bpo-42197: Don't create f_locals dictionary unless we actually need it.

[markshannon]

This tiny change seems to double the speed of coverage.py https://bugs.python.org/issue42197#msg411210

Original implementation by @fabioz. I've dropped his modifications to the sys docs, as the f_locals field is now internal, so should not be accessed directly: https://github.com/python/cpython/blob/main/Doc/whatsnew/3.11.rst

https://bugs.python.org/issue42197

[gvanrossum]

@markshannon Is there a typo in the bpo number? bpo-42917 seems to be about the size limit of the block stack.

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @markshannon for commit 281305b 🤖

If you want to schedule another build, you need to add the ":hammer: test-with-buildbots" label again.

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @markshannon for commit 0e452a1 🤖

If you want to schedule another build, you need to add the ":hammer: test-with-buildbots" label again.

[ncoghlan]

The breaking change here is technically in scope for the two fast locals PEPs, but given that 3.11 is breaking that API anyway by making frame objects private, I agree it makes sense to just go ahead and make this change independently of those.

+1 for PyFrame_GetLocals() as the replacement public API.