gh-145866: Convert SET_UPDATE to leave its inputs on the stack to be cleaned up by _POP_TOP#145979
gh-145866: Convert SET_UPDATE to leave its inputs on the stack to be cleaned up by _POP_TOP#145979Sacul0457 wants to merge 6 commits intopython:mainfrom
SET_UPDATE to leave its inputs on the stack to be cleaned up by _POP_TOP#145979Conversation
Sacul0457
requested review from
Fidget-Spinner,
markshannon,
savannahostrowski and
tomasr8
as code owners
Python/bytecodes.c
Outdated
| i = iterable; | ||
| DEAD(iterable); | ||
| if (err < 0) { | ||
| ERROR_NO_POP(); | ||
| } |
There was a problem hiding this comment.
i should transfer ownership only after the error branch case.
There was a problem hiding this comment.
Yes please. It doesn't affect the generated code at the moment, but it does affect the correctness/reasoning.
The reasoning is as follows:
- We call ERROR_NO_POP in the error branch, implying everything on the stack is a live value.
- However, the current code marks the stack value as dead before the ERROR_NO_POP.
- and 2. are contradicting each other.
|
I believe the failure isn't related to the pr |
|
@Sacul0457 after this PR, can you please raise a PR to add your name (if you feel comfortable) to https://docs.python.org/dev/whatsnew/3.15.html#upgraded-jit-compiler ? Specifically, the section on
|
|
@Sacul0457 please fix the merge conflicts. |
|
Not sure why the tests are failing, seems unrelated however |
|
I think you need to pull a later main in. The fuzzer failures sometimes indicate that the pyc magic number is out of sync. I recently updated main with a new pyc number. so maybe that's why? |
|
Sure, I'll pull in main and see if that fixes it. |
|
Welp, unfortunately the same three test seems to fail again and I have no idea why still :( |
|
Seems like #146069 should fix the CI failure. Will pull in main (and fix merge conflicts) once that is merged and hopefully the tests pass |
3 participants
_POP_TOP#145866