GH-145668: Add FOR_ITER specialization for virtual iterators. Specialize GET_ITER. by markshannon · Pull Request #147967 · python/cpython

markshannon · 2026-04-01T13:37:34Z

Add FOR_ITER specialization for virtual iterators. Specialize GET_ITER.

Add FOR_ITER_VIRTUAL to specialize FOR_ITER for virtual iterators
Add GET_ITER_SELF to specialize GET_ITER for iterators (including generators)
Add GET_ITER_VIRTUAL to specialize GET_ITER for iterables as virtual iterators
Add tp_iteritem internal slot to PyTypeObject to support a wider range of classes as virtual iterators.

This PR adds strs as virtual iterators, with the potential to add bytes, bytearray, frozenset, frozendict and others in the future.

Issue: Extend FOR_ITER specializations to cover other indexable sequences #145668

📚 Documentation preview 📚: https://cpython-previews--147967.org.readthedocs.build/

* Add FOR_ITER_VIRTUAL to specialize FOR_ITER for virtual iterators * Add GET_ITER_SELF to specialize GET_ITER for iterators (including generators) * Add GET_ITER_VIRTUAL to specialize GET_ITER for iterables as virtual iterators

Fidget-Spinner · 2026-04-01T14:18:32Z

Doesn't this change the ABI guarantees of PyTypeObject? Or are there none of those? In any case, feels strange that we are adding a public field that is opaque for our own use.

markshannon · 2026-04-01T15:43:33Z

tp_watchers is already documented as internal.
The header is public, so the fields have to go in a public header even if they aren't to be used.

…eption handling in genexpr

Fidget-Spinner

A few comments

Fidget-Spinner · 2026-04-10T16:05:07Z

Doc/c-api/typeobj.rst

   +------------------------------------------------+-----------------------------------+-------------------+---+---+---+---+
   | :c:member:`~PyTypeObject.tp_vectorcall`        | :c:type:`vectorcallfunc`          |                   |   |   |   |   |
   +------------------------------------------------+-----------------------------------+-------------------+---+---+---+---+
+   | [:c:member:`~PyTypeObject.tp_iteritem`]        | opaque function pointer           |                   |   |   |   |   |


Docs are warning that

c:member reference target not found: PyTypeObject.tp_iteritem [ref.member]

I'm not sure what I can do about that. It is defined, but Sphinx won't acknowledge it, unless I define iteritemfunc which I don't want to define, as it is internal.

I think you can remove the :c:memberand just make it a normalcode`` role then?

Name it _tp_iteritemfunc with an underscore, and don't document it.

Include/object.h

Fidget-Spinner · 2026-04-10T16:08:06Z

Python/bytecodes.c

+            _GUARD_ITER_VIRTUAL +
+            _PUSH_TAGGED_ZERO;
+
+        op(_GET_ITER_TRAD, (iterable -- iter, index_or_null)) {


what does trad mean?

Traditional

Fidget-Spinner · 2026-04-13T16:02:39Z

Do you have benchmark numbers for this?

I'm wondering of 3 extra opcodes is worth it. I think I might have a solution for the opcode problem.

markshannon · 2026-04-14T08:45:13Z

Do you have benchmark numbers for this?

In the noise. But this enables more optimizations, so neutral is good.

I'm wondering of 3 extra opcodes is worth it.

The FOR_ITER_VIRTUAL is worth it, as it allows more common iterables to be treated as virtual iterators.

By themselves, the two new GET_ITER specializations probably wouldn't be.
They're worth it because they complement the FOR_ITER and, very soon, the SEND specializations.
For those specializations to work we need GET_ITER to handle the three distinct cases:

Already an iterator: do nothing
An iterable, supporting virtual iteration: push tagged zero
Other iterables, call __iter__ to get the iterator

So the specializations follow naturally.

I think I might have a solution for the opcode problem.

That's good, but not directly relevant to this PR I think.
We aren't in imminent danger of running out of opcodes, provided we followed a reasonably principled approach to specializations.

encukou · 2026-04-14T14:13:51Z

Include/object.h

+    PyObject *object;
+    Py_ssize_t index;
+} _PyObjectIndexPair;
+typedef _PyObjectIndexPair (*iteritemfunc) (PyObject *, Py_ssize_t index);


Please name this _Py_iteritemfunc so I don't need to rename it later. If anyone complains, point at me.

encukou · 2026-04-14T14:14:29Z

Doc/c-api/typeobj.rst

   +------------------------------------------------+-----------------------------------+-------------------+---+---+---+---+
   | :c:member:`~PyTypeObject.tp_vectorcall`        | :c:type:`vectorcallfunc`          |                   |   |   |   |   |
   +------------------------------------------------+-----------------------------------+-------------------+---+---+---+---+
+   | [:c:member:`~PyTypeObject.tp_iteritem`]        | opaque function pointer           |                   |   |   |   |   |


Name it _tp_iteritemfunc with an underscore, and don't document it.

encukou · 2026-04-14T14:16:39Z

Include/object.h

 typedef PyObject *(*richcmpfunc) (PyObject *, PyObject *, int);
 typedef PyObject *(*getiterfunc) (PyObject *);
 typedef PyObject *(*iternextfunc) (PyObject *);
+typedef struct _py_object_index_pair {


AFAIK the tag is unused; add it later if that changes.

Suggested change

typedef struct _py_object_index_pair {

typedef struct {

markshannon added 2 commits April 1, 2026 13:58

Document new tp_iteritem function slot.

74b5289

markshannon requested review from Fidget-Spinner, ZeroIntensity, ericsnowcurrently, savannahostrowski and tomasr8 as code owners April 1, 2026 13:37

bedevere-app bot added the awaiting core review label Apr 1, 2026

bedevere-app bot mentioned this pull request Apr 1, 2026

Extend FOR_ITER specializations to cover other indexable sequences #145668

Open

Simplify list_iteritem

4dd6992

markshannon added skip news labels Apr 1, 2026

markshannon added 7 commits April 2, 2026 12:40

Make it clearer to MSVC that variable doesn't escape

4376246

Return a small struct in tp_iteritem to support MSVC tailcalling

5a81e54

Fix editing slip

b927de4

Put limited RESUME at start of genexpr for free-threading. Fix up exc…

044cce2

…eption handling in genexpr

Merge branch 'main' into specialize-iteration-with-jit

4ba6ecd

Merge branch 'main' into specialize-iteration-with-jit

b08dce2

Fix test_dis

830068c

markshannon requested a review from iritkatriel as a code owner April 8, 2026 16:43

markshannon added 3 commits April 9, 2026 09:12

Merge branch 'main' into specialize-iteration-with-jit

931084a

Merge branch 'main' into specialize-iteration-with-jit

87215d5

Fix magic number

73b9b87

Fidget-Spinner reviewed Apr 10, 2026

View reviewed changes

markshannon added 3 commits April 13, 2026 14:57

Address review comments

fdef628

Merge branch 'main' into specialize-iteration-with-jit

b8baca5

Make struct private

b3dc340

Try to keep sphinx happy

1c269b6

markshannon requested review from AA-Turner, StanFromIreland and hugovk as code owners April 14, 2026 13:55

encukou reviewed Apr 14, 2026

View reviewed changes

Declare tp_iteritem and iteritemfunc private by adding _ prefix to names

09ecaef

Uh oh!

Conversation

markshannon commented Apr 1, 2026 • edited by github-actions bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Fidget-Spinner commented Apr 1, 2026

Uh oh!

markshannon commented Apr 1, 2026

Uh oh!

Fidget-Spinner 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!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Fidget-Spinner commented Apr 13, 2026

Uh oh!

markshannon commented Apr 14, 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!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

markshannon commented Apr 1, 2026 •

edited by github-actions bot

Loading