Ergonomic, hermetic, user-friendly software build system for Python, Go, Java, Kotlin, Protobuf, Scala, Shell. Pants lets you fearlessly scale up your codebase!

Pants

<@U053J9JB4> I found the python chroot caching bug, I'll file an issue and get an RB up in an hour or so.

That makes me happy... because I am still failing to recreate this in a reasonably small example

<@U083H6LRL>: there is work underway to add more support for javascript (and css)... <@U071MJHQD> has an initial patch out <https://rbcommons.com/s/twitter/r/2542/>

neat, i can at least base our internal approach on that. we use webpack for our actual build so we'd probably just have one monolithic target for that whole bundle, but the node stuff is relevant

webpack or browserify are likely on the way as well.

will try to have a design doc out this week.

<@U053J9JB4>: cached chroots are supposed to be keyed off all sources and deps, so if you change a source it's not supposed to re-use the cached chroot. if it does then that's a bug.

But they nominally cancel, so its hard to figure out a repro

In short, chroots fingerprint is not transistive over the input targets

...but the chroots use default PEXBuilder linking, so they should cancel on same FS if the FS allows hard links

So I've got fixes for those and manual testing shows they do the right thing, new chroot for transitive source dep change, but a repro of the original issue is still niggling me

OH - sometimes talking helps - <@U053J9JB4> and I have been chatting offline, but this makes me think of things breaking links - <@U053J9JB4> is there any possibility of source delete - re-create for the failing cases

Say you had a custom task that downloaded python sources into the source tree such that they are owned by a target in the chroot closure

My mind warps in this way because this was done at Twitter for thrift files for example

Well, if anyone figures out Pauls failure mode in current pants - many Cokes, but I can fix these bugs anyhow

adding a new source file would reproduce the bug, no? there would be no hard link to it

what’s the cleanest way to initialize a subsystem for a test? I see how test_ivy_subsystem.py does it, but seems a little involved. is there an easier way?

I have an RB with a new utility `create_subsystem(type, **options)`

If you can wait maybe an hour, I should have Paul's fix out and then can get that branch out

Until then, you can instead try something like this:

<https://github.com/pantsbuild/pants/blob/master/tests/python/pants_test/backend/python/test_python_chroot.py#L59-59>