Reinstate failglob behaviour for most commands

[anordal]

Expand globs to zero arguments (nullglob) only for set, for and count.

The warning about failing globs, and setting the accompanying $status,
now happens regardless of mode, interactive or not.

It is assumed that the above commands are the common cases where
nullglob behaviour is desirable.
More importantly, doing this with set is a real feature enabler,
since the resulting empty array can be passed on to any command.

The previous behaviour was actually all nullglob (since commit
cab115c), but this was undocumented;
the failglob warning was still printed in interactive mode,
and the documentation was bragging about failglob behaviour.

[faho]

Without running this:

I think I'd be okay with this behavior (it's basically what I suggested).

At the very least you'll want to look at test5, which fails because you also print the error message for case *something* - you'll need to adjust it (and yes, I do think we'll want it for case as well because that's a major source of errors because it overloads the "*")

Otherwise it's worth discussing, especially:

• Do we want a way to mark a certain glob as nullable? Like a "nullglob" command?
• Do we want a way to figure out after the fact if a glob failed?

Of course both of these could be left out - we could tell people to use variables if they want that.

And one thing I'll have to test is if the error message is suppressable.

[anordal]

Thanks for the review! Yes, I read your posts on the matter, and the last one inspired me to try if this was possible.

I'm looking into the case failure, but I think it's bedtime for me now.

[anordal]

Now I get it, case *something*, from beforehand, actually checks if a file exists that matches the glob. This is also documented.

This commit accidentally "fixes" it by disallowing the syntax — what I should do is modify the expected result.

[anordal]

test5 is now fixed by expecting this error output:

No matches for wildcard '*ee*'.
fish:  case *ee*
            ^

Now, I'm struggling to reproduce the "bind.expect" failure in the clang test.

[ridiculousfish]

The bind.expect failure is a gnarly timing issue due to ASAN (I think!), though we've also had trouble reproing it so I may easily be wrong. Thanks for being conscientious and you can ignore that one.

[faho]

The test is actually unnecessary - count will return 1 if given no arguments (and 0 otherwise), so if count $foos >/dev/null works just as well. As would if set -q foos[1].

[faho]

I actually can't find anything wrong with the behavior here.

@ridiculousfish: I'd like to merge this and believe we can add any other nullglob command or syntax later. Okay?

[ridiculousfish]

Reviewed, looks most excellent. Merged with rebase as 62b76b2 . Thanks!!

[anordal]

Thank you guys, I'm delighted.