[autograd] Thread-safe Python wrapping for Node by colesbury · Pull Request #179767 · pytorch/pytorch

colesbury · 2026-04-08T22:52:55Z

Stack from ghstack (oldest at bottom):

Replace the raw PyObject* on Node with an atomic PyObjectSlot
and use the same strategy that we already use for Tensor and Storage
PyObject wrappers.

Authored with Claude.

[ghstack-poisoned]

pytorch-bot · 2026-04-08T22:52:58Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/179767

📄 Preview Python docs built from this PR
📄 Preview C++ docs built from this PR
❓ Need help or want to give feedback on the CI? Visit the bot commands wiki

Note: Links to docs will display an error until the docs builds have been completed.

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

Rolling out OSDC (ARC) runners on pull & trunk workflows in PyTorch main

✅ No Failures

As of commit fb0d22d with merge base 83de8b8 ():
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

Replace the raw PyObject* on Node with an atomic PyObjectSlot and use the same strategy that we already use for Tensor and Storage PyObject wrappers. Authored with Claude. ghstack-source-id: e467af4 Pull-Request: #179767

pytorch-bot · 2026-04-08T22:53:01Z

This PR needs a `release notes:` label

If your changes are user facing and intended to be a part of release notes, please use a label starting with release notes:.

If not, please add the topic: not user facing label.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "topic: not user facing"

For more information, see
https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.

colesbury · 2026-04-08T23:15:41Z

Got Claude to run some benchmarks on this stack of PRs and didn't see a noticeable change, maybe a tiny bit faster for a few cases, but importantly doesn't look like a regression anywhere:

https://gist.github.com/colesbury/8ac150f3d6e43d3adff2f888e9aeb5f5

[ghstack-poisoned]

Replace the raw PyObject* on Node with an atomic PyObjectSlot and use the same strategy that we already use for Tensor and Storage PyObject wrappers. Authored with Claude. ghstack-source-id: 5fa9478 Pull-Request: #179767

[ghstack-poisoned]

Replace the raw PyObject* on Node with an atomic PyObjectSlot and use the same strategy that we already use for Tensor and Storage PyObject wrappers. Authored with Claude. ghstack-source-id: 872d766 Pull-Request: #179767

[ghstack-poisoned]

Replace the raw PyObject* on Node with an atomic PyObjectSlot and use the same strategy that we already use for Tensor and Storage PyObject wrappers. Authored with Claude. ghstack-source-id: 265cdaf Pull-Request: #179767

[ghstack-poisoned]

Replace the raw PyObject* on Node with an atomic PyObjectSlot and use the same strategy that we already use for Tensor and Storage PyObject wrappers. Authored with Claude. ghstack-source-id: 165b906 Pull-Request: #179767

[ghstack-poisoned]

Replace the raw PyObject* on Node with an atomic PyObjectSlot and use the same strategy that we already use for Tensor and Storage PyObject wrappers. Authored with Claude. ghstack-source-id: 176276e Pull-Request: #179767

Replace the raw PyObject* on Node with an atomic PyObjectSlot and use the same strategy that we already use for Tensor and Storage PyObject wrappers. Authored with Claude. ghstack-source-id: 176276e Pull-Request: pytorch#179767

[ghstack-poisoned]

albanD

Sounds good!

albanD · 2026-04-15T21:16:31Z

+
+// Enable PyObject support for intrusive_ptr<Node> so that the refcount
+// machinery calls incref_pyobject/decref_pyobject on transitions.
+namespace c10::detail {


Claude seems to take offence at this one:

Missing #ifndef C10_MOBILE guard on TargetTraits specialization —
torch/csrc/autograd/node.h:707-715. Both TensorImpl.h:3096 and StorageImpl.h:407 wrap their
TargetTraits specialization in #ifndef C10_MOBILE. The intrusive_ptr refcount machinery relies on
this: in mobile builds, has_pyobject paths are disabled and guarded by
TORCH_INTERNAL_ASSERT_DEBUG_ONLY(!detail::has_pyobject(combined)) (see intrusive_ptr.h:1191). If
node.h is ever transitively included in a mobile build, can_have_pyobject = true for Node would fire
that assertion. Even if autograd headers aren't used on mobile today, the guard should be present for
consistency and safety.

I'll add the C10_MOBILE ifdef guard

albanD

Sounds good!

[ghstack-poisoned]

Stack from [ghstack](https://github.com/ezyang/ghstack/tree/0.14.0) (oldest at bottom): * #179767 * -> #179766 * #179765 * #179764 Replace std::shared_ptr<Node> with c10::intrusive_ptr<Node> throughout the autograd system. Node now inherits from c10::intrusive_ptr_target instead of std::enable_shared_from_this<Node>. Authored with Claude. Pull Request resolved: #179766 Approved by: https://github.com/albanD ghstack dependencies: #179764, #179765

Replace the raw PyObject* on Node with an atomic PyObjectSlot and use the same strategy that we already use for Tensor and Storage PyObject wrappers. Authored with Claude. [ghstack-poisoned]

Stack from [ghstack](https://github.com/ezyang/ghstack/tree/0.14.0) (oldest at bottom): * #179767 * -> #179766 * #179765 * #179764 Replace std::shared_ptr<Node> with c10::intrusive_ptr<Node> throughout the autograd system. Node now inherits from c10::intrusive_ptr_target instead of std::enable_shared_from_this<Node>. Authored with Claude. cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx kadeng chauhang amjames Lucaskabela jataylo azahed98 xmfan aditvenk [ghstack-poisoned]

Update

2131904

[ghstack-poisoned]

colesbury requested review from albanD and soulitzer as code owners April 8, 2026 22:52

This was referenced Apr 8, 2026

[c10] Consolidate PyObject preservation helpers #179764

Closed

[autograd] Extract Node class into node.h #179765

Closed

[autograd] Convert shared_ptr<Node> to intrusive_ptr<Node> #179766

Closed

colesbury added the topic: not user facing topic category label Apr 8, 2026

Update

a210ed4

[ghstack-poisoned]

Update

0fa5128

[ghstack-poisoned]

Update

7f4d00e

[ghstack-poisoned]

Update

bf1e780

[ghstack-poisoned]

Update

9630013

[ghstack-poisoned]

Update

0d2bdc7

[ghstack-poisoned]

colesbury mentioned this pull request Apr 10, 2026

[autograd] Use PyObjectSlot for PyNode/THPFunction wrapping #179924

Open

Update

7a33011

[ghstack-poisoned]

colesbury mentioned this pull request Apr 13, 2026

[autograd] Minor cleanups in autograd and compiled_autograd headers #180224

Open

albanD reviewed Apr 15, 2026

View reviewed changes

albanD approved these changes Apr 15, 2026

View reviewed changes

Update

49f5545

[ghstack-poisoned]

Update

317543f

[ghstack-poisoned]

Update on "[autograd] Thread-safe Python wrapping for Node"

fb0d22d

Replace the raw PyObject* on Node with an atomic PyObjectSlot and use the same strategy that we already use for Tensor and Storage PyObject wrappers. Authored with Claude. [ghstack-poisoned]

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[autograd] Thread-safe Python wrapping for Node#179767

[autograd] Thread-safe Python wrapping for Node#179767
colesbury wants to merge 11 commits intogh/colesbury/6/basefrom
gh/colesbury/6/head

colesbury commented Apr 8, 2026 •

edited

Loading

Uh oh!

pytorch-bot Bot commented Apr 8, 2026 •

edited

Loading

Uh oh!

pytorch-bot Bot commented Apr 8, 2026

Uh oh!

colesbury commented Apr 8, 2026

Uh oh!

albanD left a comment

Uh oh!

albanD Apr 15, 2026

Uh oh!

colesbury Apr 16, 2026

Uh oh!

albanD left a comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

colesbury commented Apr 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pytorch-bot Bot commented Apr 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/179767

❗ 1 Active SEVs

✅ No Failures

Uh oh!

pytorch-bot Bot commented Apr 8, 2026

This PR needs a release notes: label

Uh oh!

colesbury commented Apr 8, 2026

Uh oh!

albanD left a comment

Choose a reason for hiding this comment

Uh oh!

albanD Apr 15, 2026

Choose a reason for hiding this comment

Uh oh!

colesbury Apr 16, 2026

Choose a reason for hiding this comment

Uh oh!

albanD left a comment

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

colesbury commented Apr 8, 2026 •

edited

Loading

pytorch-bot Bot commented Apr 8, 2026 •

edited

Loading

This PR needs a `release notes:` label