Different implementation for nqp::substr for Moar and [JVM] (negative offset)

[usev6]

The implementation of nqp::substr differs between Moar and JVM with regard to negative offsets:

bartolin    r: use nqp; say nqp::substr("bar", -1, 1)
camelia     rakudo-moar 054aca: OUTPUT«r␤»
camelia     ..rakudo-jvm 76b061: OUTPUT«java.lang.StringIndexOutOfBoundsException: String index out of range: -1␤  in block <unit> at <tmp> line 1␤␤»

For Moar there is a comment stating explicitely that negative offsets count from the end: https://github.com/MoarVM/MoarVM/blob/ee0b54e264ca3647f55ffd115238b72c7a0729f5/src/strings/ops.c#L290

What follows is some additional discussion from #perl6-dev:

AlexDaniel m: use nqp; say nqp::substr("bar", -2, 1) camelia rakudo-moar 054aca: OUTPUT«a␤» AlexDaniel m: use nqp; say nqp::substr("bar", -4, 1) camelia rakudo-moar 054aca: OUTPUT«␤» AlexDaniel m: use nqp; say nqp::substr("bar", -5, 1) camelia rakudo-moar 054aca: OUTPUT«Substring end (-1) cannot be less than 0␤ in block at line 1␤␤» AlexDaniel m: use nqp; say nqp::substr("bar", -4, 5) camelia rakudo-moar 054aca: OUTPUT«bar␤» camelia rakudo-moar 054aca: OUTPUT«Start argument to substr out of range. Is: -1, should be in 0..3; use *-1 if you want to index relative to the end␤ in block at line 1␤␤Actually thrown at:␤ in block at line 1␤␤» FROGGS bartolin: yes, please open said ticket bartolin will do perlpilot bartolin: be sure to include a transcript of what AlexDaniel did above ... even if I can accept that it should handle negative offsets, that second to last one is just crazy

I've stumbled above this problem when I looked a failing test for rakudo-j (from S04-statements/terminator.t):

bartolin    r: {my $x = 2;
camelia     rakudo-jvm 76b061: OUTPUT«===SORRY!===␤java.lang.StringIndexOutOfBoundsException: String index 
                out of range: -1␤»
camelia     ..rakudo-moar 054aca: OUTPUT«===SORRY!=== Error while compiling <tmp>␤Missing block␤at 
                <tmp>:1␤------> {my $x = 2;⏏<EOL>␤» 

The relevant code in Rakudo's src/Perl6/Grammar.nqp (https://github.com/rakudo/rakudo/blob/4dffef7a7ab035da03803d4a85dc3ba7cccb3647/src/Perl6/Grammar.nqp#L285) is supposed to examine the preceding character -- but it calls nqp::substring with a offset of -1. (I'll open a PR for Rakudo for this specific case.)

[usev6]

I've opened a second, more general issue for the implementation of nqp::substr: https://github.com/perl6/nqp/issues/533. Nevertheless, I'll leave this issue open for now.