GeezerVison

I believe only two lines are needed.

See also #137100, which is also about the text that follows.

Some descriptions have separate signature for no argument, others merge it with a signature with one argument. See for example dict which could be written as dict(mapping_or_iterable=(), **kwargs), but is written as three semantically different signatures. I tried to be more consistent and chose the former variant. But I have no such strong preference.

I prefer fewer lines, and approve of the changes.

I'm not sure I agree here, source=b'' makes it less clear to me that source also accepts e.g. iterables of integers, buffer-protocol, etc. I would suggest:

.. class:: bytearray()
           bytearray(source, /)
           bytearray(source, /, encoding, errors='strict')

Note I have suggested to annotate 'source' as positional-only -- I think this makes more sense to users than writing e.g. bytearray(source=my_numpy_array). My IDE & type-checkers also indicates that source=... is an error, as it is annotated as positional-only in typeshed.

Some descriptions have separate signature for no argument, others merge it with a signature with one argument. See for example dict which could be written as dict(mapping_or_iterable=(), **kwargs), but is written as three semantically different signatures. I tried to be more consistent and chose the former variant. But I have no such strong preference.

It's technically true that you can pass just the encoding or errors kwargs (like str(encoding='ebcdic')), but it's not practically useful, since you just get an empty string whatever encoding you pass, and I fear it's confusing to users to put this here.

The old docs here feel clearer even if they're not as precise.

We should be technically correct.

Several years ago I proposed to deprecate this case, but my proposition was rejected. If it is worth to continue supporting this in code, it is worth to document it. Note that values of encoding and errors are ignored:

>>> str(encoding='spam', errors='ham')
''

Either approach could work but I like the way it's in the current PR; it's better to have fewer distinct signatures.