gh-149044: Implement PEP 820 – PySlot: Unified slot system for the C API by encukou · Pull Request #149055 · python/cpython

encukou · 2026-04-27T16:07:01Z

Issue: Implement PEP 820 – PySlot: Unified slot system for the C API #149044

📚 Documentation preview 📚: https://cpython-previews--149055.org.readthedocs.build/

ZeroIntensity

Documentation looks pretty good. I haven't looked at all of the C code yet.

Co-authored-by: Peter Bierma <zintensitydev@gmail.com>

read-the-docs-community · 2026-04-30T10:52:05Z

Documentation build overview

📚 cpython-previews | 🛠️ Build #32533679 | 📁 Comparing 1d09dac against main (a386a52)

🔍 Preview build

65 files changed · + 2 added · ± 63 modified

+ Added

± Modified

ngoldbaum

I wrote FFI bindings for this PR for PyO3 this afternoon and spotted several minor issues.

ngoldbaum · 2026-05-03T17:31:37Z

+#define PySlot_PTR(NAME, VALUE) \
+    {(NAME), PySlot_INTPTR, {0}, {(void*)(VALUE)}}
+
+#define PySlot_PTR_STATIC(NAME, VALUE) \
+    {(NAME), PySlot_INTPTR | PySlot_STATIC, {0}, {(void*)(VALUE)}}


Can you spell these last two with the explicit names of the struct fields like the ones above? Also while you're touching this code, IMO these two definitions should come before PySlot_END.

These macros are for code compatible with C++11 and below, which doesn't support designated initializers (that is, the explicit names).

So, both style issues are intentional :) I added a comment.

ngoldbaum · 2026-05-03T17:32:27Z

+#define PySlot_STATIC 0x02
+#define PySlot_INTPTR 0x04
+
+#define Py_slot_invalid 0xffff


for consistency should this be called Py_slot_INVALID maybe?

That would be consistent with the wrong thing. Slot IDs are named Py_tp_repr, Py_mod_exec, etc. Flags are a new thing, so they follow the current style guide: mixed prefix + uppercase.

ngoldbaum · 2026-05-03T17:34:28Z

+#define _Py_SLOT_COMPAT_VALUE(OLD, NEW) OLD
+#endif
+
+#define Py_slot_end 0


Maybe worth adding a comment above these that they're all u16 values?

In C, the preprocessor macros are just numbers :)
In PyType_Slot, these values are used as int.

ngoldbaum · 2026-05-03T17:53:46Z

+#define Py_tp_finalize 80
+#define Py_am_send 81
+#define Py_tp_vectorcall 82
+#define Py_tp_token 83


Everything through here is duplicated with typeslots.h. Was that intentional? It probably makes sense for there to be only one source of truth for these constants.

Thanks for the catch! I seem to have dropped the commit that removes the unused file.

ngoldbaum · 2026-05-04T00:52:42Z

+#ifndef _Py_HAVE_SLOTS_H
+#define _Py_HAVE_SLOTS_H
+
+typedef void (*_Py_funcptr_t)(void);


PyO3 needs to wrap this type directly so you might as well give this a public non-underscored name.

Can PyO3 use the underlying type, or better yet, some “generic C function pointer” type?

Unfortunately no, because Rust doesn't let you cast between function pointer types without using unsafe and std::mem::transmute. That said, in my second draft of this, I isolated the unsafety into a PySlot_FUNC rust macro that hides the wrapped function pointer, so I take back my comment from last night.

Keep in mind this is not the correct function type, just a function type. At this level, unsafe transmuation is exactly what you need in Rust.
(A safer layer on top of the ABI would need to have separate C-specific and C++-specific variants, and presumably a Rust-specific one as well. I didn't get that into 3.15.)

encukou · 2026-05-04T10:21:53Z

I intend to merge this today (or early tomorrow) to get it in 3.15, as postponing to 3.16 would be very inconvenient. Beta is for fixing features (or removing, if this turns out to be a bad idea after all).

ZeroIntensity

I think this is in a mergeable state. I read through all of the PEP, documentation, and code, and I couldn't find any glaring issues.

I left a number of non-blocking questions/comments; feel free to ignore them for now and address them after the beta freeze.

ZeroIntensity · 2026-05-04T14:16:07Z

        self.assertRaisesRegex(
-            SystemError,
-            "Py_tp_bases is not a tuple",
+            TypeError,
+            "metaclass conflict",
            _testcapi.create_heapctype_with_none_bases_slot
        )


Hmm, this seems like a downgrade in terms of the error message. I don't see a change to create_heapctype_with_none_bases_slot in this PR, so is there a reason the error changed?

Yeah, that's a side effect from merging Py_tp_base & Py_tp_bases -- the latter strictly required a tuple of types, so it had this message.
This is a silly restriction. I'll document lifting it.

(And I'll get SC permission to update the PEP, or revert this.)

ZeroIntensity · 2026-05-04T14:30:27Z

+                }
+                break;
        }


What happens if there's a slot other than Py_slot_invalid or Py_mod_exec here? Does _PySlotIterator_Next somehow handle that ahead-of-time?

This is PyModule_ExecDef; it ignores slots other than Py_mod_exec.

Py_slot_invalid is the error indicator here (i.e. an exception was set in _PySlotIterator_Next) -- not the standard error API but nicer than the alternatives I tried.

ZeroIntensity · 2026-05-04T14:31:22Z

+                memcpy(tp_doc, it.current.sl_ptr, len);
            }
            break;


Same question here -- do we want a default: case for debugging?

This is the first pass, most slots (especially the function pointers) are handled in the second pass.

_PySlotIterator_Next handles unknown slots, and ones that don't apply to types -- it sets an exception and reports Py_slot_invalid.

ZeroIntensity · 2026-05-04T14:37:58Z

+    _PySlot_KIND slot_struct_kind;
+} _PySlotIterator_state;
+
+#define SEEN_ENTRY_BITS (8 * sizeof(unsigned int))


It doesn't seem like this is used outside slots.c. I think that it would make more sense to move it there, but if you'd really like to keep it in the header, let's be consistent and give this a _Py prefix.

Co-authored-by: Peter Bierma <zintensitydev@gmail.com> Co-authored-by: Petr Viktorin <encukou@gmail.com>

encukou added 30 commits April 24, 2026 13:40

Add header for internal slot iterator API

3b13a93

Add public API and a generated list of slots

0396743

Add slot iterator implementation & use it for types

6da3bb1

Unify duplicate & NULL handling

a8e018f

Simplify slot handling in typeobject.c

b96acd4

Port PyType_GetSlot; remove typeslots.inc

f3a6d7c

Port module creation

5f13d2e

Simplify internal API

4660f2f

Switch modexport over to new slots

6c29000

Fix up NULL handling

dfb342c

Move PyType_FromMetaclass to slots

110464d

PyType_FromSlots

6fb0046

slots.csv update

866eafc

squash up

325f1a8

Put slot info in generated switches

e22e3af

Handle PyType_Spec specially

fea6e21

Clean up the generator

a89405c

Remove slots.csv

114a4fa

Prettier generation

8dc5d88

Tighten module creation

796dd75

Fix math

b942f10

Remove PySlot_HAS_FALLBACK

6a53b4c

Bump version

f352558

Better comments

bd78c06

Limit deprecations to new-style slots

181df86

Start on docs

644ae46

Require some slots to be static

0433d1e

Add to Stable ABI

b3131a3

blurb, versionadded, whatsnew

b35230d

Don't deprecate mixing the bases slots/args

ae0ea29

encukou requested review from AlexWaygood and hugovk as code owners April 28, 2026 12:35

bedevere-app Bot added the awaiting core review label Apr 28, 2026

ZeroIntensity reviewed Apr 28, 2026

View reviewed changes

Comment thread Doc/c-api/slots.rst Outdated

Comment thread Doc/c-api/slots.rst Outdated

AlexWaygood removed their request for review April 28, 2026 14:06

brettcannon removed their request for review April 28, 2026 16:44

encukou commented Apr 30, 2026

View reviewed changes

Comment thread Doc/c-api/slots.rst Outdated

encukou and others added 2 commits April 30, 2026 12:07

Apply suggestions from code review

986b24e

Co-authored-by: Peter Bierma <zintensitydev@gmail.com>

Merge branch 'main' into slots-pep820-mrg

3404bba

ngoldbaum mentioned this pull request May 2, 2026

Supporting 3.15.0b1 and finishing off support for the 3.15 ABI PyO3/pyo3#6012

Open

ngoldbaum reviewed May 3, 2026

View reviewed changes

ngoldbaum reviewed May 4, 2026

View reviewed changes

encukou added 2 commits May 4, 2026 08:43

Merge in the main branch

e21eb07

Address review comments

bef20e9

encukou mentioned this pull request May 4, 2026

gh-149101: Implement PEP 788 #149116

Open

ZeroIntensity approved these changes May 4, 2026

View reviewed changes

bedevere-app Bot added awaiting merge and removed awaiting core review labels May 4, 2026

encukou and others added 5 commits May 4, 2026 22:58

Apply suggestions from code review

018ce8b

Co-authored-by: Peter Bierma <zintensitydev@gmail.com> Co-authored-by: Petr Viktorin <encukou@gmail.com>

Prefix a private constant

6c9726a

Merge in the main branch

ca86080

Doc change in Py_tp_bases

a67116e

Fix badly formatted suggestion

1d09dac

encukou added the 🔨 test-with-buildbots Test PR w/ buildbots; report in status section label May 4, 2026

Merge in the main branch

06b2c3d

encukou merged commit 508b498 into python:main May 5, 2026
93 of 103 checks passed

bedevere-app Bot removed the awaiting merge label May 5, 2026

encukou deleted the slots-pep820-mrg branch May 5, 2026 07:18

Uh oh!

Conversation

encukou commented Apr 27, 2026 • edited by github-actions Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ZeroIntensity left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

read-the-docs-community Bot commented Apr 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Documentation build overview

Uh oh!

ngoldbaum left a comment

Choose a reason for hiding this comment

Uh oh!

ngoldbaum May 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

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!

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!

encukou commented May 4, 2026

Uh oh!

ZeroIntensity left a comment

Choose a reason for hiding this comment

Uh oh!

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!

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!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

encukou commented Apr 27, 2026 •

edited by github-actions Bot

Loading

read-the-docs-community Bot commented Apr 30, 2026 •

edited

Loading

ngoldbaum May 3, 2026 •

edited

Loading