When would we expect `experimental_shell_command` ...
# general
b
When would we expect
experimental_shell_command
to rerun? I'm seeing it run essentially every daemon restart, which isn't ideal. 🧵
w
cc @ancient-vegetable-10556: ^
b
I'm using
build_file_dir
I would think that wouldn't do it though 🤔
w
Chris identified this same bug about an hour ago, and has started looking at it
b
I can try anything out locally because I'm currently using
pants_from_sources
and also I have a forked impl for
experimental_shell_command
in my repo 😛
Last time I saw this it was an issue with using sets
The debug log lines are the exact same 🤔
It's weird too because it's actually running the command. We don't hit the process cache either
w
yea, Chris determined that this might be a bug in Snapshot digests (!!)… repro is on https://github.com/pantsbuild/pants/issues/16825#issuecomment-1311073810 . i was going to look Monday
b
With my track record recently it's probably related to symlinks 😅🤣
w
that was the first thing that came to mind, yea… the bisect was tricky because it required two
--no-pantsd
builds per commit, so i’m not sure we know exactly when it was introduced. but i’ll probably just work it from the repro on
main
.
to close the loop on this one: it was https://github.com/pantsbuild/pants/pull/17549