Two interesting things this week: • I just wrote a script to validate our Bazel metadata is correct (code doesn't rely on indirect deps) by using Pants' metadata. Found ~385 targets missing ~1k dependencies total. • That means my previous estimation of 18k BUILD lines was off by about 1k lines 😂 This may seem like some "fun" numbers for trivia. But consider someone refactoring code and removing a dependency from their Python file and their metadata. If some Python file indirectly relied on that dep, now that file's tests fail. E.g. refactoring

a.py

makes `f.py`'s tests fail. If you're the refactorer you'll be wasting your time trying to figure out why `f.py`'s tests are suddenly failing on a missing import.

A key insight for why we are so into dep inference. Maintaining manual metadata is hard to get right!

This insight sounds like it should be a blog …

I agree. @bitter-ability-32190 I'd love to have you write about this for the blog. It's a very good point, and a gotcha that is all too common.

🙈

No rush if you've got too much to think about already. Want me to circle back to the idea in N weeks?

I tried figuring out how much sauce I could actually publish and didn't get anywhere when navigating the behemoth. Yeah circle back later