tarfile.TarFile.errors is initialized for some reason as None even though it really only accepts str

[kbaikov]

Bug report

Bug description:

See the exact line here:

cpython/Lib/tarfile.py

Line 1687 in ee57221

errors = None # Error handler for unicode conversion.

After that the errors is used only as string in encode/decode. And is initialized in the dunder init with default errors="surrogateescape".

So perhaps it would be a good idea to use errors="" in #L1687 ? or directly with the default errors="surrogateescape" ?

This was discovered as part of the typeshed, where tarfile.TarFile.errors was added into the allowed list since mypy complained: https://github.com/python/typeshed/blob/2ccc53bb6716463680fcf1d204642c46299b7f88/stdlib/%40tests/stubtest_allowlists/common.txt#L46C1-L46C16

I will add a PR here and to the typeshed if this gets approved.

CPython versions tested on:

3.13

Operating systems tested on:

Windows

[sobolevn]

I don't think that this is really a bug. It more looks like a refactoring. Can it affect existing users?

[rruuaanng]

I think it is initialized to None and there seems to be nothing wrong with it, after all, int and str are a class.

[picnixz]

I think it is initialized to None and there seems to be nothing wrong with it, after all, int and str are a class.

This is not the issue. The issue concerns typing and the fact that errors must be typed str | None in this case, instead of just str. As for whether having a None value makes sense at runtime except for default-detection, the question remains and may affect existing users as Nikita hinted.