`plistlib.dumps()` raises `TypeError` when `skipkeys=True` and dict has mixed key types

[stefanzetzsche]

Issue

The documentation for dump() states:

When sort_keys is true (the default) the keys for dictionaries will be
written to the plist in sorted order, otherwise they will be written in
the iteration order of the dictionary.

When skipkeys is false (the default) the function raises :exc:TypeError
when a key of a dictionary is not a string, otherwise such keys are skipped.

plistlib.dumps() with skipkeys=True and sort_keys=True (the default) currently raises TypeError when the dict contains both string and non-string keys. The expected behavior per the documentation is: non-string keys are skipped, and the remaining (string) keys are written in sorted order.

Reproducer

import plistlib

d = {1: 'a', 'z': 'b', 'a': 'c'}

plistlib.dumps(d, skipkeys=True)
# Raises: TypeError: '<' not supported between instances of 'str' and 'int'
# Expected: valid plist with {"a": "c", "z": "b"}

Both FMT_XML and FMT_BINARY formats raise the same error.

CPython versions tested on:

CPython main branch

Operating systems tested on:

No response

Linked PRs

• gh-145856: Fix plistlib.dumps() skipkeys behavior with mixed key types #145867
• gh-145856: Fix plistlib.dumps() skipkeys behavior with mixed key types #150109

[stefanzetzsche]

I believe this is an implementation error rather than a documentation issue. The fix is likely to reverse the order of operations: filter non-string keys first, then sort. Happy to file a PR if confirmed a bug.

[stefanzetzsche]

For context, json has a similar issue: #69643.