set -l in `if` scope should be consistent with new for scope

[mqudsi]

Since #1935 has been committed, the following behavior changes from "enhancement" to "bug" since there's now an inconsistency in fish's syntax:

~> if not set -l output (somecmd && echo "success")
                      echo "command failed!" 1>&2
      end
      echo "Output is: "$output
Output is: 

[mqudsi]

@faho @zanchey are there other keywords that would need this change?

[faho]

I'd imagine while and possibly begin. Of the other keywords, command and exec can't possibly set a local variable, the booleans already work like this, and the others don't take a command.

Logically, either begin is special or

begin set -l foo bar
# do stuff
end

is different from

begin
set -l foo bar
# do stuff
end

[mqudsi]

Shouldn't the first form just be illegal, like a command after end? It's too syntactically ambiguous, a ; should be required after it if trying to execute on the same line (which is how I've always used it).

[faho]

In #986 we decided to just keep it as-is, as

it seems relatively harmless

[mqudsi]

You are a treasure trove of fish history, @faho 👍

[mqudsi]

I think #4443 is the opposite of this, a scope is being created for eval.

[ridiculousfish]

I went ahead and fixed this for if and while but left begin alone, since there's no reason why you would set a variable inside a begin block if you wanted it outside.

[mqudsi]

Awesome, thanks @ridiculousfish