Also worth taking a look at https://rt.perl.org/Ticket/Display.html?id=126569
On 2017-09-29 14:06:58, b...@abrij.org wrote:
> On Sun, 28 May 2017 00:08:18 -0700, sml...@gmail.com wrote:
> > This bug is still present in
> >
> > This is Rakudo version 2017.05-134-g0c5fe56cc built on MoarVM version
> > 2017.05-25-g62bc54e9
> > implementing Perl 6.c.
>
>
> OK, keeping in mind that I have entirely, absolutely no idea what I am
> doing...
>
> I figured out that the block inside the string inside the parens is
> being annotated
> to a statement ID one greater than that of the whole statement_mod,
> which means
> it doesn't get "migrated". I also figured out that it seems to be the
> case that
> no blocks from inside the predicate of the for/given appear in the
> list considered
> for migration. I don't know if the annotation is correct or not, and
> I don't know
> if there is some contorted syntax that could put a block in the list
> of considered
> items for migration which would be erroneously migrated, but simply
> changing
> make_topic_block_ref to also grab all in_stmt_mod blocks with a
> statement ID higher
> than the one it is handed seems to pass all spectests and "fix" this
> problem.
>
> Resulting in the following one-character patch (drumroll please):
>
> diff --git a/src/Perl6/Actions.nqp b/src/Perl6/Actions.nqp
> index 36fa59b2b..6576f61b7 100644
> --- a/src/Perl6/Actions.nqp
> +++ b/src/Perl6/Actions.nqp
> @@ -8953,7 +8953,7 @@ class Perl6::Actions is HLL::Actions does
> STDActions {
> $*W.pop_lexpad();
> if nqp::defined($migrate_stmt_id) {
> migrate_blocks($*W.cur_lexpad(), $block, -> $b {
> - !$b.ann('in_stmt_mod') && ($b.ann('statement_id') //
> -1) == $migrate_stmt_id
> + !$b.ann('in_stmt_mod') && ($b.ann('statement_id') //
> -1) >= $migrate_stmt_id
> });
> }
> ($*W.cur_lexpad())[0].push($block);
>
> Now, all that said, I cannot begin to properly emphasize the massive
> extent of the heck of which I do not know here.
>
> So, this is either the right fix, or a very wrong fix which might just
> break things in a way
> roast test writers are just not twisted and deviant enough to think
> of, or a fix that doesn't
> catch derivative cases, or...
>
> Other cases tested with this patch:
>
> $ perl6 -e 'say ("{$_}") given <aa bb>'
> aa bb
> $ perl6 -e 'say ("{$_}") if True given <aa bb>'
> aa bb
> $ perl6 -e 'say ("{$_}") if ({True}) given <aa bb>'
> aa bb
> $ perl6 -e 'say ("{$_}") if ("{True}") given <aa bb>'
> aa bb
> $ perl6 -e 'say ("{$_}") if "{True}" given <aa bb>'
> aa bb
> $ perl6 -e 'say ("{$_}") if "{so $_}" given <aa bb>'
> aa bb
> $ perl6 -e 'say ("{$_}") if ("{$_ ~~ /bb/ ?? q|1| !! q|| }") for <aa
> bb>'
> bb
> $ perl6 -e 'say ("{$_}") for ("{$_}" for <aa bb>)'
> aa
> bb
> $ perl6 -e 'my $a = 13; say ("{$_}") for {my $a = 42; ("{$a}" for <aa
> bb>)}()'
> 42
> 42
