On Sun Mar 20 06:39:39 2016, elizabeth wrote:
> ==========================
> my $a = 14;
> while (True) {
> my $z = (2..13).first(-> $x { !($a %% $x) });
> last if (!$z);
> $a += 14
> }
> say $a
> ==========================
>
> The above code “verbatim” segfaults on OS X and Linux on HEAD.
>
> Replacing either the (True) by True, or (!$z) by !$z, or adding an
> additional say after the loop, will make the code succeed.
>
>
> Looking at the output of —target=optimize between (!$z) and !$z, I see
> only this significant difference:
>
> - - QAST::Op(callstatic &prefix:<!>) :BY<modifier_expr W>
> :statement_id<?> :WANTED !
> + - QAST::Op(callstatic &prefix:<!>) :BY<modifier_expr W> :WANTED !
>
> I would have sorta expected that the () would not make a difference
> for the generated code at all, but apparently it does?
>
> Also, disabling spesh or JIT does not seem to make a difference.
>
>
> In any case, I thought this would be a nice thing for the right
> people. :-)
>
> Discussion at: http://irclog.perlgeek.de/perl6/2016-03-20#i_12212186
>
Was a GC invariant violation that occasionally led to memory corruption (thus
why it was easy to make it vanish with silly changes). Actually a regression
within the last month or two. Fixed it, and add a test to
integration/weird-errors.t.
/jnthn
