Correction -- the "lock-up" I observed might have been due to paging (possibly related to emacs - I'll have to investigate later). Toggling auto-revert-avoid-polling appeared to reduc emacs CPU usage, according to top(1) - it's a bit difficult to tell because there seems to be some lag in when the command takes effect and my compilation process switches a certain amount between CPU-intensive and IO-intensive. Anyway, CPU seemed to drop from ~60% to ~30% (please take this observation with a large grain of salt). I'll try more things in the future and report if anything interesting shows up. - peter On Mon, 6 Jan 2020 at 16:01, Peter Ludemann wrote: > If anything, auto-revert-avoid-polling makes responsiveness worse -- the > window locked up a few times on me while doing ctrl-N while running a > CPU-intensive (and possibly IO-intensive) compilation. > Although I might not have set the value correctly ... I did M-x > customise-variable RET auto-revert-avoid-polling RET then "toggle" then > "apply". > . > > On Sun, 5 Jan 2020 at 11:57, Mattias Engdegård wrote: > >> 5 jan. 2020 kl. 20.31 skrev Peter Ludemann : >> >> > Which version of Emacs would you like me to try this with? And what >> result are you expecting/hoping to see? (e.g., might it reduce the current >> 30-80% CPU load for polling with emacs 28.0.50?) There are only a few open >> files directly under /tmp, so would this have any effect or does it >> propagate down to subdirectories? >> > >> > [Also, I'd need a few more details (not being an emacs-internals >> person) ... should I add this to my .emacs and restart, or execute in a >> scratch buffer, or ...?] >> >> 'auto-revert-avoid-polling' is a single global customisable variable, so >> you would set it using >> >> M-x customise-variable RET auto-revert-avoid-polling RET >> >> , then turn it on and apply the change. I believe it was introduced in >> Emacs 27. >> >> The idea is to save CPU by not having to look at files periodically to >> see if they have changed. I have no idea if you would see any improvement >> at all, but it shouldn't make anything worse. >> >>