From @timo

17​:50 < timotimo> nqp​: say("\r\na" ~~ /<?[\w]>"\r\na"/);
17​:50 <+camelia> nqp-moarvm​: OUTPUT«Substring length (-6) cannot be negative␤Str␤»
17​:50 <+camelia> ..nqp-jvm, nqp-parrot​: OUTPUT«␤»
17​:50 < timotimo> nqp​: say("\r\na" ~~ /"\r\na"/);
17​:50 <+camelia> nqp-moarvm, nqp-jvm, nqp-parrot​: OUTPUT«

the implementation of CCLASS_NEWLINE doesn't have a condition for the addition to pos if it matched a \r\n.

I'll fix this bug on a train ride today.

From @timo

actually, the regex just has to have \n, but it b0rks with \w as well, because that mixes it in.

18​:03 < timotimo> nqp​: say("\r\na" ~~ /<?[\n]>"\r\na"/);
18​:03 <+camelia> nqp-moarvm​: OUTPUT«Substring length (-6) cannot be negative␤Str␤»
18​:03 <+camelia> ..nqp-jvm, nqp-parrot​: OUTPUT«␤»
18​:03 < timotimo> nqp​: say("\r\na" ~~ /"\r\na"/);
18​:03 <+camelia> nqp-moarvm, nqp-jvm, nqp-parrot​: OUTPUT«

From @timo

solved the bug, but it needs to get a test.

ideally, the test would land in qregex/somewhere, but that's giving me trouble with \r\n and different levels of escaping ...

From @FROGGS

Added test to roast instead of nqp​: Raku/roast@c82bbd55ff

@FROGGS - Status changed from 'new' to 'resolved'