Strange behaviour of <<!=>>

[lizmat]

The Problem

From https://stackoverflow.com/questions/71690554/strange-behavior-of-a-hyperoperator

say 0 != 0;                 # False

my @a = 0, 0, 0;
say @a «==» @a;             # [True True True]
say @a «!=» @a;             # [True True True] <--- why?
say @a «!==» @a;            # [False False False]

Expected Behavior

say @a «!=» @a;             # [False False False]

Actual Behavior

say @a «!=» @a;             # [True True True] <--- why?

This is on Rakudo HEAD.

[lizmat]

The problem appears to be in the codegen of «!=», as a direct call to METAOP_HYPER yields the correct result:

my @a = 0, 0, 0;
say METAOP_HYPER(&infix:<<!=>>)(@a,@a);  # [False False False]

This seems to be caused by the decision at https://github.com/rakudo/rakudo/blob/master/src/Perl6/Actions.nqp#L8121 to go for the given AST, rather than generating the proper call to &infix:<!=>.

Making that code always generate the call to &infix:<!=> makes the above code produce the correct result. But breaks spec tests, so there is more to it.

[lizmat]

Turns out it was all wrong to begin with:

my @a = 1,2,3;
my @b = 4,5,6;
say @a >>!=<< @b;
say @a;  # [False False False]
say @b;  # [4 5 6]

So it actually created a meta-op using assignment. As can be seen with:

say (1,2) >>!=<< (3,4);
# Cannot modify an immutable Int (1)

@jnthn++ for noting. Fixed for now with the PR, should have special attention in the RakuAST branch in the future.

[raiph]

cf Raku/old-issue-tracker#3322