News for hackers. Stuff that matters.
Re:Bringing GNU Emacs to Native Code (2020) by unhammer on Tuesday May 04, 2021 @08:30AM (#27036850)
Similarly, I wanted to check what was taking up time on redoing org-agenda, profiler says the majority of the time is spent doing `kill-all-local-variables`

         524  57%            - kill-all-local-variables
          56   6%             + magit-wip-after-save-mode-cmhh
          56   6%             + global-evil-quickscope-mode-cmhh
          56   6%             + global-eldoc-mode-cmhh
          52   5%             + global-page-break-lines-mode-cmhh
          52   5%             + global-font-lock-mode-cmhh
          48   5%             + smartparens-global-mode-cmhh
          44   4%             + global-prettify-symbols-mode-cmhh
          40   4%             + magit-auto-revert-mode-cmhh
          40   4%             + global-tree-sitter-mode-cmhh
          40   4%             + evil-mode-cmhh
          40   4%             + global-evil-collection-unimpaired-mode-cmhh
These are all third-party packages, have nothing to do with org-mode. Pretty much any globalized minor mode you enable will add some constant time to switching into fundamental-mode, and it all adds up.

(Also, half the time when I run the profiler, the culprit ends up being evil-mode. Unfortunately it's so useful …)

Re:Bringing GNU Emacs to Native Code (2020) by TeMPOraL on Tuesday May 04, 2021 @11:49AM (#27038889)
Wow, that's an interesting result, I never noticed it directly - I've seen -cmhh functions before, but never bothered to check what they are before.

As it turns out, these are automatically created by `define-globalized-minor-mode' macro - per docstring, it's used to define a global mode for buffer-local minor modes. I.e. that's how you get all these `global-XXX-mode` functions that turn on XXX-mode in every (relevant) buffer.

One of the things this macro does is create a `global-XXX-cmhh' function, and attaches it to `change-major-mode-hook', which is invoked by `kill-all-local-variables' whenever buffer's mode is changed. I imagine that, as part of generating org mode's agenda, files get opened and modes get switched a lot. Curiously, it also seems that (digging into `kill-all-local-variables' source in C), each call to it forces mode-line redisplay, so further computation may be caused by whatever is in your mode line (which can also be nontrivial). I haven't checked that though, maybe it redisplays mode line only once.

Thanks for bringing this up, it's another place where pretty much every mode stuff some hook, that I didn't know about :).

EDIT: curiously enough, I just profiled my agenda - both doing from scratch and redoing it - and `kill-all-local-variables' isn't even on its radar. I wonder what runs in your hooks (and/or modeline) then. The result from refreshing an existing agenda was roughly what I expected. Building it anew, after killing all open org buffers, was interesting. In my case, 69% of total time was spent opening files (`find-file-noselect'), before even doing any mode switching. In that, `projectile-find-file-hook-function' accounted for 50% of total agenda building time, half of that spent updating the mode line!

From what I see, it's just trying to determine the project to which a file belongs. I'll probably look for an option to make projectile stop being interested in files that aren't in the "known projects" subtrees without explicitly instructing it so.