`[[[[abc]` generate "incomplete token" error (regression from #12903)

Problem

In 4.8.1:

$ echo [[[[abc]
[[[[abc]

In master (commit abaf54f0f8e2f331e4c3038c1cfb21cfbdcc7a04 from PR #12903):

$ echo [[[[abc]
fish: Unexpected end of string, square brackets do not match
echo [[[[abc]
       ^

Notes:

• This however never worked:

  fish复制

  $ echo [[[[abc  # no closing ]
  fish: Unexpected end of string, square brackets do not match
  echo [[[[abc
          ^

  So in a way, the error is just applied more consistently.

• This is related to #2485 and boils down to: When is [ special? When is it just a regular character?

  • In $var1[$var2[... and (...)[..., it is obviously special, and the first is what the PR was fixing.
  • In [[..., it is obviously not (although arguably, it could be reserved for character ranges/classes like in globbing regex/...)
  • In a[b[..., it's unclear. Usually it has no meaning:
    • currently set does rely on it being special (set var[1 2] ...), at least for the 1st level of nesting
    • but echo -e \\e[4mfoo fails because of the missing of ], even though this ought to be valid

  And if [ is some times special and some times not, what happens with mixed cases, e.g. $var[1 b[...] 2] (which ] should close $var[? Technically, it should be the 1st ], since the 2nd [ would have no special meaning, but a user might expect the second).

Solution

I think the regression is the lesser evil, at least for now. The PR fixed a valid use case. And if one needs unbalanced brackets, they can use quotes to get around the error. The quotes are already necessary when there are no ] at all.

Maybe a line in the changelog?