[JVM] Creation of Bag objects broken

[usev6]

The Problem

A lot of spectests for Bag fail on the JVM backend.

The breakage can be seen with a simple command (which should print Bag(42)):

$ ./rakudo-j -e 'say bag(42)'
java.lang.RuntimeException: No such attribute '$!value' for this object
  in block <unit> at -e line 1
$ ./rakudo-j --ll-exception -e 'say bag(42)'
java.lang.RuntimeException: No such attribute '$!value' for this object
  in ADD-ITERATOR-TO-BAG (gen/jvm/CORE.c.setting:10734)
  in create-from-iterator (gen/jvm/CORE.c.setting:60275)
  in dispatch:<!> (gen/jvm/CORE.c.setting:2083)
  in new (gen/jvm/CORE.c.setting:60288)
  in new (gen/jvm/CORE.c.setting:1175)
  in bag (gen/jvm/CORE.c.setting:61844)
  in <unit> (-e:1)
  in <unit-outer> (-e:1)
  in eval (gen/jvm/stage2/NQPHLL.nqp:1216)
  in <anon> (gen/jvm/stage2/NQPHLL.nqp:1328)
  in command_eval (gen/jvm/stage2/NQPHLL.nqp:1325)
  in command_eval (gen/jvm/Compiler.nqp:119)
  in command_line (gen/jvm/stage2/NQPHLL.nqp:1309)
  in MAIN (gen/jvm/rakudo.nqp:133)
  in <mainline> (gen/jvm/rakudo.nqp:121)
  in <anon> (gen/jvm/rakudo.nqp)

The problem was introduced a couple of days agow with 7c0a556e9c. With that commit reverted things work as expected.

I think the following code is a golf of the underlying problem:

$ ./rakudo-m -e 'use nqp; my $pair := nqp::create(Pair); nqp::bindattr($pair,Pair,"\$!value",nqp::add_i(nqp::getattr(($pair := Pair.new(42,0)),Pair, "\$!value"),1)); say $pair'
42 => 1
$ ./rakudo-j -e 'use nqp; my $pair := nqp::create(Pair); nqp::bindattr($pair,Pair,"\$!value",nqp::add_i(nqp::getattr(($pair := Pair.new(42,0)),Pair,"\$!value"),1)); say $pair'
42 => 0

Please note, that I'm using my $pair := nqp::create(Pair); instead of just my $pair; as it is done in https://github.com/rakudo/rakudo/blob/776f1a626d/src/core.c/Rakudo/QuantHash.pm6#L579
The latter fails with the failure mode from the original example:

$ ./rakudo-j -e 'use nqp; my $pair; nqp::bindattr($pair,Pair,"\$!value",nqp::add_i(nqp::getattr(($pair := Pair.new(42,0)),Pair,"\$!value"),1)); say $pair'
java.lang.RuntimeException: No such attribute '$!value' for this object
  in block <unit> at -e line 1

So AFAICS, the order of execution differs between the two backends. On MoarVM the binding to $pair (within nqp::getattr) seems to be executed first and the ǹqp::bindattr runs afterward -- overwriting $!value with 1.
On the JVM the nqp::bindattr seems to be executed first and the binding to $pair (within nqp::getattr) overwrites ``$!valuewith0```.

Environment

• Compiler version (perl6 -v or raku -v): Welcome to Rakudo(tm) v2021.03-12-gb4367a8f9.
  Implementing the Raku(tm) programming language v6.d.
  Built on JVM.

[lizmat]

Oops, I guess Github being a little overzealous. I think it fixes the issue on the JVM backend, but am not 100% sure as no way to test.

[usev6]

Spectest looks good again. \o/
I'm closing this issue.