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

Pants

...exact same `filedeps` with/without `MODE=debug`

```F401:ERROR   src/python/pants/engine/scheduler.py:010 'threading' imported but unused
     |import threading```

(... the thing above with the test count was a test caching thing...)

```./pants --tag='-integration' test --cache-ignore tests/python/pants_test/engine:

MODE=debug
  ==== 178 passed, 22 skipped in 43.02 seconds =====
MODE=release
  ==== 178 passed, 22 skipped in 12.49 seconds =====```

It’s actually a little upsetting that the scheduler has that much overhead without optimisation :S

not sure it's overhead... those tests are all tests that actually invoke the scheduler to do things

I thought it was just some random tests, in which case an extra 31 seconds is scary

and in general, so are the rest of pants' tests (integration certainly).... so yea, CI should definitely be using release builds

...big caveat: most of the unit tests aren't yet =(  ... <https://github.com/pantsbuild/pants/pull/4867>

need to pick that back up soon. broken window.

<@U6YPB4SJX>: since you have yet to escape: the "Under graph lock." bit here <https://github.com/pantsbuild/pants/pull/5178> refers to the fact that if we received an EntryId under a (read only) graph lock, it should be guaranteed to be valid

i actually started diving into making that a type-level check (which would be hella cool, and maybe i still should), but for now am thinking of just having a `unsafe_entry_for_id` or similar to do the unwrap with a useful error message

...yea, maybe i just do that now, since it's icing time.

Sounds good :slightly_smiling_face: unsafe_entry_for_id works for me, as long as the string in its expect is somewhat verbose :slightly_smiling_face:

yea... basically, i think what'd you do is newtype up a `SafeEntryId` wrapper, with a lifetime tied to a `ReadLockGuard`, such that a SafeEntryId is always valid

(...and switch the `Mutex` to a `ReadWriteLock`, which i've done at various points without seeing any performance difference)

I would happily review that code :slightly_smiling_face: