bit shift broken at 32 bits

[p6rt]

Migrated from rt.perl.org#131278 (status was 'resolved')

Searchable as RT131278$

[p6rt]

From curt@tilmes.org

<curt_> m​: my $i = -0x8000000000000000; say ($i div 2**12) == ($i +> 12) ??
'good' !! 'bad';
<camelia> rakudo-moar 6bb1b5​: OUTPUT​: «good␤»
<curt_> m​: my $i = -0x8000000000000000; say ($i div 2**37) == ($i +> 37) ??
'good' !! 'bad';
<camelia> rakudo-moar 6bb1b5​: OUTPUT​: «bad␤»
<curt_> m​: my $i = -0x8000000000000000; say ($i div 2**23) == ($i +> 23) ??
'good' !! 'bad';
<camelia> rakudo-moar 6bb1b5​: OUTPUT​: «good␤»
<curt_> m​: my $i = -0x8000000000000000; say ($i div 2**42) == ($i +> 42) ??
'good' !! 'bad';
<camelia> rakudo-moar 6bb1b5​: OUTPUT​: «bad␤»
<TimToady> loks like a bug
<TimToady> at the 32-bit boundary
<TimToady> m​: my $i = -0x8000000000000000; say ($i +> 31)
<camelia> rakudo-moar 6bb1b5​: OUTPUT​: «-4294967296␤»
<TimToady> m​: my $i = -0x8000000000000000; say ($i +> 32)
<camelia> rakudo-moar 6bb1b5​: OUTPUT​: «-1␤»
<TimToady> please report bug

[p6rt]

From @zoffixznet

On Tue, 09 May 2017 09​:49​:37 -0700, curt@​tilmes.org wrote​:

<curt_> m​: my $i = -0x8000000000000000; say ($i div 2**12) == ($i +> 12) ??
'good' !! 'bad';
<camelia> rakudo-moar 6bb1b5​: OUTPUT​: «good␤»
<curt_> m​: my $i = -0x8000000000000000; say ($i div 2**37) == ($i +> 37) ??
'good' !! 'bad';
<camelia> rakudo-moar 6bb1b5​: OUTPUT​: «bad␤»
<curt_> m​: my $i = -0x8000000000000000; say ($i div 2**23) == ($i +> 23) ??
'good' !! 'bad';
<camelia> rakudo-moar 6bb1b5​: OUTPUT​: «good␤»
<curt_> m​: my $i = -0x8000000000000000; say ($i div 2**42) == ($i +> 42) ??
'good' !! 'bad';
<camelia> rakudo-moar 6bb1b5​: OUTPUT​: «bad␤»
<TimToady> loks like a bug
<TimToady> at the 32-bit boundary
<TimToady> m​: my $i = -0x8000000000000000; say ($i +> 31)
<camelia> rakudo-moar 6bb1b5​: OUTPUT​: «-4294967296␤»
<TimToady> m​: my $i = -0x8000000000000000; say ($i +> 32)
<camelia> rakudo-moar 6bb1b5​: OUTPUT​: «-1␤»
<TimToady> please report bug

Looking at the source code, this behaviour is due to a kludge to fix RT#​126942[^1],
according to which long bit shifts of negative integers should always produce -1

I'm not sure where the value of 32 bits came into play, other than it's the value
where RT#​126942's bug came around.

[1] https://rt-archive.perl.org/perl6/Ticket/Display.html?id=126942
[2] https://github.com/rakudo/rakudo/blob/6435933/src/core/Int.pm#L412

[p6rt]

The RT System itself - Status changed from 'new' to 'open'

[p6rt]

From @zoffixznet

Found where the problem was. Sent a PR[^1] to MoarVM to fix it and will commit Rakudo's patch and tests once that's merged.

[1] MoarVM/MoarVM#593

[p6rt]

From @zoffixznet

Thanks for the report. This is now fixed.

Fix​:
  MoarVM/MoarVM@362277b79f
  Raku/nqp@9763728215
  rakudo/rakudo@66e8e72c02
  rakudo/rakudo@ef29bb9f41

Tests​:
  Raku/roast@02d1f9d0f6
  rakudo/rakudo@6409ee582e

[p6rt]

@zoffixznet - Status changed from 'open' to 'resolved'