Performance Improvements

[jgkamat]

I originally tested alda-mode on small-ish files, but on larger files, it really slows down due to font lock. (such as https://github.com/joshhting/alda-music/blob/master/PotentialForAnything.alda)

I need to do a bit of work on the regexes, and see if I can get any performance improvements. If not, I'll look at reducing the default font lock level.

[wasamasa]

Can you provide actual hard numbers? It's otherwise hard to tell whether there's been an improvement by a change. I have one in mind that might help with this issue and am preparing a PR for it.

[jgkamat]

I'm not sure what you're asking for in terms of hard numbers, but:

• The problem goes away for me when I turn off font-lock-mode, which means (on my end) it's a problem with font lock.
• Here's a profile of holding down 'j/k' in evil-mode in the above file:

- command-execute                                                 817  67%
 - call-interactively                                             817  67%
  - funcall-interactively                                         690  57%
   - evil-previous-line                                           305  25%
    - evil-line-move                                              303  25%
     - previous-line                                              303  25%
      - line-move                                                 303  25%
         line-move-1                                              303  25%
   - evil-next-line                                               289  23%
    - evil-line-move                                              289  23%
     - next-line                                                  289  23%
      - line-move                                                 289  23%
         line-move-1                                              289  23%
   + execute-extended-command                                      76   6%
   + minibuffer-complete-and-exit                                  20   1%
  + byte-code                                                     127  10%
- redisplay_internal (C function)                                 260  21%
 - eval                                                           256  21%
  - sml/generate-minor-modes                                      184  15%
   - rm--mode-list-as-string-list                                 102   8%
    - mapcar                                                       53   4%
       rm-format-mode-line-entry                                   37   3%
     + rm--propertize                                              12   0%
    + mapconcat                                                    49   4%
   + sml/fill-width-available                                      74   6%
  + list                                                           31   2%
    sml/generate-modified-status                                   17   1%
  + eyebrowse-mode-line-indicator                                  12   0%
   file-remote-p                                                    4   0%
+ ...                                                              60   4%
+ sp--save-pre-command-state                                       25   2%
+ evil--jump-hook                                                  24   1%
+ evil-repeat-pre-hook                                              9   0%
  evil-repeat-post-hook                                             4   0%
  internal-timer-start-idle                                         4   0%
  tooltip-hide                                                      4   0%

I'm a bit confused at the profile, since it dosen't point to font-lock or redisplay as the primary curplit (maybe this is just my config, then). I have vim-style scrolling in my config, which is definetly making things worse.

When I made this issue originally, I wanted to try and play around with supporting lower font-lock levels and seeing if that helps with the problem at all, since I had similar problems with c++ mode, when I turned it up to the highest font-lock level.

[wasamasa]

I mean something like running font-lock-profiler-buffer on the problematic buffer.

edit: Results:

Modified:

   0             0             0%  ("^" (0 form-feed--font-lock-face t))
  10             0             0%  ("[{}]" (0 font-lock-builtin-face))
  12             0             0%  ((lambda (_) (let (case-fold-search) (and (re-search-forward hl-todo-regexp nil t) (nth 8 (syntax-ppss))))) (1 (hl-todo-get-face) t t))
   3             6             2%  ("\\([a-zA-Z]\\{2\\}[A-Za-z0-9_-]*\\)\\( *\\(\"[A-Za-z0-9_-]*\"\\)\\)?:" (0 font-lock-type-face))
   8             6             0%  ("[@%][a-zA-Z]\\{2\\}[a-zA-Z0-9()+-]*" (0 font-lock-builtin-face))
   2            58             2%  ("\\([Vv][0-9]+\\):" (1 font-lock-function-name-face))
   5            94             2%  ("([a-zA-Z-]+!? +\\(?:[0-9]+\\|\\[\\(?::[a-zA-Z]+ ?\\)+\\]\\))" (0 font-lock-variable-name-face))
   6           119             2%  ("o[0-9]+" (0 font-lock-constant-face))
   1           174             6%  ("|" (0 font-lock-comment-face))
   9           265             4%  ("\\*[0-9]+" (0 font-lock-builtin-face))
  11           532             8%  ("[][]" (0 font-lock-builtin-face))
   7           765            11%  ("[<>]" (0 font-lock-constant-face))
   4          2939            63%  ("[a-gA-GrR]\\([-+_]*\\)\\([~.0-9/]*\\(?:m?s\\)?\\)" (1 font-lock-type-face) (2 font-lock-builtin-face))

Original:

   3             6             3%  ("\\([a-zA-Z]\\{2\\}[A-Za-z0-9_-]*\\)\\( *\\(\"[A-Za-z0-9_-]*\"\\)\\)?:" (1 font-lock-type-face))
   7             6             0%  ("\\([@%][a-zA-Z]\\{2\\}[a-zA-Z0-9()+-]*\\)" (1 font-lock-builtin-face))
  12            21             2%  ("\\([a-gA-GrR] *[-+]+\\)" (1 font-lock-preprocessor-face))
   2            58             1%  ("\\([Vv][0-9]+\\):" (1 font-lock-function-name-face))
   4            94             2%  ("\\(([a-zA-Z-]+!? +\\(\\([0-9]+\\)\\|\\(\\[\\(:[a-zA-Z]+ ?\\)+\\]\\)\\))\\)" (1 font-lock-variable-name-face))
   5           119             2%  ("\\(o[0-9]+\\)" (1 font-lock-constant-face))
   1           174             3%  ("\\(|\\)" (1 font-lock-comment-face))
   9           265             5%  ("\\(\\*[0-9]+\\)" (1 font-lock-builtin-face))
  11           532            10%  ("\\(\\[\\|\\]\\)" (1 font-lock-builtin-face))
   6           765            15%  ("\\(>\\|<\\)" (1 font-lock-constant-face))
   8          2939            56%  ("[a-gA-GrR][ +-]*\\([~.0-9 /]*\\(m?s\\)?\\)" (1 font-lock-builtin-face))

From this I can tell that if anything, you'd be best off to omit the most often applied rule for a less fancy font-lock level.