Hi, I recently came across a problem in Bazel where you have a rule that outputs a directory of randomly named files instead of a few known files. This means you can't add it to the build graph because that only deals in files, not directories. The currently recommended workaround is to zip/unzip the directory before/after running the rule. Not idea. I was wondering if Pants handles this any better?

Mmmm you'd have to be more specific on the problem at hand to get a complete answer. Pants certainly can handle various things at the directory (and file) level, but depending on the situation your ability to cache results of operations may diminish

Well in Bazel when you define a rule you'd like to say something like this:

ctx.actions.run(
        ...
        outputs = [work_dir],
    )

Where

work_dir

is a directory that gets a load of randomly named files written to it. But it doesn't really work because the build graph operates on files with names that are all known at analysis time.

So Pants defintely supports directory-as-a-unit both at the BUILD-level (

experimental_shell_command

comes to mind) and at the Plugin level (using a

Digest

). Depending on where/how the directory is captured is the razor for caching. So in that regard YMMV

Ah ok cool that's good to know!