On Fri, 21 Apr 2017 05:44:48 -0700, consult...@jnthn.net wrote:
> This is golfed from a File::Ignore regression. This reproduces it:
>
> grammar Parser {
> token TOP { <matcher>+ }
> proto token matcher { * }
> token matcher:sym<[]> { '[' <( <-[\]]>+ )> ']' }
> token matcher:sym<lit> { <-[/*?[]>+ }
> }
> class RuleCompiler {
> # Simply commenting out this causes the matcher:sym<lit> to get the
> .txt it should. With
> # this, it gets ].txt instead.
> method matcher:sym<[]>($/) { }
> }
> say Parser.parse('[AB].txt', :actions(RuleCompiler));
>
> Commenting out the action method makes the final match tree correct.
> Without it, the match tree ends up with the second matcher wrongly
> capturing the ].
>
> I've had trouble bisecting it due to unbuildable commits; the log is here:
>
> https://gist.github.com/Whateverable/119eb26d8ca245bf59ec05bc014c9fc2
>
> It was fine in 2017.03 and broken in 2017.04, perhaps around the time of
> the uncurse merge. I'm somewhat suspecting that, since it unified Match and
> Cursor. This code may be a tricky edge case for the unification since it
> uses the <( and )> constructs, which make the captured text and the text
> matched by the cursor differ.
Fixed by TimToady++ and test added in S05-metasyntax/angle-brackets.t.
