
 package.json      |    2 +-

With magit-delta

Without magit-delta

(use-package magit-delta
  :disabled
  :if (executable-find "delta")
  :hook
  ((magit-mode . (lambda () (magit-delta-mode +1))))
  :config
  ;; Cannot use line number feature of delta in magit. refer to https://github.com/dandavison/magit-delta/issues/13
  (setq magit-delta-delta-args (append magit-delta-delta-args '("--features" "magit-delta"))))

.gitconfig

[interactive]
    diffFilter = delta --color-only
[delta]
    syntax-theme = Sublime Snazzy
    features = line-numbers custom-style
    whitespace-error-style = 22 reverse
[delta "custom-style"]
    plus-style = syntax "#203624"
    minus-style = syntax "#362020"
[delta "magit-delta"]
    whitespace-error-style = 22 reverse

[dandavison]

Thanks, interesting! Is it a public repo? If so a commit / any other instructions to reproduce would be great.

[braineo]

That will be trivial

Any nodejs projects will do, for example

git clone git@github.com:apollographql/apollo-server.git
git reset HEAD~20

If it freezes emacs just change how many commits you want to leap back 😛

[sankalp-khare]

It really slowed down my magit, even though the same stuff seems to work fine on command line git. had to disable the mode (these were some not-so-large .tfstate files)

[mpereira]

I had to disable it too. Having magit-delta enabled (with delta 0.6.0) caused creating magit buffers showing diffs much slower. Profiling shows that magit-git-wash is the culprit.

[ntharim]

The workaround I'm doing now is enabling magit-delta unless there are large diffs in the buffer:

(defvar nth/magit-delta-point-max 50000)
;; Disable mode if there are too many characters
(advice-add 'magit-delta-call-delta-and-convert-ansi-escape-sequences :around
            (defun nth/magit-delta-colorize-maybe-a (fn &rest args)
              (if (<= (point-max) nth/magit-delta-point-max)
                  (apply fn args)
                (magit-delta-mode -1))))
;; Re-enable mode after `magit-refresh' if there aren't too many characters
(add-hook 'magit-post-refresh-hook
          (defun nth/magit-enable-magit-delta-maybe-h (&rest _args)
            (when (and (not magit-delta-mode)
                       (<= (point-max) nth/magit-delta-point-max))
              (magit-delta-mode +1))))

[dandavison]

Sorry for not giving this much attention. It's not because I'm unaffected -- I use magit-delta daily and I encounter performance issues too (moving point in magit-status-mode slows to a crawl), but they go away on restarting Emacs (no idea why yet, but it has made me slow to get to the bottom of it).

@mpereira can you give instructions for profiling? It sounds like you have done this much more successfully than me -- I have so far failed to profile in a way that points to any relevant function.

[mpereira]

No worries @dandavison, and no pressure 🙂

moving point in magit-status-mode slows to a crawl

This happens to me sporadically, even without using delta or magit-delta. I consistently fix it by

1. M-x ibuffer
2. filtering buffers by name with the pattern "magit" (ibuffer-filter-by-name)
3. marking all buffers for deletion (ibuffer-mark-for-delete)
4. ibuffer-do-kill-on-deletion-marks

can you give instructions for profiling?

What I did was

1. install and configure delta/magit-delta
2. go to a git repository with lots of staged/unstaged changes
3. start a profile with M-x profiler-start
4. M-x magit-status
5. wait for the buffer to open (depending on the diff size, it might take a while)
6. M-x profiler-stop followed by M-x profiler-report

That should show you the function tree with the number of cycles samples collected in each function. In my case I could see magit-git-wash (which I think renders gif diffs) taking a big chunk of the time.