Omit wheels from lockfile based on --exclude-newer

[charliermarsh]

Summary

We respect --exclude-newer during resolution, but we weren't applying it to individual files when writing the lockfile. As a result, if wheels were added to a distribution after its initial release, we'd end up including them in the lockfile, even if they were uploaded after the --exclude-newer date.

Closes #12296.

[zanieb]

Is there going to be a user-facing consequence here for people with --locked?

[charliermarsh]

No.

[charliermarsh]

There's a panic in the benchmark script setup that looks real.

[charliermarsh]

As a reminder (partly for myself), I think the reason we don't filter these out entirely (which would be way simpler) is because we want to be able to show them in error messages (i.e., we want to be able to say "There were wheels, but they were too new").

[charliermarsh]

This depends on https://pypi.org/project/methodtools/0.4.7/#files, which added a wheel on August 23, 2024. So we can bump this to avoid building it from source.

[codspeed-hq]

CodSpeed Performance Report

Merging #12299 will improve performances by 20.51%

_{Comparing charlie/exc (0964b22) with main (a95f4cf)}

Summary

⚡ 1 improvements
✅ 13 untouched benchmarks

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
⚡	resolve_warm_airflow	702.2 ms	582.7 ms	+20.51%

[zanieb]

The benchmark change is pretty funny. I'll review this tomorrow, thanks for the quick turnaround.

[zanieb]

(oops)

[konstin]

I can reproduce a speedup in the right range by bumping the cutoff date:

Benchmark 1: uv pip compile --exclude-newer 2024-08-08 a.txt
  Time (mean ± σ):      92.2 ms ±   2.2 ms    [User: 100.6 ms, System: 108.0 ms]
  Range (min … max):    87.9 ms …  96.7 ms    32 runs
 
Benchmark 2: uv pip compile --exclude-newer 2024-09-01 a.txt
  Time (mean ± σ):      80.5 ms ±   4.5 ms    [User: 83.2 ms, System: 102.8 ms]
  Range (min … max):    74.2 ms …  97.3 ms    39 runs

[Gankra]

Is there a reason to not consider all Incompatible() instances "excluded"?