From @labster

multi infix​:<eq> (Int $a, Int $b) { $a * $b == 0 };
say 1 eq 0, 1 ne 0, 1 !eq 0;
TrueTrueFalse

1 ne 0 looks like it's calling .Str here, or otherwise it would follow my
strange definition of integer <eq>.

if 'a' ne 'a'|'b'|'c' { say True }
True
if 'a' !eq 'a'|'b'|'c' { say True }
Nil
'a' ne 'a'|'b'|'c'
False

I suspect these bugs are related.

From @moritz

On Fri Apr 12 01​:59​:19 2013, labster wrote​:

multi infix​:<eq> (Int $a, Int $b) { $a * $b == 0 };
say 1 eq 0, 1 ne 0, 1 !eq 0;
TrueTrueFalse

1 ne 0 looks like it's calling .Str here, or otherwise it would follow my
strange definition of integer <eq>.

Sure, you didn't define an infix​:<ne> for integers, which is why the
built-in one is used, which stringifies. As specified. (Afaict the specs
don't contain any wording about infix ne forwarding to infix eq ones,
which would make infix ne slower than necessary. So you must override
infix​:<ne> yourself).

if 'a' ne 'a'|'b'|'c' { say True }
True
if 'a' !eq 'a'|'b'|'c' { say True }
Nil
'a' ne 'a'|'b'|'c'
False

I'm pretty sure both should be the as as

!('a' eq 'a'|'b'|'c')

and thus return False.

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

From @moritz

It's an optimization bug​:

$ cat junc.pl
if 'a' ne 'a'|'b'|'c' { say True }
$ ./perl6 --optimize=0 junc.pl
$ ./perl6 junc.pl
True
$

From @timo

My optimization didn't check the arguments of the operator to see if they
accept Any or Mu. Thus it autothreaded the ne although it should have
passed the junction as an argument.

There now is a bit of test code in roast and the bug is fixed.

Thanks!

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