struct: non-ASCII bytes format raises UnicodeEncodeError instead of struct.error on 3.13/3.14

[corona10]

Bug report

#153835 (backport of gh-85943) introduces a regression on 3.13 and 3.14. A bytes format with a non-ASCII byte now raises UnicodeEncodeError instead of struct.error:

>>> import struct
>>> struct.unpack(b'\xc5', b'')
UnicodeEncodeError: 'ascii' codec can't encode character '\udcc5' in position 0: ordinal not in range(128)

3.14.0 raises struct.error: bad char in struct format. main is not affected since gh-145850 already rejects non-ASCII formats with ValueError.

This aborts fuzz_struct_unpack, so CIFuzz fails on any 3.14 PR that touches C files, e.g. gh-158162: https://github.com/python/cpython/actions/runs/36136424084/job/108075564322

UnicodeEncodeError: 'ascii' codec can't encode character '\udcc5' in position 0: ordinal not in range(128)
==75== ERROR: libFuzzer: deadly signal
    #6 0x55d07f8c7884 in _run_fuzz /src/cpython3/Modules/_xxtestfuzz/fuzzer.c:575:9
    #7 0x55d07f8c7884 in LLVMFuzzerTestOneInput /src/cpython3/Modules/_xxtestfuzz/fuzzer.c:640:11
Base64: xQA=

Linked PRs

• [3.14] gh-158172: Preserve struct.error for non-ASCII bytes formats #158185

[corona10]

cc @serhiy-storchaka

[StanFromIreland]

More generally, for CIFuzz things we have #152211.