Skip repeated directories in PATH when searching for Python interpreters

[zanieb]

Closes #12302

The change is visible in this commit.

[zanieb]

Indented, best viewed with https://github.com/astral-sh/uv/pull/12367/files?diff=unified&w=1

[charliermarsh]

Should this be implemented by de-duping the interpreters rather than the parent directories? I assume not, but it'd be good to note why.

[zanieb]

Should this be implemented by de-duping the interpreters rather than the parent directories?

We probably could but I'm not sure what it gets us. We'd spend more time collecting interpreters we've already seen. We could dedupe interpreter paths instead of parent directories but that also just adds redundant directory item enumeration.

[charliermarsh]

We could dedupe interpreter paths instead of parent directories but that also just adds redundant directory item enumeration.

I think that cost of that will be really low compared to the cost of querying the interpreters. But is it even better? Do we want to show symlinks of adjacent Python interpreters (like python vs python3 in the same directory)? I can't rememeber.

[zanieb]

I think that cost of that will be really low compared to the cost of querying the interpreters.

I agree. I don't see why it'd be preferable though.

But is it even better? Do we want to show symlinks of adjacent Python interpreters (like python vs python3 in the same directory)?

Yes, I think it's important we can represent this. There are a couple reasons off the top of my head...

• It's meaningful to show that python and python3 are both available on the PATH and invoke the same interpreter
• The name of the executable is relevant for some downstream logic around "default" versions, i.e., have you opted into free-threaded Python by installing it as python3?

We could deduplicate those later, e.g., during the uv python list display (we already do some deduplication there). However, during discovery it's important to avoid filtering too aggressively.

[codspeed-hq]

CodSpeed Performance Report

Merging #12367 will degrade performances by 10.19%

_{Comparing zb/python-seen (ef6a569) with main (be3d5df)}

Summary

❌ 1 regressions
✅ 13 untouched benchmarks

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
❌	wheelname_tag_compatibility[flyte-short-incompatible]	771.4 ns	858.9 ns	-10.19%