[bug-gawk] Strange matching behavior

[Davide Brini]

I'm not sure what's going on here. Per POSIX, "]" can appear inside a
bracket expression if it's the first character following the opening [.
Indeed, this works (and always has, AFAIR):

$ printf '%s\n' '[' ']' '.' '*' '$' | awk '/[][]/ { print "<" $0 ">" }'
<[>
<]>

However, if I add some other characters to the list ("." and "*" here), gawk
fails:

$ printf '%s\n' '[' ']' '.' '*' '$' | awk '/[][.*]/ { print "<" $0 ">" }'
awk: cmd. line:1: error: Unmatched [ or [^: /[][.*]/

The gawk documentation says:

"To include one of the characters '\', ']', '-', or '^' in a bracket
expression, put a '\' in front of it."

I think that, as said, that escape should not be necessary, but anyway,
let's try it:

$ printf '%s\n' '[' ']' '.' '*' '$' | awk '/[\][.*]/ { print "<" $0 ">" }'
awk: cmd. line:1: error: Unmatched [ or [^: /[\][.*]/


But if I add another "$" (without escaping anything), it works again:

$ printf '%s\n' '[' ']' '.' '*' '$' | awk '/[][$.*]/ { print "<" $0 ">" }'
<[>
<]>
<.>
<*>
<$>

In fact, it works if *any* character is added to the expression instead of
"$"; however, it has to be exactly in that spot. If it's added after the
"." or after the "*" it does not work; it must be after the "[".

Finally, if I escape *both* square brackets, it works in the problematic
case too:

$ printf '%s\n' '[' ']' '.' '*' '$' | awk '/[\]\[.*]/ { print "<" $0 ">" }'
<[>
<]>
<.>
<*>

$ gawk --version
GNU Awk 4.1.1, API: 1.1 (GNU MPFR 3.1.2-p10, GNU MP 6.0.0)
...

-- 
D.