and in fact I'm running tests in the toolchain repo, not even pants's own tests

The best idea I have for this would be to keep the check in

./pants

for if the CLI args include

test.*

, then if this the case and the hash has changed for the source files, run

./pants --tag=integration list globs-originally-passed-to-pants-script

. If that returns with something, then regen the PEX. Otherwise, we know it’s just unit tests and can ignore Alternatively, introduce an env var like

NO_REGEN_PEX

to turn off the mechanism 🙂 -- The reason for all this complication is that we have this logic live at the bash script level. We considered having it instead live in

pants-plugins

or `src`—i.e. the logic of whether to call `./pants binary src/python/pants:local_binary && mv dist/pants.pex pants.pex`—but we decided this was too complicated. So, because this logic lives at the bash level, we can’t simply access information like `does this glob contain any targets with the tag

integration

?

I guess we can just set a NO_REGEN_PEX in our in-repo "pants_from_sources" script

Currently, we check

CI=true

. As a temporary workaround, you can set that (not sure if that impacts anything else)

TRAVIS=true

it looks like