Oh, you're so funny. What you've been "editing" buddy, your gramma's cookie recipes? Just the other day I transcribed a three-hour long video of a meetup, picking the model, setting parameters, running the process; then scrolling through Reddit while waiting for it to get done; then I opened the transcript in subtitle-editing mode that allowed me to "follow" the video - I would scroll through the subs and it would move the player pointer, I'd watch the part, pause, move back and forth, etc. I then sent a request to a model asking it to summarize it, extracted pieces for my notes. I explored URLs from the meeting by OCRing them automatically - I didn't have to manually type jackshit, I'd point to a frame, select the region on the screen, and voila - the url appears in the buffer. I then reverse-url-searched those links, I was curious of what people been talking about that stuff on HN. Then I grabbed the links shared in each thread - single key-press for me. All that without leaving Emacs - no context switching, no focus-wandering, no lollygagging. And that's just a regular Wednesday. The crazy part, that I could just ask an LLM do all that - it could actively drive the process, directly opening things, moving between the buffers, gathering data, etc. Or I could ask it to make it into a deterministic script - if I wanted to have it for automation or whatever.
You "chose to use an editor", pfff... bro, my key-switches are dying of laughing.
I thought Emacs was terrible to use until I realized it was extremely fast on my home computer. All the pauses that EDR does combined with Emacs mostly single threadedness is a bummer. I still daily drive it now for my IDE at work because it's workflows work better for me than others, but it is such a bummer.
I definitely notice an EDR tax on macOS generally, but I haven't noticed Emacs having issues specifically. It's usually certain I/O operations like Git or tarball extraction that become slow.
I notice it on macOS, which I use for work. I use Linux personally but no EDR there :)
There are just so many little operations that cause stuttering on macOS with an unfortunately over-eager crowdstrike configuration. Magit is the worst offender but it causes paper cuts all over :(
So, on Mac there are two main problems - displaying and fork/exec.
Rendering on Mac is slower because NS port display backend just sucks. Emacs's redisplay emits many small drawing operations per frame - one per glyph run, cursor, or fringe bitmap. X11/cairo and pgtk on Linux batch those very efficiently - Cocoa has nothing comparable here.
You can try installing one of the forks of Mitsuharu's port - they usually have --with-metal flag and some other improvements, but they are typically one-two versions lagging behind, and in my experience they are not always stable. In general, possibility of Emacs dying on Mac is not zero, on Linux is much more rock solid.
And then the forking problem. Emacs has a big heap, so each fork is expensive. On Mac it triggers all sorts of crap - code-signature validation, Gatekeeper/XProtect scanning, TCC checks. That's dozens of milliseconds per subprocess and Linux has no such crap to deal with - shit is just fast. Magit calls git dozens of times per refresh. Now multiply that and then you get usual complains - pestilence on poor Jonas Bernoulli: "I love Magit, but why is it so damn slow..." etc.
Here are some things that I do to mitigate:
- Do these and never touch keyboard rate settings in the UI, just do it, reboot and thank me later.
# "Set a blazingly fast keyboard repeat rate, "
defaults write NSGlobalDomain KeyRepeat -int 1
# "Set a shorter Delay until key repeat"
defaults write NSGlobalDomain InitialKeyRepeat -int 9
- I turned off Spotlight indexer completely - I don't search for files through it nor through Alfred.
sudo mdutil -a -i off
- Set `core.fsmonitor true` in gitconfig. Add `core.untrackedCache true` to go with `status.showuntrackedfiles all`, otherwise the fsmonitor win is partly eaten by the untracked scan.
- Remove sections from Magit you don't need all the time, e.g.:
;; who cares if tags not displayed in magit-refs buffer?
(remove-hook 'magit-refs-sections-hook #'magit-insert-tags)
There are some other things, but that's what I remember off the top of my head, if I recall some more, will update.
It's possible to determine good values automatically: Measure your typing speed.
You can do it in the background with:
https://pavpanchekha.com/blog/emacs-keylogger.html
and periodically update the latency parameters to be just over your natural typing speed.
I just chose to use an editor whose very architecture doesn't require nonsense like this so the editor doesn't freeze on me.
Oh, you're so funny. What you've been "editing" buddy, your gramma's cookie recipes? Just the other day I transcribed a three-hour long video of a meetup, picking the model, setting parameters, running the process; then scrolling through Reddit while waiting for it to get done; then I opened the transcript in subtitle-editing mode that allowed me to "follow" the video - I would scroll through the subs and it would move the player pointer, I'd watch the part, pause, move back and forth, etc. I then sent a request to a model asking it to summarize it, extracted pieces for my notes. I explored URLs from the meeting by OCRing them automatically - I didn't have to manually type jackshit, I'd point to a frame, select the region on the screen, and voila - the url appears in the buffer. I then reverse-url-searched those links, I was curious of what people been talking about that stuff on HN. Then I grabbed the links shared in each thread - single key-press for me. All that without leaving Emacs - no context switching, no focus-wandering, no lollygagging. And that's just a regular Wednesday. The crazy part, that I could just ask an LLM do all that - it could actively drive the process, directly opening things, moving between the buffers, gathering data, etc. Or I could ask it to make it into a deterministic script - if I wanted to have it for automation or whatever.
You "chose to use an editor", pfff... bro, my key-switches are dying of laughing.
Oh, you use Lazarus? Cool, it's an amazing project. Too bad it's a bit of an unsung hero, it deserves it's day in the limelight.
Cool story bro.
I've got an emacs uptime measured in years, so I have no idea what you're talking about, but I'm happy that you're happy.
You know what makes Emacs feel slow?
Corporate EDR BS.
I thought Emacs was terrible to use until I realized it was extremely fast on my home computer. All the pauses that EDR does combined with Emacs mostly single threadedness is a bummer. I still daily drive it now for my IDE at work because it's workflows work better for me than others, but it is such a bummer.
(I think EDR here means https://en.wikipedia.org/wiki/Endpoint_detection_and_respons... )
Do you notice this on macOS or Linux?
I definitely notice an EDR tax on macOS generally, but I haven't noticed Emacs having issues specifically. It's usually certain I/O operations like Git or tarball extraction that become slow.
I notice it on macOS, which I use for work. I use Linux personally but no EDR there :)
There are just so many little operations that cause stuttering on macOS with an unfortunately over-eager crowdstrike configuration. Magit is the worst offender but it causes paper cuts all over :(
> I notice it on macOS
So, on Mac there are two main problems - displaying and fork/exec.
Rendering on Mac is slower because NS port display backend just sucks. Emacs's redisplay emits many small drawing operations per frame - one per glyph run, cursor, or fringe bitmap. X11/cairo and pgtk on Linux batch those very efficiently - Cocoa has nothing comparable here.
You can try installing one of the forks of Mitsuharu's port - they usually have --with-metal flag and some other improvements, but they are typically one-two versions lagging behind, and in my experience they are not always stable. In general, possibility of Emacs dying on Mac is not zero, on Linux is much more rock solid.
And then the forking problem. Emacs has a big heap, so each fork is expensive. On Mac it triggers all sorts of crap - code-signature validation, Gatekeeper/XProtect scanning, TCC checks. That's dozens of milliseconds per subprocess and Linux has no such crap to deal with - shit is just fast. Magit calls git dozens of times per refresh. Now multiply that and then you get usual complains - pestilence on poor Jonas Bernoulli: "I love Magit, but why is it so damn slow..." etc.
Here are some things that I do to mitigate:
- Do these and never touch keyboard rate settings in the UI, just do it, reboot and thank me later.
- disable App Nap for Emacs: - I turned off Spotlight indexer completely - I don't search for files through it nor through Alfred. - Set `core.fsmonitor true` in gitconfig. Add `core.untrackedCache true` to go with `status.showuntrackedfiles all`, otherwise the fsmonitor win is partly eaten by the untracked scan.- Remove sections from Magit you don't need all the time, e.g.:
There are some other things, but that's what I remember off the top of my head, if I recall some more, will update.For me it can be slow, especially refreshing magit (which probably is blocked by git, blocked by the bs)(on MacOS)
Try checking what M-x magit-toggle-verbose-refresh has to say, you may be able to identify what's slow and optimize around it
magit is brutal when EDR is involved.