Add `limit=` keyword to `int()` and `str()` functions to avoid contention on global `PYTHONINTMAXSTRDIGITS`

Enhancement

With the mitigations for CVE-2020-10735 in place, there's tension between:

• security-minded folks which need to protect against DoS attacks
• scientific (and other) use-cases which need to process large numbers in some form

Those are both vying for the one setting (PYTHONINTMAXSTRDIGITS) that controls this, which is unfortunately global state.

Since several such libraries are often loaded in the same runtime, this makes it very tricky to balance.

The idea is simply to add a limit=None keyword to int() and str(), which defaults to PYTHONINTMAXSTRDIGITS if None, but can be set to any other value. Setting it to 0 (for example, or -1) would make it unlimited.

Pitch

This would not solve all the fallout right away, but it would provide a sane path forward, where the default can be safe, and callsites can explicitly opt into saying "I want this call to have a different limit".

In particular, it would not require global state to balance the divergent needs of different parts of the ecosystem anymore.

In case a spectacular new conversion algorithm comes along, the default could eventually be set to infinite again, but the API wouldn't stand out as useless/harmful, but still serve a clear purpose, and so this proposal is independent of future string algorithms or default limits.

Previous discussion

https://discuss.python.org/t/int-str-conversions-broken-in-latest-python-bugfix-releases/18889/
and many more places linked from there

Commencez par la proposition liée sur discuss.python.org et la discussion qu’elle référence, puis examinez comment CPython applique actuellement PYTHONINTMAXSTRDIGITS aux conversions int() et str(). L’issue ne nomme aucun fichier source ni aucun test ; le comportement de l’API, le périmètre de l’implémentation et la couverture devraient donc être définis d’un commun accord avant de commencer le travail.