GH-94851: fix immortal objects refcounting in compiler

[kumaraditya303]

• Issue: Check for ref counting bugs in debug mode caused by immortal objects. #94851

Found while working on #95001

[corona10]

@kumaraditya303 Thanks Kumar

I was also curious about this code also when you submitted the related patch before.

1. It will be helpful to review if you can suggest reproducible code as you did before.
2. There are several _Py_STR related codes that don't increase reference count when they are used.
   Should they be updated? and if yes, Would you like to submit the one-shot patch?

[kumaraditya303]

There are several _Py_STR related codes that don't increase reference count when they are used.

it depends on its usage, if you are owning the object or returning it from a function then incref is required else it is not required.

[kumaraditya303]

It will be helpful to review if you can suggest reproducible code as you did before.

If you follow the macro, it is capturing a reference to the new object so incref is required.

See comment : /* Same as ADDOP_LOAD_CONST, but steals a reference. */

[corona10]

@kumaraditya303

a = 1
print(f'{a:d} {a:d}  {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d}')

./python.exe -X showrefcount tt.py
1 1  1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
[-1 refs, 0 blocks]

[kumaraditya303]

a = 1
print(f'{a:d} {a:d}  {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d} {a:d}')

./python.exe -X showrefcount tt.py
1 1  1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
[-1 refs, 0 blocks]

Nice repro, thanks!

[corona10]

And I checked that your patch fix this issue.

./python.exe -X showrefcount tt.py
1 1  1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
[0 refs, 0 blocks]

[corona10]

LGTM.

Please submit the reproducible code if possible it can reduce the reviewer's time and help understanding context :)
I know your patch is correct but reproducible code makes sure what your patch fixes exactly.

[miss-islington]

Thanks @kumaraditya303 for the PR, and @corona10 for merging it 🌮🎉.. I'm working now to backport this PR to: 3.11.
🐍🍒⛏🤖

[bedevere-bot]

GH-95049 is a backport of this pull request to the 3.11 branch.

[corona10]

I will merge backport patch after my dinner :)