This fix doesn't work, I get a warning using GCC 10 on Fedora 32:

gcc -pthread -c -Wno-unused-result -Wsign-compare -g -Og -Wall -O0   -std=c99 -Wextra -Wno-unused-result -Wno-unused-parameter -Wno-missing-field-initializers -Werror=implicit-function-declaration -fvisibility=hidden  -I./Include/internal  -I. -I./Include    -DPy_BUILD_CORE -o Objects/frameobject.o Objects/frameobject.c
In file included from ./Include/Python.h:61,
                 from Objects/frameobject.c:3:
Objects/frameobject.c: In function 'frame_setlineno':
./Include/pyport.h:313:35: warning: comparison of integer expressions of different signedness: 'long int' and 'long unsigned int' [-Wsign-compare]
  313 |     (assert((WIDE)(NARROW)(VALUE) == (VALUE)), (NARROW)(VALUE))
      |                                   ^~
Objects/frameobject.c:400:15: note: in expansion of macro 'Py_SAFE_DOWNCAST'
  400 |     int len = Py_SAFE_DOWNCAST(
      |               ^~~~~~~~~~~~~~~~

2^30 opcodes in a single code object is extremely unlikely

I'm not comfortable with relying on such assumption that things "should not happen in practice". I would prefer to implement a concrete runtime check in PyCode_NewWithPosOnlyArgs() to reject co_code if (size_t)PyBytes_GET_SIZE(f->f_code->co_code)/sizeof(_Py_CODEUNIT) > (size_t)INT_MAX is true. With that, you can replace Py_SAFE_DOWNCAST with (int) cast with a comment referring to PyCode_NewWithPosOnlyArgs() check.

Hm, I got this warning too when I use gcc9.3 on Red Hat 4.8.5-36.

Aah sorry about that. I was trying to go with a narrow fix and didn't want to touch too many files. It looks like there is an assertion here https://github.com/python/cpython/blob/master/Python/ceval.c#L1328
but I agree with vstinner that this should probably just be a runtime check in codeobject. Let me open up a PR with that.

No problem :-) Follow-up PR: PR #20628.