Add float type caster and revert type hint changes to int_ and float_ #5839

gentlegiantJGC · 2025-09-16T07:55:40Z

Description

The casting rules and type hints for py::float_ and py::int_ should match the python type hint conventions.
An input typed with int can only accept int types and float can accept float or int types as defined in as per PEP 484.
when an argument is annotated as having type float, an argument of type int is acceptable

Expected behaviour

Class	can cast from	type hint
py::int_	int	int
py::float_	float or int	float

Current behavour

Class	can cast from	type hint
py::int_	int	typing.SupportsInt
py::float_	float	typing.SupportsFloat

The type hints were changed in #5540 by @InvincibleRMC but the classes do not support these casting rules.

m.def("test_int", [](py::int_ i) { return i; });
m.def("test_float", [](py::float_ f) { return f; });

test_int(5.5)

Traceback (most recent call last):
  File "<input>", line 1, in <module>
TypeError: test_int(): incompatible function arguments. The following argument types are supported:
    1. (arg0: typing.SupportsInt) -> int
Invoked with: 5.5

test_float(5)

Traceback (most recent call last):
  File "<input>", line 1, in <module>
TypeError: test_float(): incompatible function arguments. The following argument types are supported:
    1. (arg0: typing.SupportsFloat) -> float
Invoked with: 5

Fixes #5840

Suggested changelog entry:

Fix py::float_ casting and py::int_ and py::float_ type hints

These two types do not support casting from int-like and float-like types.

gentlegiantJGC · 2025-09-16T08:07:21Z

I feel like py::float_ should handle automatically casting from a python int as per PEP 484
when an argument is annotated as having type float, an argument of type int is acceptable
But that should be another pull request.

Edit: After debugging this a little py::float_ can be constructed from a py::int_ but the type caster will only cast if the type is an instance of float

The default py::object caster only works if the object is an instance of the type. py::float_ should accept python int objects as well as float. This caster will pass through float as usual and cast int to float. The caster handles the type name so the custom one is not required.

gentlegiantJGC · 2025-09-16T15:48:23Z

I don't know what has happend with those tests but I feel it is beyond the scope of the changes I have made.

InvincibleRMC · 2025-09-16T16:45:15Z

Interesting py::float_ doesn't convert the same as float. Nice fix!

rwgk

Looks good to me. @InvincibleRMC could you please also review & formally approve? I'll merge this PR after I see your approval.

rwgk · 2025-09-26T15:14:42Z

tests/test_pytypes.cpp

    m.def("get_tuple_from_iterable", [](const py::iterable &iter) { return py::tuple(iter); });
    // test_float
    m.def("get_float", [] { return py::float_(0.0f); });
+    m.def("cast_float", [](py::float_ f) { return f; });


Could you please rename this to roundtrip_float or float_roundtrip or similar?

(cast is super ambiguous in the context pybind11; C/C++ cast, to-Python, from-Python, all in the same name bucket unfortunately)

gentlegiantJGC · 2025-09-26T15:21:21Z

@rwgk do you have any idea what is causing the tests to fail?
I can't see what I have changed that would cause the segfault

These are the failing tests
macos-latest, graalpy-24.2 https://github.com/pybind/pybind11/actions/runs/18040422028/job/51337588233
ubuntu-latest, graalpy-24.1 https://github.com/pybind/pybind11/actions/runs/18040422028/job/51337588534

rwgk · 2025-09-26T15:33:22Z

@msimacek Could you please take a look at these graalpy failures? We believe they are unrelated to this PR.

https://github.com/pybind/pybind11/actions/runs/17764708439/job/50485236713

InvincibleRMC

Looks good to me. Nice fix.

rwgk · 2025-09-26T17:25:38Z

I'm using PR #3939 to see if current master has the same issue with the two graalpy tests.

gentlegiantJGC · 2025-09-26T17:49:38Z

I'm using PR #3939 to see if current master has the same issue with the two graalpy tests.

Looks like it only ran half the tests and not the ones that failed here.

rwgk · 2025-09-26T18:36:24Z

I'm using PR #3939 to see if current master has the same issue with the two graalpy tests.

Looks like it only ran half the tests and not the ones that failed here.

Yeah :-(
I forgot about that. I converted to Ready for review to trick it into running everything.

rwgk · 2025-09-26T18:38:37Z

I just manually cancelled the workflow, to not block those two runner slots for several more hours until the tests hit the timeout.

rwgk · 2025-09-26T19:29:50Z

CI passes under #3939, but I'm a bit confused at the minute TBH:

3939 — All checks have passed 3 skipped, 76 successful checks
Here — Some checks haven't completed yet, 2 cancelled, 2 skipped, 81 successful checks

So that's 79 total vs 85

I need to look more carefully, probably I'll get a chance only tomorrow or so. If there is anything you see, please let me know.

@henryiii JIC that's something obvious to you

gentlegiantJGC · 2025-09-26T19:47:49Z

The tests that failed here have passed in your PR. It must be something I changed but I have no idea what it is.

The output is the same up to this line

 ../../tests/test_pytypes.py::test_surrogate_pairs_unicode_error[str_from_std_string_input] XFAIL [ 76%]

This is the first test that did not run

../../tests/test_pytypes.py::test_bytes PASSED                           [ 76%]

Are your tests run sequentially? Is this where it is failing?

The only potential break I can see is the removal of handle_type_name for float_.
I thought it was not needed with it added to the custom type_caster but perhaps something needs it?

gentlegiantJGC · 2025-09-26T19:54:07Z

I will try running the tests tomorrow in my branch and see where it breaks

rwgk · 2025-09-26T20:22:38Z

Are your tests run sequentially?

Sorry I'm not sure I understand the question. I did not run anything locally, or change anything in PR #3939, except changing one character in the main README. I.e. all tests should run identically to your PR. But then again, there is the difference in the total number of tests that's still a mystery to me.

rwgk · 2025-09-26T20:35:51Z

Trying another super quick and easy thing: PR #5850

Just started. But now I see already that the total number of tests is 85, like here.

rwgk · 2025-09-26T21:58:44Z

All 83 tests pass under #5850: it seems very certain now that there is a problem with this PR, unfortunately, or maybe this PR is triggering a latent bug in graalpy.

@gentlegiantJGC I think it could save you time (potentially a lot) if you wait until we hear back from @msimacek.

gentlegiantJGC · 2025-09-27T15:15:58Z

It was a reference counting mistake I had made.
I have fixed that and simplified the implementation a bit.

Another test is now stuck indefinetly. No error for this one though.

You may want to add a more sensible maximum time limit for workflows.
The default of 6 hours is a bit high.
https://docs.github.com/en/actions/reference/workflows-and-actions/workflow-syntax#jobsjob_idtimeout-minutes

rwgk · 2025-09-27T16:02:36Z

Another test is now stuck indefinetly. No error for this one though.

I think I have seen that one a few times before, most likely it's unrelated to this PR. I just cancelled the job and then triggered a rerun.

You may want to add a more sensible maximum time limit for workflows. The default of 6 hours is a bit high. https://docs.github.com/en/actions/reference/workflows-and-actions/workflow-syntax#jobsjob_idtimeout-minutes

It'd be awesome if you could send a PR for that. 90 minutes will probably never time out unless there is an actual bug.

rwgk · 2025-09-27T16:12:56Z

It was a reference counting mistake I had made.

Ah, I missed that there was a reinterpret_steal before. Interesting that it didn't crash on any other platforms.

Revert type hint changes to int_ and float_

60e8528

These two types do not support casting from int-like and float-like types.

Fix tests

b8a619d

gentlegiantJGC marked this pull request as ready for review September 16, 2025 09:08

gentlegiantJGC mentioned this pull request Sep 16, 2025

[BUG]: SupportsInt and SupportsFloat changes cause type checking failures on python classes that override these functions #5767

Open

3 tasks

gentlegiantJGC and others added 2 commits September 16, 2025 11:22

style: pre-commit fixes

4bab567

gentlegiantJGC changed the title ~~Revert type hint changes to int_ and float_~~ Add float type caster and revert type hint changes to int_ and float_ Sep 16, 2025

gentlegiantJGC added 3 commits September 16, 2025 11:41

Fix name

35948ce

Fix variable

a8a4d00

Try satisfying the formatter

4b694ec

Merge branch 'master' into fix-int-float-type-hint

eaa18f1

rwgk reviewed Sep 26, 2025

View reviewed changes

Rename test function

53ca8a9

InvincibleRMC approved these changes Sep 26, 2025

View reviewed changes

Simplify type caster

46e9e74

Fix reference counting issue

50e81a9

rwgk approved these changes Sep 27, 2025

View reviewed changes

rwgk merged commit 8ed0dab into pybind:master Sep 27, 2025
147 of 148 checks passed

github-actions bot added the needs changelog Possibly needs a changelog entry label Sep 27, 2025

gentlegiantJGC deleted the fix-int-float-type-hint branch September 27, 2025 16:56

Add float type caster and revert type hint changes to int_ and float_ #5839

Add float type caster and revert type hint changes to int_ and float_ #5839

Conversation

gentlegiantJGC commented Sep 16, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Suggested changelog entry:

Uh oh!

gentlegiantJGC commented Sep 16, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

gentlegiantJGC commented Sep 16, 2025

Uh oh!

InvincibleRMC commented Sep 16, 2025

Uh oh!

rwgk left a comment

Choose a reason for hiding this comment

Uh oh!

rwgk Sep 26, 2025

Choose a reason for hiding this comment

Uh oh!

gentlegiantJGC Sep 26, 2025

Choose a reason for hiding this comment

Uh oh!

gentlegiantJGC commented Sep 26, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

rwgk commented Sep 26, 2025

Uh oh!

InvincibleRMC left a comment

Choose a reason for hiding this comment

Uh oh!

rwgk commented Sep 26, 2025

Uh oh!

gentlegiantJGC commented Sep 26, 2025

Uh oh!

rwgk commented Sep 26, 2025

Uh oh!

rwgk commented Sep 26, 2025

Uh oh!

rwgk commented Sep 26, 2025

Uh oh!

gentlegiantJGC commented Sep 26, 2025

Uh oh!

gentlegiantJGC commented Sep 26, 2025

Uh oh!

rwgk commented Sep 26, 2025

Uh oh!

rwgk commented Sep 26, 2025

Uh oh!

rwgk commented Sep 26, 2025

Uh oh!

gentlegiantJGC commented Sep 27, 2025

Uh oh!

rwgk commented Sep 27, 2025

Uh oh!

rwgk commented Sep 27, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

gentlegiantJGC commented Sep 16, 2025 •

edited

Loading

gentlegiantJGC commented Sep 16, 2025 •

edited

Loading

gentlegiantJGC commented Sep 26, 2025 •

edited

Loading

rwgk commented Sep 27, 2025 •

edited

Loading