^stu had some great comments above, especially som...
# development
^stu had some great comments above, especially some very very interesting remarks on the mypy caching ticket: https://github.com/pantsbuild/pants/issues/10864#issuecomment-699689828. i'm putting the doc on hold for a bit until i can answer the questions raised, and i will take a look into
today i think as per the linked comment
Yea, the brfs angle is maybe a thing...1) It would allow the input paths to all be stable, 2) could fake all the timestamps to something stable
🔥 1
It might look like an atomic swap below the running process, and then re-run.
ah omg yes
didn't think about metadata like timestamps but that would cover even such ridiculous caches as the intellij cache
The only thing is, FUSE does not send file watching events on Linux (it does on OSX), so I'm not sure that that strategy would work for mypy's daemon
IntelliJ might have an issue with that too
i'm also under the impression that the FUSE driver itself is something we can hack on. i suppose in linux that would literally have to involve a PR to the kernel so will have to balance that with expected benefit. i will review the different FUSE capabilities across systems since all of that was news to me