Cannot import an extension module if its path is not encodable in UTF-8

Bug report

Importing an extension module fails with UnicodeEncodeError if its path contains characters unencodable in UTF-8, e.g. surrogate escapes of a file name undecodable in the filesystem encoding.

$ .../nd\udcffwt/python -c "import _json"
Traceback (most recent call last):
  File "<string>", line 1, in <module>
UnicodeEncodeError: 'utf-8' codec can't encode character '\udcff' in position 45: surrogates not allowed

As a result, CPython cannot be run in a directory whose name contains undecodable bytes, and cannot even be built there -- make fails in checksharedmods. On 3.13 and 3.14 every import of a shared extension module from such a path fails.

The cause is the key of the extensions cache: hashtable_key_from_2_strings() in Python/import.c encodes the module file name with _PyUnicode_AsUTF8NoNUL(). It was added in gh-105699.

The raw content of the strings (UCS1, UCS2 or UCS4) can be used for the key instead of the UTF-8 encoding.

Linked PRs

• gh-155249

Commencez dans Python/import.c, au niveau de hashtable_key_from_2_strings(), où l’issue identifie l’encodage UTF-8 du nom de fichier du module d’extension comme étant la cause. Examinez la PR liée gh-155249 ainsi que l’échec de checksharedmods décrit dans le rapport. Le travail terminé doit inclure des imports et des builds réussis de modules d’extension depuis des chemins contenant des octets de nom de fichier indécodables.