From @masak

<tadzik> r​: for Range.new(0, 1) -> $p { when $p { say $_.perl; say $_.path } }
<p6eval> rakudo 39087a​: OUTPUT«use of uninitialized variable $a of
type Any in numeric context in block <anon> at
/tmp/bhcZACNkRT​:1␤␤Null PMC access in find_method('perl') [...]
<tadzik> there we are
* masak submits rakudobug
<pmichaud> ...why is that a bug, ooc?
<masak> pmichaud​: Null PMC access
<tadzik> NPA
<pmichaud> oh, I see.
<masak> r​: for ^1 -> $p { when $p { .foo } }
<p6eval> rakudo 39087a​: OUTPUT«use of uninitialized variable $a of
type Any in numeric context in block <anon> at
/tmp/HzMxDYTCaC​:1␤␤Null PMC access in find_method('foo') [...]
<pmichaud> oh, it's the .ACCEPTS bug
<masak> oh?
<pmichaud> sure, it's the when $p that is attempting to convert
Any into a number
<pmichaud> in order to smart match it against $p
<masak> the interesting thing is that the Range, the 'when' and the
$_.foo call are all irreducible parts of triggering the bug.
<masak> remove any of them, and the Null PMC Access disappears.
<pmichaud> r​: for (0,1) -> $p { when $p { .foo } }
<p6eval> rakudo 39087a​: OUTPUT«use of uninitialized variable $a of
type Any in numeric context in block <anon> at
/tmp/VgCIK_kuoP​:1␤␤Null PMC access in find_method('foo') [...]
<masak> maybe not. :)
<pmichaud> r​: for 0 -> $p { when $p { .foo } }
<masak> ok, the Range is not essential, then.
<p6eval> rakudo 39087a​: OUTPUT«use of uninitialized variable $a of
type Any in numeric context in block <anon> at
/tmp/2BcvX6lmnC​:1␤␤Null PMC access in find_method('foo') [...]
<masak> huh.
<masak> tadzik​: I'm pretty sure we tested that.
<tadzik> masak​: yeah, same. Odd
<pmichaud> given 0 -> $p { when $p { .foo } }
<pmichaud> r​: given 0 -> $p { when $p { .foo } }
<p6eval> rakudo 39087a​: OUTPUT«use of uninitialized variable $a of
type Any in numeric context in block <anon> at
/tmp/SJ1JbHQSej​:1␤␤Null PMC access in find_method('foo') [...]
<pmichaud> r​: (-> $p { when $p { .foo } }).(0)
<p6eval> rakudo 39087a​: OUTPUT«use of uninitialized variable $a of
type Any in numeric context in block <anon> at /tmp/5zdWF1O_Fa​:1␤␤No
such method 'foo' for invocant of type 'Any' [...]
<masak> there. it disappeared.
<masak> I think iteration is involved somehow, anyway.
<pmichaud> well, not with 'given'
<pmichaud> there's no iteration with the .given
<masak> hm, troo.
<masak> so, why does it trigger with 'for' and 'given', but not with
that last one, with a block?
<pmichaud> r​: my $ps = -> $p { when $p { .foo } }; $ps(0);
<p6eval> rakudo 39087a​: OUTPUT«use of uninitialized variable $a of
type Any in numeric context in block <anon> at
/tmp/Ap9xU9dVVo​:1NLNLNo such method 'foo' for invocant of type 'Any'
[...]
<pmichaud> r​: 0.map( -> $p { when $p { .foo } }).eager
<p6eval> rakudo 39087a​: OUTPUT<<use of uninitialized variable $a of
type Any in numeric context in block <anon> at
/tmp/JRkifpk737​:1NLNLNo such method 'foo' for invocant of type 'Any'
[...]
<pmichaud> the NPA has to be something in the way the closures are
being built when part of 'for' or 'given'
<pmichaud> which means it's a bug in <xblock>
<pmichaud> (in Actions.pm)
<pmichaud> here's another, then
<pmichaud> r​: if 1 -> $p { when $p { .foo } }
<p6eval> rakudo 39087a​: OUTPUT<<use of uninitialized variable $a of
type Any in numeric context in block <anon> at
/tmp/C8Mba0rmGT​:1NLNL>>
<pmichaud> hmm.
<pmichaud> okay, so, the "unintialized variable $a..." is coming from
an ACCEPTS call
<pmichaud> I'm guessing it's this one​:
<pmichaud> src/core/Numeric.pm​:4​: multi method ACCEPTS(Numeric​:D​: $a) {
<pmichaud> The NPA is because the pointy block is somehow getting the
wrong outer $_
<masak> ah.
<pmichaud> but only when being used as part of 'for' or 'given' (so
far... could be other structures)
<pmichaud> r​: $_ = Bool; if 1 -> $p { when $p { .say } }
<p6eval> rakudo 39087a​: OUTPUT<<use of uninitialized variable $a of
type Bool in numeric context in block <anon> at
/tmp/amepJyEJYf​:1NLNL>>
<pmichaud> r​: Bool.say
<p6eval> rakudo 39087a​: OUTPUT<<Bool()NL>>
<pmichaud> *there*
<masak> \o/
<pmichaud> the .say got a $_ that isn't the one with the Bool
<pmichaud> even though the one with the Bool is the one that was used
for the smartmatching 'when'
<masak> I'm lost, but I'm including this in the report anyway :P
<pmichaud> please do, it's pretty well golfed.
<masak> pmichaud++

From @pmichaud

Never mind the 'if' example above; it doesn't really evoke the error.

This does, however​:

  > $_ = 0; for ^1 -> $p { when $p { say 'matched'; $_.say } }
  matched
  Null PMC access in find_method('say')
  > $_ = 1; for ^2 -> $p { when $p { say 'matched'; $_.say } }
  matched
  Null PMC access in find_method('say')
  >

I'm guessing the problem has something to do with the way that 'when' is
topicalizing the evaluation of the $p expression and not restoring $_
properly.

Pm

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

From @pmichaud

I'm guessing the problem has something to do with the way that 'when'
is
topicalizing the evaluation of the $p expression and not restoring $_
properly.

This guess is wrong; there's currently no topicalization taking
place around $p . (See --target=pir for proof.)

So, it's something about the way $_ is being left unset in the when block
itself. Perhaps a problem with the signature of the when block (failing
to set $_ = OUTER​::$_), it not being closured properly, or something like
that.

Pm

From @masak

<zjmarlow> r​: my $line; for < one two three >.values -> $line { if True
{ .say } }
<p6eval> rakudo bf472b​: OUTPUT«Null PMC access [...]
<masak> huh.
<masak> we may have this in RT already, actually.
<masak> r​: for 1 -> $ { if 1 { .say } }
<p6eval> rakudo bf472b​: OUTPUT«Null PMC access [...]
<masak> yes, here​: https://rt-archive.perl.org/perl6/Ticket/Display.html?id=113904
<masak> though this particular case has an 'if', not a 'when'.
<masak> r​: $_ = 42; .say
<p6eval> rakudo bf472b​: OUTPUT«42␤»
<masak> r​: $_ = 42; for 1 { .say }
<p6eval> rakudo bf472b​: OUTPUT«1␤»
<masak> r​: $_ = 42; for 1 { if 1 { .say } }
<p6eval> rakudo bf472b​: OUTPUT«1␤»
<masak> r​: $_ = 42; for 1 -> $p { .say }
<p6eval> rakudo bf472b​: OUTPUT«42␤»
<masak> r​: $_ = 42; for 1 -> $p { if 1 { .say } }
<p6eval> rakudo bf472b​: OUTPUT«Null PMC access [...]
<masak> this one illustrates it fairly well. I'll paste it into the
ticket.

From @coke

One of the nom regressions (S02-magicals/dollar-underscore.t) is caused by
this NPA​:

19​:12 < [Coke]> r​: $_ = 1; my $tracker = ''; for 12 -> $a { if $_ == 1 {
  $tracker ~= "1 : $_|"; $_ = 2; } }
19​:12 <+camelia> rakudo de2080​: OUTPUT«Null PMC access in
  find_method('Stringy')␤ in block at /tmp/HWQMO6B4wa​:1␤
in
  method reify at src/gen/CORE.setting​:5737␤ in method
reify at
  src/gen/CORE.setting​:5632␤ in method gimme at
  src/gen/CORE.setting​:6053␤ in method eager at
  src/gen/CORE.setting​:6032␤ …

--
Will "Coke" Coleda

From @FROGGS

it only comes if there is a condition within a for-loop with a named
iterator

<FROGGS> r​: $_ = 1; for 12 -> $a { if 1 { "$_" } }
<camelia> rakudo de2080​: OUTPUT«Null PMC access in
find_method('Stringy')␤ in block at /tmp/Bify_dvuSE​:1␤ in method
reify at src/gen/CORE.setting​:5737␤ in method reify at
src/gen/CORE.setting​:5632␤ in method gimme at
src/gen/CORE.setting​:6053␤ in method eager at src/gen/CORE.setting​:6032␤ …

<FROGGS> r​: $_ = 1; for 12 -> $a { unless 0 { "$_" } }
<camelia> rakudo de2080​: OUTPUT«Null PMC access in
find_method('Stringy')␤ in block at /tmp/2A9OMuXLSv​:1␤ in method
reify at src/gen/CORE.setting​:5737␤ in method reify at
src/gen/CORE.setting​:5632␤ in method gimme at
src/gen/CORE.setting​:6053␤ in method eager at src/gen/CORE.setting​:6032␤ …

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

From @FROGGS

$_ is not an SMO​:

r​: for 12 -> $a { if 1 { say $_.WHAT } }
<camelia> rakudo de2080​: OUTPUT«Can only use get_what on a
SixModelObject␤ in block at /tmp/SdPsnKQ9XU​:1␤ in method reify at
src/gen/CORE.setting​:5737␤ in method reify at
src/gen/CORE.setting​:5632␤ in method gimme at
src/gen/CORE.setting​:6053␤ in method eager at src/gen/CORE.setting​:6032␤ …

From @usev6

This Null PMC Access error does not occur any longer. Also the SixModelObject error is gone. (Output of following commands is identical for Moar, Parrot and JVM.)

$ perl6 -e '$_ = 1; for ^2 -> $p { when $p { say "matched"; $_.say } }'
matched
1

$ perl6-m -e '$_ = 42; for 1 -> $p { if 1 { .say } }'
42

$ perl6 -e 'for 12 -> $a { if 1 { say $_.WHAT } }'
Nil

I extended the existing tests in S02-magicals/dollar-underscore.t a bit with the following commit​: Raku/roast@0e579ae7c7

I'm closing this ticket now.

@usev6 - Status changed from 'open' to 'resolved'