neonene
neonene
@ncoghlan > 1. capsule reference is acquired > 2. module is reloaded > 3. capsule reference is used I think I've found a regression case. A user can keep the...
FYI, `_datetime` code owners are not involved in PEP-687 implementation (no response).
[gh-118337](https://github.com/python/cpython/pull/118337) fully implements: >1.(regular) Multi-phase-init with per-module heaptypes (and C-API). abi: Store a reference (to the C-API struct) in PyInterpreterState. It passes refleak bot tests where the module is reloaded...
I would llke an SC decision about how much a module can depend on `PyInterpreterState`. `_datetime` and other modules are not so different except the presence of a capsule C-API,...
>* By contrast, `PyCapsule_GetPointer` is explicitly an export API. If `datetime` API includes the direct use of `PyCapsule_GetPointer()`, a capsule should be held until the interpreter finishes (by using `PyInterpreterState_GetDict()`?)....
Did @erlend-aasland point out that the problem applies to `_datetime`?
`_datetime` is the last remaining extension module that need to be migrated. While we can suggest that the new generic C-API is needed, there seems to be no individual discussion...
@erlend-aasland Can you explain how the discussion resolves `_datetime ` isolation? If it is really effective, why didn't you either propose nor apply it before the **3.12** beta?
Then, why didn't you suggest it on your spare time?
`_datetime` needs rationales if per-interpreter heaptypes are required.