ci: Add 3.15 in CI #5518

clin1234 · 2025-10-15T15:23:02Z

No description provided.

davidhewitt · 2025-10-16T11:06:39Z

I added full build, if you want to fixup merge conflict we can see what initial results are.

.github/workflows/ci.yml

bschoenmaeckers · 2025-10-21T07:54:15Z

You should bump cpython’s max-version in pyo3-ffi/cargo.toml to 1.15.

Tpt · 2025-10-22T07:03:45Z

You should bump cpython’s max-version in pyo3-ffi/cargo.toml to 1.15.

cpython 3.15 C-API is I think still moving, wouldn't it better to keep the error message and use UNSAFE_PYO3_SKIP_VERSION_CHECK until the API is stabilized? I am a bit afraid we ship a PyO3 version that end up broken on cpython 3.15 released versions

davidhewitt · 2025-11-01T14:53:19Z

You should bump cpython’s max-version in pyo3-ffi/cargo.toml to 1.15.

cpython 3.15 C-API is I think still moving, wouldn't it better to keep the error message and use UNSAFE_PYO3_SKIP_VERSION_CHECK until the API is stabilized? I am a bit afraid we ship a PyO3 version that end up broken on cpython 3.15 released versions

Agreed, we should not declare support for 3.15 properly until 3.15 is in beta.

In #5093 (comment) I wrote some thoughts about maybe how we could support 3.15 for development.

ngoldbaum · 2026-01-08T20:35:49Z

@clin1234 looks like it's that time of year again. Do you mind if I push to this branch to fixup things?

clin1234 · 2026-01-09T01:18:50Z

Sure, considering I'm job hunting ATM

Co-authored-by: Bas Schoenmaeckers <7943856+bschoenmaeckers@users.noreply.github.com>

ngoldbaum · 2026-01-09T20:58:06Z

pyo3-ffi/src/object.rs

+// use a macro to substitute _PyObject_MIN_ALIGNMENT definition above?
+
+#[cfg_attr(not(Py_3_15), repr(C))]
+#[cfg_attr(Py_3_15, repr(C, align(4)))]


I guess I need a macro here to avoid hard-coding the value of _PyObject_MIN_ALIGNMENT in the cfg_attr macro?

Maybe not, this is rust-lang/rust#52840

It feels like it should be possible to write a static assertion here to check

e.g.

const _: () = assert!(std::mem::align_of::<PyObject>() == _PyObject_MIN_ALIGNMENT);

Good call! Just one note that it's the minimum alignment, so the assertion is >=, not ==.

ngoldbaum · 2026-01-09T22:01:22Z

It looks like CI is green here 🎉

davidhewitt

Thanks for getting this moving!

davidhewitt · 2026-01-10T14:59:41Z

pyo3-build-config/Cargo.toml

 abi3-py313 = ["abi3-py314"]
-abi3-py314 = ["abi3"]
+abi3-py314 = ["abi3-py315"]
+abi3-py315 = ["abi3"]


Is it too early to add this feature? My instinct is that we don't want users to be able to build for the 3.15 stable ABI until that's actually nailed down.

Ultimately my whole goal here is to enable testing with the new opaque PyObject stable ABI. I could wait to do that until after finishing up the wrappers for the PEP 793 module initialization API, though.

I suspect this year we might need to enable this earlier than beta 1, but I'll think about how to handle that without possibly causing downstream users to ship abi3 wheels too early.

I think we can set UNSAFE_PYO3_SKIP_VERSION_CHECK in the CI and start testing without exposing it to users.

davidhewitt · 2026-01-10T15:01:41Z

pyo3-ffi/src/object.rs

+// use a macro to substitute _PyObject_MIN_ALIGNMENT definition above?
+
+#[cfg_attr(not(Py_3_15), repr(C))]
+#[cfg_attr(Py_3_15, repr(C, align(4)))]


It feels like it should be possible to write a static assertion here to check

e.g.

const _: () = assert!(std::mem::align_of::<PyObject>() == _PyObject_MIN_ALIGNMENT);

davidhewitt · 2026-01-10T15:02:13Z

pyo3-ffi/src/object.rs

 pub type PyObjectObRefcnt = Py_ssize_t;

+const _PyGC_PREV_SHIFT: usize = 2;
+pub const _PyObject_MIN_ALIGNMENT: usize = 1 << _PyGC_PREV_SHIFT;


Is there a need for public export of the private prefixed symbol?

Suggested change

pub const _PyObject_MIN_ALIGNMENT: usize = 1 << _PyGC_PREV_SHIFT;

const _PyObject_MIN_ALIGNMENT: usize = 1 << _PyGC_PREV_SHIFT;

Since it is declared in the public API header: https://github.com/python/cpython/blob/aa8578dc54df2af9daa3353566359e602e5905cf/Include/object.h#L117, it's probably safer to make it a public export

Underscored symbols are not part of the public API, even if they're in public headers (they're present there as ABI implementation details).

We've generally been removing such exports across the whole of pyo3-ffi, e.g. #5592

pyo3-ffi/src/object.rs

davidhewitt · 2026-01-14T15:05:25Z

pyo3-ffi/build.rs

    max: PythonVersion {
        major: 3,
-        minor: 14,
+        minor: 15,


I think it'd be worth briefly discussing what the alternatives to bumping this number are. Could we make a rule that max + 1 will always build and run with warnings (maybe with an env var to turn off the warnings)?

My worry is that packages built against PyO3 0.28 might not be using 3.15 yet, but in the future when 3.15's ABI has both changed and finalised, users might install packages built with PyO3 0.28 on 3.15.0, get no warnings, and then get an instant crash.

Could we make a rule that max + 1 will always build and run with warnings (maybe with an env var to turn off the warnings)?

Let me see how painful this is to set up. Seems sensible.

By "build with warnings" are you talking about the same kind of warning we already generate when someone tries to build a stable ABI extension on the free-threaded build?

ngoldbaum · 2026-01-14T19:30:23Z

pyo3-ffi/build.rs

+                warn!(
+                    "Using experimental support for the Python {}.{} ABI. \
+                     Build artifacts may not be compatible with the final release of CPython, \
+                     so do not distribute them.",


Anyone have language tweaks here? Maybe I should also be warning that the build artifacts might lead to crashes with this python if it's ABI incompatible with the FFI definitions.

I think experimental is probably good enough hint that is might crash. Let's not block this PR any further for getting testing running, we can edit this copy later trivially.

davidhewitt · 2026-01-16T11:33:09Z

pyo3-ffi/build.rs

+                warn!(
+                    "Using experimental support for the Python {}.{} ABI. \
+                     Build artifacts may not be compatible with the final release of CPython, \
+                     so do not distribute them.",


I think experimental is probably good enough hint that is might crash. Let's not block this PR any further for getting testing running, we can edit this copy later trivially.

davidhewitt added the CI-build-full label Oct 16, 2025

bschoenmaeckers reviewed Oct 20, 2025

View reviewed changes

.github/workflows/ci.yml Outdated Show resolved Hide resolved

clin1234 force-pushed the 3.15 branch 2 times, most recently from 0397ab6 to 8b88136 Compare November 11, 2025 13:30

clin1234 force-pushed the 3.15 branch from 8b88136 to f4da3c9 Compare November 27, 2025 16:51

ngoldbaum mentioned this pull request Jan 8, 2026

Metaissue: 3.15 support #5610

Open

4 tasks

clin1234 and others added 18 commits January 9, 2026 12:42

Add 3.15 testing in ci.yml

2c1fd27

Create 5518-packaging.md

2b1bc17

Update .github/workflows/ci.yml

fdffee3

Co-authored-by: Bas Schoenmaeckers <7943856+bschoenmaeckers@users.noreply.github.com>

Update Python version compatibility in pyo3-ffi/Cargo.toml

761db77

Typo

7439eea

Add abi3-py315 in pyo3-build-config/Cargo.toml

aa4c3d4

Add abi3-py315 in Cargo.toml

57f95d1

Bump max supported CPython version in noxfile.py

cf0bcac

Bump max CPython version in pyo3-ffi/build.rs

af8eb8c

Lint noxfile.py encountered by PyLance

23f680f

Closer match of PyObject's definition in 3.15t

9662b22

tmp

331b886

fix ffi-check on the free-threaded build

78137d4

fix test failures on the free-threaded build

c919a1a

Update ABI3_MAX_MINOR

94fe4bb

revert most changes to the noxfile

79546bd

Fix free-threaded build on <3.15

d72f1a3

revert changes to moduleobject bindings

215c416

adjust ob_flags setup and fix conditional compilation

9ce3ce0

ngoldbaum reviewed Jan 9, 2026

View reviewed changes

ngoldbaum mentioned this pull request Jan 9, 2026

Supporting in-development Python versions #5093

Open

davidhewitt reviewed Jan 10, 2026

View reviewed changes

ngoldbaum added 3 commits January 12, 2026 11:13

disable abi3-py315 for now

b873ad9

rework PyObject 4-byte alignment to more closely follow upstream

43fd6c7

disable abi3-py315 for now

00441c4

ngoldbaum reviewed Jan 12, 2026

View reviewed changes

pyo3-ffi/src/object.rs Show resolved Hide resolved

ngoldbaum added 3 commits January 12, 2026 11:40

fix check-feature-powerset and test-version-limits

af4bb11

really fix test-version-limits

2a33e2d

run ruff

34bd3d5

ngoldbaum added FFI free-threading labels Jan 12, 2026

ngoldbaum added 2 commits January 12, 2026 12:11

Remove free-threaded versions from ABI3_PY_VERSIONS

4cc0aff

fix ffi-check on free-threaded build

032d532

ngoldbaum reviewed Jan 12, 2026

View reviewed changes

pyo3-ffi/src/object.rs Show resolved Hide resolved

ngoldbaum mentioned this pull request Jan 13, 2026

orjson thread safety under the free-threaded interpreter ijl/orjson#591

Closed

davidhewitt reviewed Jan 14, 2026

View reviewed changes

Allow experimental builds of alpha releases of CPython

6c4ce7f

ngoldbaum reviewed Jan 14, 2026

View reviewed changes

davidhewitt approved these changes Jan 16, 2026

View reviewed changes

davidhewitt added this pull request to the merge queue Jan 16, 2026

Merged via the queue into PyO3:main with commit 49e8772 Jan 16, 2026
92 checks passed

clin1234 deleted the 3.15 branch January 17, 2026 15:29

This was referenced Jan 18, 2026

doc unit tests failure with 3.15 #5614

Closed

test_module failure with 3.15 #5613

Closed

test_declarative_module failure with 3.15 #5612

Closed

test_module_append_to_inittab failure with 3.15 #5611

Closed

	pub const _PyObject_MIN_ALIGNMENT: usize = 1 << _PyGC_PREV_SHIFT;
	const _PyObject_MIN_ALIGNMENT: usize = 1 << _PyGC_PREV_SHIFT;

ci: Add 3.15 in CI #5518

ci: Add 3.15 in CI #5518

Conversation

clin1234 commented Oct 15, 2025

Uh oh!

davidhewitt commented Oct 16, 2025

Uh oh!

Uh oh!

bschoenmaeckers commented Oct 21, 2025

Uh oh!

Tpt commented Oct 22, 2025

Uh oh!

davidhewitt commented Nov 1, 2025

Uh oh!

ngoldbaum commented Jan 8, 2026

Uh oh!

clin1234 commented Jan 9, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ngoldbaum commented Jan 9, 2026

Uh oh!

davidhewitt left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants