gh-126072: do not add None to co_consts if no docstring

[xuantengh]

Hi @markshannon this is my attempt to implement #126072. Initially the value of CO_HAS_DOCSTRING is set as 0x0040, but I found it has been used by the CO_NOFREE in inspect.

cpython/Lib/inspect.py

Lines 409 to 411 in 85799f1

	co_flags bitmap: 1=optimized | 2=newlocals | 4=*arg | 8=**arg
	| 16=nested | 32=generator | 64=nofree | 128=coroutine
	| 256=iterable_coroutine | 512=async_generator

And I've tested it with the following code: See the test cases in changed test files

I've also updated the related documentation, but maybe there is something I missed. Please help me improve this PR, thank you!

• Issue: The method for storing docstrings in code objects is awkward and prevents optimizations. #126072

---

📚 Documentation preview 📚: https://cpython-previews--126101.org.readthedocs.build/

[bedevere-app]

Most changes to Python require a NEWS entry. Add one using the blurb_it web app or the blurb command-line tool.

If this change has little impact on Python users, wait for a maintainer to apply the skip news label instead.

[JelleZijlstra]

Please add some tests, probably to test_code.py.

You'll also probably need to change the __doc__ descriptor on function objects so it continues to work correctly.

[xuantengh]

Maybe the optimization pass should be updated as the hypothesis may not hold:

cpython/Python/flowgraph.c

Lines 2117 to 2119 in 9b14083

	// The first constant may be docstring; keep it always.
	index_map[0] = 0;

The generated bytecodes for the same function are different before and after this PR.

[markshannon]

FYI, we are in the process of removing small ints from the co_consts tuple: #125972

Can you change the ints in your tests to strings. Doing so will also help test that non-docstring strings do not get interpreted as docstrings.

[markshannon]

Maybe the optimization pass should be updated as the hypothesis may not hold:

cpython/Python/flowgraph.c

Lines 2117 to 2119 in 9b14083

	// The first constant may be docstring; keep it always.
	index_map[0] = 0;

The generated bytecodes for the same function are different before and after this PR.

There are a few improvements we can make to the bytecode compiler once this PR is done.
It is probably best to do those in other PRs, once this is working.

[xuantengh]

It is probably best to do those in other PRs, once this is working.

Yeah, what I concern is whether the constant removal optimization will lead to a unexpected results when the first element in co_consts is no longer None if there is no docstring.

[xuantengh]

Currently, as the const removal optimization unconditionally keeps the first element in consts, None will present in co_consts even if co has no docstring if there is only one const (i.e., None itself).

[willingc]

Looks good. One small edit for the language reference.

[markshannon]

Thanks for doing this.

This was one of those little things that's been bothering me for a while.
Glad to see it fixed

[xuantengh]

Update doc inspect.rst as well.

[willingc]

One small comment on inspect.rst.

[markshannon]

The UI shows 2 jobs still in progress. But they have completed successfully 🤷‍♂️

[xuantengh]

Thanks for your review, it's my first PR merged into CPython! And I'm also interested in the following related optimizations.

[iritkatriel]

Thanks for your review, it's my first PR merged into CPython! And I'm also interested in the following related optimizations.

Congratulations! in your next PR, add your name to Misc/ACKS (we should have done it here, but too late now).