allow predicate functions to lstrip etc. and default to stripping unicode whitespace

[simonbyrne]

This is similar to #27232, except that the predicate function is the second argument. It does not require any deprecations.

Fixes #27211.

[ararslan]

The predicate function should always go first per our conventions, documented in the style guide in the manual.

[KristofferC]

It is a "general rule" not a "must". Having the function argument second here seems much more natural to me.

[fredrikekre]

[ararslan]

I disagree wholeheartedly; it seems very unnatural to me.

[ararslan]

I don't think we should deviate from our argument order conventions for this case.

[StefanKarpinski]

I agree with @ararslan here: predicate function as the first argument, character sets as the second; we can also allow a predicate as either the first or second argument as the user prefers since a string argument isn't callable.

[simonbyrne]

Note that we do allow predicate function as a second argument to split/rsplit (this was not changed in #27252).

[ararslan]

That should probably be fixed.

[Keno]

I'm also with @ararslan. It doesn't really bother me that the predicate and the chars are in different places in the argument list.

[JeffBezanson]

I'm also OK with adding a predicate-as-first-argument version and just leaving the others.

[StefanKarpinski]

@simonbyrne, your PR, your call. Just do whatever you see fit and merge it once CI passes.

[ararslan]

Can I make one last plea to put the function argument first?

[ararslan]

Seems there was broad support for at least allowing that.

[simonbyrne]

@ararslan I am sympathetic, and I have thought a lot about it, but my opinion is that:

1. keeping argument order somewhat consistent is preferable to adhering to the function first rule
2. if we really want to allow predicate function also as a first argument, we can add that later without breakage.
3. it is too late to change the argument order altogether, but it is possible to do that via deprecation without silent breakage (since we don't allow chars to be a string).

So my call is to merge this PR.

[simonbyrne]

Also, I got rid of the other uses of _default_delimiters.

[ararslan]

keeping argument order somewhat consistent is preferable to adhering to the function first rule

I very adamantly disagree with this, since the argument order is then not consistent. This will be a case of "functions always come first, except for this one weird case." We had that with finalizer and that was subsequently changed.

if we really want to allow predicate function also as a first argument, we can add that later without breakage

I can submit a PR to add it once this is merged, but it really makes more sense to me to just have lstrip(f, s) and lstrip(s, c) rather than lstrip(f, s), lstrip(s, c), and lstrip(s, f). But if that's unacceptable I'll just add lstrip(f, s).

I do appreciate your work here and that you've given a lot of thought to this--I don't want you to think otherwise.

[ararslan]

PR to fix the argument order: #27605

[JeffBezanson]

DelimitedFiles still attempts to import _default_delims (though doesn't seem to use it).