Fix braces

[faho]

Description

This is a double-whammy to remove some of the annoyances from brace/bracket-expansion (the code calls it "brackets", I've only ever heard "braces" elsewhere).

It changes a literal {} to expand to itself, which fixes issue #1109, and makes find -exec nicer to use.

It also changes it so successive commas in braces are seen as successive empty items, rather than alternating between an empty item and a literal ",". So x{,,,}y is no longer seen as xy x,y xy (because the second "," is seen as a literal "," rather than a separator), but as "xy xy xy xy". That fixes #3002. (To get the literal comma, it can be \\ escaped or quoted still)

Both of these are in the "theoretically backwards-incompatible, but unlikely to break stuff" category. The former especially will only break code that does useless stuff - currently a {} could be removed everywhere without any changes, so why have it at all?

TODOs:

• Changes to fish usage are reflected in user documentation/manpages - the first has been, the second behavior was never documented AFAICT.
• Tests have been added for regressions fixed
• User-visible changes noted in CHANGELOG.md

[faho]

Note: The {,,,} behavior dates back to 149594f - the initial revision. The comment has been added later on, but I have no idea if there's been any prior discussion on it that justified the feature. To me, it just seems wrong(tm).

[mqudsi]

That is awesome, thanks for tackling this! I agree the alternating comma behavior is weird, I'm presuming \, still specifies an explicit comma, which would be the obvious approach.

Any chance you could implement #3802 while you're at it?

[faho]

I'm presuming , still specifies an explicit comma, which would be the obvious approach.

Yes. Which is why I'm oblivious as to why the current behavior was a thing in the first place, and why this is rather unlikely to cause any issues in practice.

Any chance you could implement #3802 while you're at it?

No. That would require touching the parser/tokenizer (since expand.cpp receives BRACKET_END for the space IIUC), and I do not understand that yet.

[faho]

Merged as 55ebf44..d67b4d6 (that last was because of some weirdness, possibly related to git rerere).