GeezerVison

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

_{Comment @coderabbitai help to get the list of available commands.}

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

@copilot make a proper patch please

@youknowone The model claude-haiku-5.5 is not available for your account. This can happen if the model was disabled by your organization's policy or if your Copilot plan doesn't include access to it.

You can try again without specifying a model (just @copilot) to use the default, or choose a different model from the model picker.

If you want to contact GitHub about this error, please mention the following identifier so they can better serve you: 46d66e99-fa1b-4c4a-98cc-3cf9e6d541e1

@dependabot rebase

Merging this PR will not alter performance

✅ 62 untouched benchmarks
⏩ 4 skipped benchmarks¹

_{Comparing dependabot/cargo/libc-0.2.190 (53c45ac) with main (1be1628)²}

@codspeedbot explain performance differences

@codspeedbot explain performance differences

Neither change comes from the libc bump. Both benchmarks are noisy, so this PR can be merged and the loop_append.py regression acknowledged.

The PR only touches Cargo.lock (libc 0.2.189 → 0.2.190). For x86_64 Linux, that release adds bindings and Default impls and changes nothing the interpreter runs.

The report also uses an older base. CodSpeed had no result for main at 2a09ee0, so it compared against b429155. That base doesn't include #9105 (the CPython 3.15 bytecode/VM change), but this branch does. A better baseline is main at 1269277: it has the same VM code as this PR, just with an indexmap bump instead of libc.

⚡ cmp.py +11% (69.6 ms → 62.5 ms)

The base was the outlier. cmp.py usually measures about 62.5–65 ms on main, but b429155 measured 69.6 ms because of extra time in Dict::_get_inner. This PR's 62.5 ms is normal. Compared with 1269277, cmp.py doesn't change.

❌ loop_append.py −11.7% (2.3 ms → 2.7 ms)

Compared with 1269277, which has the same VM code, the flamegraphs show:

The interpreter loop, int allocation, list.append and the libc/malloc functions cost about the same in both runs. All of the extra time is in the module globals dict lookups for i and obj (the loop runs at module level, so these are LOAD_NAME/STORE_NAME). The instruction count in those two dict functions roughly doubles, which points to longer hash probe chains, not slower code.

The cause is the benchmark setup. benches/microbenchmarks.rs uses Settings::default(), which leaves hash_seed = None, so every benchmark process picks a random string hash secret. Depending on the seed, the names in the globals dict collide more or less, and benchmarks like loop_append.py, loop_string.py, strings.py and cmp.py move by 10–17% between runs. The same thing has shown up on several earlier PRs.

Possible fix

Setting a fixed seed in the microbenchmark harness (settings.hash_seed = Some(0)) would stop these jumps, so results would only change when the code does.

Comment @codspeedbot pin the hash seed in the microbenchmark harness and I'll open a PR for it.

_{Learnings updated. See what CodSpeed remembers about this repository.}

_{Was this response helpful? React with 👍 or 👎 to help us improve CodSpeedBot.}

@codspeedbot pin the hash seed in the microbenchmark harness

@codspeedbot pin the hash seed in the microbenchmark harness

I opened #9126 to pin the hash seed: #9126

It's a separate PR so this dependabot branch only contains the libc bump.

Change: one line in benches/microbenchmarks.rs. bench_rustpython_code now sets settings.hash_seed = Some(0), so every microbenchmark run uses the same string hash secret. Before, the secret was random for each process, which made dict-heavy benchmarks like loop_append.py, cmp.py, loop_string.py and strings.py move by 10–17% between runs with no code change.

Not tested yet: I only checked formatting. The full benchmark build didn't finish in my environment, so I haven't compiled or run the change. The CodSpeed CI run on #9126 will be the first test.

What to expect: that first CI run may still move those benchmarks once, because the fixed seed can collide differently than the random seed in the base run. After it merges, these benchmarks should only change when the code does.

benches/execution.rs sets up its interpreter the same way, but I left it unchanged.

Comment @codspeedbot pin the hash seed in benches/execution.rs too and I'll add it to #9126.

_{Was this response helpful? React with 👍 or 👎 to help us improve CodSpeedBot.}

@youknowone Clippy failure is from #9152.

TODO(me): consolidate the unix and windows dependency groups in .github/dependabot.yml into an os dependency group, adding libc and rustix to that group.

CI failure for aarch64-apple-ios is due to nix-rust/nix#2830, which was fixed by nix-rust/nix#2821.

@SnoozeThis https://crates.io/crates/nix

I would wait until https://crates.io/crates/nix has a new version and then add a comment. But an error occurred while snoozing: crate nix not found on crates.io

Superseded by #9240.