Skip to content

GH-126910: Revert "Make _Py_get_machine_stack_pointer return the stack pointer (#147945)"#147994

Merged
encukou merged 1 commit intopython:mainfrom
encukou:revert-255026d
Apr 2, 2026
Merged

GH-126910: Revert "Make _Py_get_machine_stack_pointer return the stack pointer (#147945)"#147994
encukou merged 1 commit intopython:mainfrom
encukou:revert-255026d

Conversation

@encukou
Copy link
Copy Markdown
Member

@encukou encukou commented Apr 2, 2026

This reverts commit 255026d.

I suspect this commit broke a tier-1 buildbot; I'll use this PR to confirm.
Per PEP-11, any breakage should be fixed or reverted immediately.

@encukou
Copy link
Copy Markdown
Member Author

encukou commented Apr 2, 2026

!buildbot AMD64.CentOS9.NoGIL

@bedevere-bot
Copy link
Copy Markdown

🤖 New build scheduled with the buildbot fleet by @encukou for commit 6803150 🤖

Results will be shown at:

https://buildbot.python.org/all/#/grid?branch=refs%2Fpull%2F147994%2Fmerge

The command will test the builders whose names match following regular expression: AMD64.CentOS9.NoGIL

The builders matched are:

  • AMD64 CentOS9 NoGIL Refleaks PR
  • AMD64 CentOS9 NoGIL PR

@encukou
Copy link
Copy Markdown
Member Author

encukou commented Apr 2, 2026

@Fidget-Spinner @markshannon Are you able to investigate/fix this today?
If not, let's revert so the buildbot can catch other issues, and re-apply later.

@markshannon
Copy link
Copy Markdown
Member

markshannon commented Apr 2, 2026

I can't reproduce the failure on either fedora x86-64 or ubuntu Aarch64.

Note, that #147945 doesn't change the way frames are laid out or any aspect of the program, except:

  • How JIT frames are laid out (there is no JIT active on the failing buildbot)
  • How deep the stack can go before a RecursionError is raised.

Does CentOS have a glibc missing a way to tell the stack size?

I want to try increasing the estimated stack size a bit, as I suspect the combination of debug build, no-gil and perf trampolines is causing the stack to grow very fast.

@markshannon
Copy link
Copy Markdown
Member

#148002

@markshannon
Copy link
Copy Markdown
Member

It doesn't look like it is a stack issue.

It looks like py_trampoline_evaluator isn't calling the generated function for the code object. I don't know why.
It looks like whatever the problem is is being swallowed, not reported:
https://github.com/python/cpython/blob/main/Python/perf_trampoline.c#L452

@pablogsal any idea why py_trampoline_evaluator is not working?

Copy link
Copy Markdown
Member

@markshannon markshannon left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@encukou go ahead with the revert. I doubt this will be fixed before the weekend

@encukou encukou merged commit 9b08f8c into python:main Apr 2, 2026
83 of 84 checks passed
@encukou
Copy link
Copy Markdown
Member Author

encukou commented Apr 2, 2026

I also can't reproduce on a centos container (on a Fedora kernel).

@encukou encukou deleted the revert-255026d branch April 2, 2026 14:53
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants