Fix/Add check_output_timestamps to PythonEnvironment

[Luc-Mcgrady]

restat defaults to 1 (Though the docs don't make it seem that way?) meaning that .check_output_timestamps() doesn't do anything.

This, for me, fixes the python environment rebuilding every run.

[iamllama]

I can't seem to reproduce this, n2 (from ./tools/install-n2) reports no work to do after a second ./run

[Luc-Mcgrady]

Here is the difference this pr makes for me:

Peek.2025-06-21.23-53.mp4

Is there something to do with ./tools/install-n2 that I'm meant to do to avoid this behaviour?

On main this is the output of ./ninja -d explain -v

    Finished `release` profile [optimized] target(s) in 0.15s
ninja explain: output /home/luc/Programming/forks/anki/out/pyenv/bin/black older than most recent input /home/luc/Programming/forks/anki/out/rust/release/runner (1750527715838813748 vs 1750546233608145463)
ninja explain: /home/luc/Programming/forks/anki/out/pyenv/bin/protoc-gen-mypy is dirty
ninja explain: /home/luc/Programming/forks/anki/out/pylib/anki/ankidroid_pb2.py is dirt
ninja explain: /home/luc/Programming/forks/anki/out/pylib/anki/ankidroid_pb2.pyi is dirty
...

[iamllama]

Is there something to do with ./tools/install-n2 that I'm meant to do to avoid this behaviour?

n2 is intended to be used instead of ninja (which doesn't support the hide_success rule var that ninja_gen emits, for example), and is installed via that script

[Luc-Mcgrady]

using n2 fixes this for me but It's still worth fixing for ninja users right?

[iamllama]

Dug into this a bit (discussed over discord), and from looking at ninja's source it seems to check for the restat rule var's presence (not its value) as the docs state
https://github.com/ninja-build/ninja/blob/656412538b6fc102b809a61e0efce422e5a20534/src/graph.cc#L514
https://github.com/ninja-build/ninja/blob/656412538b6fc102b809a61e0efce422e5a20534/src/build.cc#L912

Which means that every statement ends up getting restat'ed, which fixes the issue but not in an expected way 🤔

[dae]

The build scripts try to support both n2 and ninja. @iamllama I'm not sure I'm following your comment - I thought you meant we were doing restat = 0 (which wouldn't work), but the code appears to be only conditionally writing it:

    if action.check_output_timestamps() {
        stmt.rule_variables.push(("restat".into(), "1".into()));
    }

Could you clarify?

[dae]

Thanks Luc!

[iamllama]

I thought you meant we were doing restat = 0 (which wouldn't work), but the code appears to be only conditionally writing it:

@dae I was referring to the pr's original impl ace2e5e which would have always defined restat

[dae]

Ah, gotcha. I was familiar with this issue, as I made the same mistake with assuming 0 = off with n2 in the past. 😅