release: 0.29.1 - #6151
Draft
davidhewitt wants to merge 43 commits into
Draft
Conversation
Signed-off-by: person93 <person93.person93@gmail.com>
…_MEMBERS` fail with address sanitizer (#6145) * PyHeapType_GET_MEMBERS: cast etype to *mut u8 * PyObject_GET_WEAKREFS_LISTPTR: cast o to *mut u8 * Add newsfragement entry * fix newsfragement entry * Use byte_offset instead of casting * PyObject_GET_WEAKREFS_LISTPTR: link from CPython
* fix typos & add spell check * add typos to dependency-groups * uvx ruff format .
* fix: compile with PyPy when nightly feature is enabled * add changelog * add check-nightly * Update .github/workflows/ci.yml --------- Co-authored-by: David Hewitt <mail@davidhewitt.dev>
Bumps [actions/checkout](https://github.com/actions/checkout) from 6.0.3 to 7.0.0. - [Release notes](https://github.com/actions/checkout/releases) - [Commits](actions/checkout@v6.0.3...v7) --- updated-dependencies: - dependency-name: actions/checkout dependency-version: 7.0.0 dependency-type: direct:production update-type: version-update:semver-major ... Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
* ci: update / clean up wasm build * drop `--quiet` * fix CC variable conflict * switch wasm jobs to uv * debug * fix `test-wasm` path
Bumps [actions/cache](https://github.com/actions/cache) from 5 to 6. - [Release notes](https://github.com/actions/cache/releases) - [Changelog](https://github.com/actions/cache/blob/main/RELEASES.md) - [Commits](actions/cache@v5...v6) --- updated-dependencies: - dependency-name: actions/cache dependency-version: '6' dependency-type: direct:production update-type: version-update:semver-major ... Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Updates the requirements on [codspeed-criterion-compat](https://github.com/CodSpeedHQ/codspeed-rust) to permit the latest version. - [Release notes](https://github.com/CodSpeedHQ/codspeed-rust/releases) - [Commits](CodSpeedHQ/codspeed-rust@v4.0.0...v5.0.1) --- updated-dependencies: - dependency-name: codspeed-criterion-compat dependency-version: 5.0.1 dependency-type: direct:production ... Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Update nox command for generating lcov file for vscode plugin for coverage visualization.
* fix lifetime of return value in `PyClassGuardMutSuper::as_super` * newsfragment
Merging this PR will improve performance by 16.51%
|
| Mode | Benchmark | BASE |
HEAD |
Efficiency | |
|---|---|---|---|---|---|
| ⚡ | Simulation | extract_u128_small |
1,171.7 ns | 986.1 ns | +18.82% |
| ⚡ | Simulation | extract_u128_zero |
1.2 µs | 1 µs | +18.28% |
| ⚡ | Simulation | extract_i128_small_pos |
1.2 µs | 1 µs | +17.77% |
| ⚡ | Simulation | extract_i128_zero |
1.3 µs | 1.1 µs | +17.28% |
| ⚡ | Simulation | extract_u128_u32_max |
1.3 µs | 1.1 µs | +17.05% |
| ⚡ | Simulation | extract_i128_small_neg |
1.3 µs | 1.1 µs | +16.75% |
| ⚡ | Simulation | extract_u128_u64_max |
1.4 µs | 1.2 µs | +15.46% |
| ⚡ | Simulation | extract_u128_max |
1.4 µs | 1.2 µs | +15.07% |
| ⚡ | Simulation | extract_i128_pos_max |
1.5 µs | 1.3 µs | +14.38% |
| ⚡ | Simulation | extract_i128_neg_min |
1.5 µs | 1.3 µs | +14.35% |
Tip
Curious why this is faster? Comment @codspeedbot explain why this is faster on this PR, or directly use the CodSpeed MCP with your agent.
Comparing release-0.29.1 (8663a9a) with release-0.29 (646a38d)1
Footnotes
Contributor
|
I just opened #6192. The first commit there should be backported here. |
Member
|
Could you also include the fix in #6154? |
* update `PySet_GET_SIZE` and similar for free-threaded Python * split `cpython` portion of `setobject.h` off
* Fix #[pyclass(dict)] leak on Python 3.11 and 3.12 * apply review suggestion * Don't use sys.getallocatedblocks in test * add release note * fix clippy * apply David's suggestions * Apply David's suggestion with cfg_select
* Fix compilation with GraalPy 3.13 * Exclude mypy from optional-dependencies on GraalPy * Add GraalPy to check-all
…6252) Nightly rustdoc's `rustdoc::invalid-html-tags` lint gained an "improperly nested Markdown paragraph" check, which broke the netlify-build CI job (`nox -s check-guide` runs `cargo doc` with `-Dwarnings`). The warning box on `PyModule::from_code` contained a blank doc line before `</pre></div>`. A blank line terminates a CommonMark HTML block, so the closing tags landed in a separate block from the tags they close. Keep the markup on contiguous doc lines so it forms a single HTML block. This also drops a stray `//` (rather than `///`) line, which was silently omitting content from the rendered docs. Claude-Session: https://claude.ai/code/session_0176PdmBb2j8euHTFEv4huzz Co-authored-by: Claude <noreply@anthropic.com>
Contributor
|
I just pushed backports for all the merged PRs listed in the description. I backported a few more PRs than the ones listed to get CI passing. I also had to do manual fixes for Python 3.8 compat and pinning ruff. |
… is `cfg`-ed out (#6255) The `,` separating two elements of an introspection list was written unconditionally as soon as the element was not the first one, even when the elements before it were all behind a `#[cfg]` that is disabled. A `#[pymodule]` whose first member is `cfg`-ed out therefore serialized its member list as `[,"..."]`, and since `find_introspection_chunks_in_*` propagates the parse error, one such module makes `pyo3-introspection` fail to read the *whole* extension — no stubs at all. This is reachable in `pyo3_pytests` as it ships: `pyfunctions` has `#[cfg(feature = "experimental-async")] use super::with_async;` as its first member, so building it with `experimental-inspect` but without `experimental-async` produces a binary `pyo3-introspection` refuses: Failed to parse introspection chunk: "{...,\"members\":[,\"134042... `nox -s test-introspection` never catches it because it always enables both features together. The separator is now gated behind an `any(..)` of the `cfg`s of all the elements before the one it precedes, on top of that element's own. The two cases that look like they need special handling fall out of `cfg` syntax instead: `any()` is false, so the first element gets no separator, and `all()` is true, so an element that is always compiled in makes every later separator unconditional too. Verified in both the `abi3` and non-`abi3` configurations, for a `cfg`-ed out first member, middle member and last member. `experimental-inspect`: cover `cfg`-ed out `#[pymodule]` members in `pytests` `nox -s test-introspection` builds `pytests` with `experimental-async` and `experimental-inspect` together, so the only `cfg`-ed out member the crate has as it ships is always compiled in and no CI run ever exercised a `#[pymodule]` member list with a hole in it -- which is how the invalid JSON fixed two commits ago went unnoticed. `pyfunctions` gains a first and a last member behind `#[cfg(any())]`, which is never true. This changes neither the module at runtime nor the checked-in stubs, but it makes every introspection run walk the path where an element is removed, and a regression there fails loudly: `pyo3-introspection` refuses the whole extension and the session never reaches the stub comparison. Rename
* fix: register in-place sequence operations in distinct slots * docs: add newsfragment for PR 6260
…g an instance of an extension type (#6224) * pycell: decref the reference to the type held by the instance when deallocating the instance * add newsfragment * fix type being decrefed and add tests * check type is heap type before forcing decref and use the type system to encode the decref of the type, * remove useless import * address last comment * clippy fix * proper gating to make clippy happy * improve release notes * remove separate type decref for types with free list * disable reference based count tests on free threaded builds * cfg out macro helper used only in test that are cfg out on free threaded build
* fix: return error for oversized time durations * docs: add time duration overflow news fragment
…-endian (#6150) * fix 3.15+ `append_to_inittab` on platforms which aren't 64-bit little-endian * fix test conditional code * fix missing import * newsfragment * implement legacy slots in terms of new slots * don't build module def slots on abi3t * promote constants to statics * fix test
…Generator and ConvertField (#6277) * Remove IMPLEMENTS_INTOPYOBJECT from PyClassGetterGenerator and ConvertField Unused, enables to remove IsIntoPyObject traity * Fix UI tests
…d magic methods (#6239) * fix: emit positional-only arguments for slot-backed magic methods `experimental-inspect` emitted the arguments of magic methods as positional-or-keyword, but the slot wrappers CPython installs for type slots reject keyword arguments entirely, e.g. `__eq__` is exposed as `($self, value, /)`. `mypy.stubtest` therefore reported every such method as inconsistent with the runtime. `PythonSignature` already distinguishes positional-only, positional-or- keyword and keyword-only parameters, so record the truth when the signature is built rather than teaching introspection a new mode: set `positional_only_parameters`, the same assignment a trailing `/` in `#[pyo3(signature = ...)]` performs, now factored into `make_all_parameters_positional_only`. Introspection is untouched. `SlotDef::arguments_are_positional_only` is the single source of truth. It excludes the `TpNew`/`TpInit` calling conventions, since `__new__` and `__init__` are called with `args`/`kwargs` and keep the signature declared in Rust. `__call__` behaves the same way, so this ends up being exactly the set of magic methods for which `#[pyo3(signature = ...)]` is allowed. Writing to the spec's signature is inert for codegen: `impl_arg_params` is reached from the slot path only for `TpNew`/`TpInit`, and `__text_signature__` is rejected on every magic method but `__new__`. Runtime text signatures are unchanged. The parameter of the comparison dunders synthesized by `#[pyclass(eq)]` and `#[pyclass(ord)]` is also renamed from `other` to `value` to match the name CPython uses in the `tp_richcompare` slot wrappers. Closes #6235 Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * Proper newsfragments name * Address review comments --------- Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
…ame (#6254) * `experimental-inspect`: introspect `#[pyfunction]` under its Python name `impl_wrap_pyfunction` passed `func.sig.ident`, the Rust identifier, as the function name in the introspection data. `#[pyfunction(name = "x")]` was therefore declared in the stubs under a name the module does not have at runtime: #[pyfunction(name = "solve")] fn rust_solve() -> usize { 42 } generated `def rust_solve() -> int: ...` for a module exporting `solve`. `pytests` gains a renamed `#[pyfunction]` so the checked-in stubs cover it. `experimental-inspect`: pin the renamed `#[pyfunction]` to its Python name The preceding commit checks the name a renamed `#[pyfunction]` is declared under in the checked-in stubs, but nothing checks that this is the name the module actually exports -- which is the invariant the bug broke. Assert both from Python. Also drop that commit's mention of `__all__`: the stub generator never emits one, so only the declaration has to use the Python name. `experimental-inspect`: drop the newsfragment suffix The `.1` suffix distinguished this fragment from the one of the `cfg` separator fix, which now lives on its own branch. Both still need renumbering to their real pull request. Rename * Rename
… indentation (#6270) * fix(inspect): don't pad blank docstring lines with the body indentation `class_stubs` and `function_stubs` re-indent each docstring line into the body by writing the indentation and then the line. For an empty line — a paragraph break, which every non-trivial docstring has — that leaves a line consisting only of spaces. A 1200-line stub generated for a real project came out with 95 of them, all flagged `W293` by ruff and flake8. It is not fixable downstream: the file is generated, so any hand edit is reverted by the next run, and the formatter's autofix is classified unsafe inside a string literal. The three nested-indentation sites used `str::replace('\\n', "\\n ")`, which has the same problem for any nested element whose body contains a blank line. Both now go through `push_indented`/`push_docstring`, which leave an empty line empty. Indentation of non-empty lines is unchanged, so the checked-in `pytests/stubs/*.pyi` are untouched. * Rename
* pyo3-introspection Fixes wheel publishing - The artifact directory was wrong during publishing - Validate wheels after build - Remove license files that were not properly included in source wheels (the `License: MIT OR Apache-2.0` metadata line is still there) * Properly install uvx --------- Co-authored-by: David Hewitt <mail@davidhewitt.dev>
* add an AI contributions policy * drop extra word * typo * reword to be more direct and forbid use of AI on easy issues * signpost to contributing guide on pull request template Co-authored-by: Bruno Kolenbrander <59372212+mejrs@users.noreply.github.com> --------- Co-authored-by: Bruno Kolenbrander <59372212+mejrs@users.noreply.github.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
18 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Backports for the 0.29.1 release.
Before I prep the release to ship:
PyTuple::newon wrongsizehintwhen compiling for RustPython #6154PySet_GET_SIZEand similar for free-threaded Python #6230#[pyo3(get)]can fail to compile some types withexperimental-inspect#6233tp_traverseto traverse type object #6269experimental-inspect: introspect#[pyfunction]under its Python name #6254experimental-inspect: don't pad blank docstring lines with the body indentation #6270experimental-inspect: emit positional-only arguments for slot-backed magic methods #6239