gh-134768: Fix definition of mt_continue_should_break()

[thesamesam]

In 121ed71, mt_continue_should_break was changed to be guarded by Py_DEBUG, but it's used in compress_mt_continue_lock_held with just assert, so it needs to be available when NDEBUG is undefined too.

Py_DEBUG implies NDEBUG is undefined, so we can check just that.

Fixes: 121ed71

• Issue: Build failure on main and latest Python 3.14 beta (missing mt_continue_should_break) #134768

[emmatyping]

Just verifying my understanding, because "NDEBUG" is rather confusingly named it seems: the failure case is you compile in debug (not --enable-optimizations) which defines Py_DEBUG, but also enable assertions via --enable-assertions (which defines NDEBUG)?

[mgorny]

No. The failure is that you compile regular (which doesn't set Py_DEBUG) but also --with-assertions (which implies -UNDEBUG).

[thesamesam]

Other way around:

• --with-pydebug (Py_DEBUG) => -UNDEBUG -DPy_DEBUG -> -DPy_DEBUG (works on main)
• --with-assertions (-UNDEBUG) => nothing (needs this PR because assert is not compiled out)
• nothing (-DNDEBUG) => -DNDEBUG (works on main, because the assert is compiled out)

It's really stupid and it's to do with how assert is defined. By default, assert does something. If NDEBUG is defined, it doesn't. It's very common to set NDEBUG for performance and CPython does that unless you pass --with-assertions.

[emmatyping]

Okay, makes sense, thanks for the explanations

[mgorny]

Could we merge this, please? It just bit me again as I was about to recheck a failing assertion on main.

[miss-islington-app]

Thanks @thesamesam for the PR, and @gpshead for merging it 🌮🎉.. I'm working now to backport this PR to: 3.14.
🐍🍒⛏🤖

[bedevere-app]

GH-134916 is a backport of this pull request to the 3.14 branch.