Inconsistent results when building for 3.14

[clin1234]

Within CI tests for 3.14 within my fork, I am getting inconsistent results. When cargo test --no-default-features --lib --tests is run within GH Actions, which fetches the (non-GIL_ Python from https://github.com/actions/python-versions/releases, I get the following error

failures:

---- ffi::tests::test_inc_dec_ref_immortal stdout ----
thread 'ffi::tests::test_inc_dec_ref_immortal' panicked at src/ffi/tests.rs:323:9:
assertion `left == right` failed
  left: 5529[770](https://github.com/clin1234/pyo3/actions/runs/13242668921/job/36961533481#step:15:771)39373
 right: 552977039372
stack backtrace:
   0: rust_begin_unwind
             at /rustc/e71f9a9a98b0faf423844bf0ba7438f29dc27d58/library/std/src/panicking.rs:665:5
   1: core::panicking::panic_fmt
             at /rustc/e71f9a9a98b0faf423844bf0ba7438f29dc27d58/library/core/src/panicking.rs:76:14
   2: core::panicking::assert_failed_inner
   3: core::panicking::assert_failed
             at /home/runner/.rustup/toolchains/stable-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/core/src/panicking.rs:373:5
   4: pyo3::ffi::tests::test_inc_dec_ref_immortal::{{closure}}
             at ./src/ffi/tests.rs:323:9
   5: pyo3::marker::Python::with_gil
             at ./src/marker.rs:412:9
   6: pyo3::ffi::tests::test_inc_dec_ref_immortal
             at ./src/ffi/tests.rs:315:5
   7: pyo3::ffi::tests::test_inc_dec_ref_immortal::{{closure}}
             at ./src/ffi/tests.rs:314:31
   8: core::ops::function::FnOnce::call_once
             at /home/runner/.rustup/toolchains/stable-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/core/src/ops/function.rs:250:5
   9: core::ops::function::FnOnce::call_once
             at /rustc/e71f9a9a98b0faf423844bf0ba7438f29dc27d58/library/core/src/ops/function.rs:250:5
note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace.

---- types::mappingproxy::tests::test_set_item_refcnt stdout ----
thread 'types::mappingproxy::tests::test_set_item_refcnt' panicked at src/types/mappingproxy.rs:225:17:
assertion `left == right` failed
  left: 552977039379
 right: 552977039380
stack backtrace:
   0: rust_begin_unwind
             at /rustc/e71f9a9a98b0faf423844bf0ba7438f29dc27d58/library/std/src/panicking.rs:665:5
   1: core::panicking::panic_fmt
             at /rustc/e71f9a9a98b0faf423844bf0ba7438f29dc27d58/library/core/src/panicking.rs:76:14
   2: core::panicking::assert_failed_inner
   3: core::panicking::assert_failed
             at /home/runner/.rustup/toolchains/stable-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/core/src/panicking.rs:373:5
   4: pyo3::types::mappingproxy::tests::test_set_item_refcnt::{{closure}}
             at ./src/types/mappingproxy.rs:225:17
   5: pyo3::marker::Python::with_gil
             at ./src/marker.rs:412:9
   6: pyo3::types::mappingproxy::tests::test_set_item_refcnt
             at ./src/types/mappingproxy.rs:216:9
   7: pyo3::types::mappingproxy::tests::test_set_item_refcnt::{{closure}}
             at ./src/types/mappingproxy.rs:215:30
   8: core::ops::function::FnOnce::call_once
             at /home/runner/.rustup/toolchains/stable-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/core/src/ops/function.rs:250:5
   9: core::ops::function::FnOnce::call_once
             at /rustc/e71f9a9a98b0faf423844bf0ba7438f29dc27d58/library/core/src/ops/function.rs:250:5
note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace.


failures:
    ffi::tests::test_inc_dec_ref_immortal
    types::mappingproxy::tests::test_set_item_refcnt

test result: FAILED. 653 passed; 2 failed; 0 ignored; 0 measured; 0 filtered out; finished in 4.67s

error: test failed, to rerun pass `--lib`

Reproduced when building against the same kind of interpreter obtained from this PPA.

When building against a freethreaded interpreter, however, all the tests pass. Tested within interpreter obtained from the above PPA and also from https://src.fedoraproject.org/rpms/python3.14

[clin1234]

Update: above error is resolved, but a new one arises; specifically, when cargo +stable run --manifest-path=pyo3-ffi-check/Cargo.toml is run:

error[E0609]: no field `_padding` on type `_object`
  --> src/main.rs:45:13
   |
45 |             pyo3_ffi_check_macro::for_all_fields!($name, check_field);
   |             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
...
70 |     pyo3_ffi_check_macro::for_all_structs!(check_struct);
   |     ---------------------------------------------------- in this macro invocation
   |
   = note: this error originates in the macro `pyo3_ffi_check_macro::for_all_fields` which comes from the expansion of the macro `pyo3_ffi_check_macro::for_all_structs` (in Nightly builds, run with -Z macro-backtrace for more info)

error[E0609]: no field `ob_base` on type `bindings::_PyGenObject`
  --> src/main.rs:45:13
   |
45 |             pyo3_ffi_check_macro::for_all_fields!($name, check_field);
   |             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
...
70 |     pyo3_ffi_check_macro::for_all_structs!(check_struct);
   |     ---------------------------------------------------- in this macro invocation
   |
   = note: this error originates in the macro `pyo3_ffi_check_macro::for_all_fields` which comes from the expansion of the macro `pyo3_ffi_check_macro::for_all_structs` (in Nightly builds, run with -Z macro-backtrace for more info)

error[E0609]: no field `gi_weakreflist` on type `bindings::_PyGenObject`
  --> src/main.rs:45:13
   |
45 |             pyo3_ffi_check_macro::for_all_fields!($name, check_field);
   |             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
...
70 |     pyo3_ffi_check_macro::for_all_structs!(check_struct);
   |     ---------------------------------------------------- in this macro invocation
   |
   = note: this error originates in the macro `pyo3_ffi_check_macro::for_all_fields` which comes from the expansion of the macro `pyo3_ffi_check_macro::for_all_structs` (in Nightly builds, run with -Z macro-backtrace for more info)

error[E0609]: no field `gi_name` on type `bindings::_PyGenObject`
  --> src/main.rs:45:13
   |
45 |             pyo3_ffi_check_macro::for_all_fields!($name, check_field);
   |             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
...
70 |     pyo3_ffi_check_macro::for_all_structs!(check_struct);
   |     ---------------------------------------------------- in this macro invocation
   |
   = note: this error originates in the macro `pyo3_ffi_check_macro::for_all_fields` which comes from the expansion of the macro `pyo3_ffi_check_macro::for_all_structs` (in Nightly builds, run with -Z macro-backtrace for more info)

error[E0609]: no field `gi_qualname` on type `bindings::_PyGenObject`
  --> src/main.rs:45:13
   |
45 |             pyo3_ffi_check_macro::for_all_fields!($name, check_field);
   |             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
...
70 |     pyo3_ffi_check_macro::for_all_structs!(check_struct);
   |     ---------------------------------------------------- in this macro invocation
   |
   = note: this error originates in the macro `pyo3_ffi_check_macro::for_all_fields` which comes from the expansion of the macro `pyo3_ffi_check_macro::for_all_structs` (in Nightly builds, run with -Z macro-backtrace for more info)

error[E0609]: no field `gi_exc_state` on type `bindings::_PyGenObject`
  --> src/main.rs:45:13
   |
45 |             pyo3_ffi_check_macro::for_all_fields!($name, check_field);
   |             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
...
70 |     pyo3_ffi_check_macro::for_all_structs!(check_struct);
   |     ---------------------------------------------------- in this macro invocation
   |
   = note: this error originates in the macro `pyo3_ffi_check_macro::for_all_fields` which comes from the expansion of the macro `pyo3_ffi_check_macro::for_all_structs` (in Nightly builds, run with -Z macro-backtrace for more info)

error[E0609]: no field `gi_origin_or_finalizer` on type `bindings::_PyGenObject`
  --> src/main.rs:45:13
   |
45 |             pyo3_ffi_check_macro::for_all_fields!($name, check_field);
   |             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
...
70 |     pyo3_ffi_check_macro::for_all_structs!(check_struct);
   |     ---------------------------------------------------- in this macro invocation
   |
   = note: this error originates in the macro `pyo3_ffi_check_macro::for_all_fields` which comes from the expansion of the macro `pyo3_ffi_check_macro::for_all_structs` (in Nightly builds, run with -Z macro-backtrace for more info)

error[E0609]: no field `gi_hooks_inited` on type `bindings::_PyGenObject`
  --> src/main.rs:45:13
   |
45 |             pyo3_ffi_check_macro::for_all_fields!($name, check_field);
   |             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
...
70 |     pyo3_ffi_check_macro::for_all_structs!(check_struct);
   |     ---------------------------------------------------- in this macro invocation
   |
   = note: this error originates in the macro `pyo3_ffi_check_macro::for_all_fields` which comes from the expansion of the macro `pyo3_ffi_check_macro::for_all_structs` (in Nightly builds, run with -Z macro-backtrace for more info)

error[E0609]: no field `gi_closed` on type `bindings::_PyGenObject`
  --> src/main.rs:45:13
   |
45 |             pyo3_ffi_check_macro::for_all_fields!($name, check_field);
   |             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
...
70 |     pyo3_ffi_check_macro::for_all_structs!(check_struct);
   |     ---------------------------------------------------- in this macro invocation
   |
   = note: this error originates in the macro `pyo3_ffi_check_macro::for_all_fields` which comes from the expansion of the macro `pyo3_ffi_check_macro::for_all_structs` (in Nightly builds, run with -Z macro-backtrace for more info)

error[E0609]: no field `gi_running_async` on type `bindings::_PyGenObject`
  --> src/main.rs:45:13
   |
45 |             pyo3_ffi_check_macro::for_all_fields!($name, check_field);
   |             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
...
70 |     pyo3_ffi_check_macro::for_all_structs!(check_struct);
   |     ---------------------------------------------------- in this macro invocation
   |
   = note: this error originates in the macro `pyo3_ffi_check_macro::for_all_fields` which comes from the expansion of the macro `pyo3_ffi_check_macro::for_all_structs` (in Nightly builds, run with -Z macro-backtrace for more info)

error[E0609]: no field `gi_frame_state` on type `bindings::_PyGenObject`
  --> src/main.rs:45:13
   |
45 |             pyo3_ffi_check_macro::for_all_fields!($name, check_field);
   |             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
...
70 |     pyo3_ffi_check_macro::for_all_structs!(check_struct);
   |     ---------------------------------------------------- in this macro invocation
   |
   = note: this error originates in the macro `pyo3_ffi_check_macro::for_all_fields` which comes from the expansion of the macro `pyo3_ffi_check_macro::for_all_structs` (in Nightly builds, run with -Z macro-backtrace for more info)

error[E0609]: no field `gi_iframe` on type `bindings::_PyGenObject`
  --> src/main.rs:45:13
   |
45 |             pyo3_ffi_check_macro::for_all_fields!($name, check_field);
   |             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
...
70 |     pyo3_ffi_check_macro::for_all_structs!(check_struct);
   |     ---------------------------------------------------- in this macro invocation
   |
   = note: this error originates in the macro `pyo3_ffi_check_macro::for_all_fields` which comes from the expansion of the macro `pyo3_ffi_check_macro::for_all_structs` (in Nightly builds, run with -Z macro-backtrace for more info)

[ngoldbaum]

I think we resolved this by updating the FFI definitions in your PR.

[clin1234]

Upstream CI passes, but not within my fork; my PR adds CPython 3.14 for CI testing, and there are still FFI test errors. I'll have time to tackle this early next week

[ngoldbaum]

Ah yes there's still the issue with the PyObject layout on the GIL-enabled build. The ffi-check macro likely needs a special case.

[clin1234]

Also, how does the ffi-check handle pseudo-anonymous structs, like those introduced in 1e627f9?

[ngoldbaum]

That's what I was referring to here:

#4811 (comment)

The special-case in ffi-check needs to be updated. Debugging macros is no fun and I'm not familiar enough with bindgen to know how to spell it.

This isn't really a problem with the FFI bindings per se, just a missing special case in the test to account for bindgen not handling this nicely.