[perl #123776] Binding a variable at BEGIN time doesn't stick aruond for runtime

On Tue, 30 Jan 2018 19:03:13 -0800, [email protected] wrote:
> On Tue, 30 Jan 2018 15:08:38 -0800, [email protected] wrote:
> > On Fri, 01 Dec 2017 12:09:05 -0800, [email protected] wrote:
> > > Still reproducible (2017.11, HEAD(5929887)), but is supposed to
> > > work?
> >
> > Don't see any reason why that'd be questionable.
> >
> > > On 2015-02-09 17:43:52, [email protected] wrote:
> > > > m: my $a; BEGIN { $a := 1; say $a; }; say $a;
> > > > rakudo-{parrot,moar} 0cb22e: OUTPUT«1␤(Any)␤»
> > > >
> > > > m: my $a; BEGIN { $a = 1; say $a; }; say $a; # works with
> > > > assignment
> > > > rakudo-{parrot,moar} 0cb22e: OUTPUT«1␤1␤»
> >
> >
> > Another manifestation of the same bug:
> >
> > <Zoffix__> m: BEGIN { my $a1 := 42; say $a1; }
> > <camelia> rakudo-moar deffe54b8: OUTPUT: «42␤»
> > <Zoffix__> m: BEGIN { my $a1 = 42; say $a1; }
> > <camelia> rakudo-moar deffe54b8: OUTPUT: «42␤»
> > <Zoffix__> m: BEGIN my $a1 = 42; say $a1;
> > <camelia> rakudo-moar deffe54b8: OUTPUT: «42␤»
> > <Zoffix__> m: BEGIN my $a1 := 42; say $a1;
> > <camelia> rakudo-moar deffe54b8: OUTPUT: «(Mu)␤»
> 
> 
> 
> Tracked this to Perl6::World.compile_in_context but leaving it there,
> as it's getting too hard.
> Need to gain a few levels first.
> 
> The comment in the method reads:
>    # Create outer lexical contexts with all symbols visible. Maybe
>    # we can be a bit smarter here some day. But for now we just make a
>    # single frame and copy all the visible things into it.
> 
> So I imagine the binding gets made to this faked out block and never
> propagates back to the source where all the lexicals were copied from.


More comments: https://irclog.perlgeek.de/perl6/2018-02-02#i_15769787

15:11 	jnthn 	Zoffix: Yeah. Lexpads are immutable, and when we compile a BEGIN we obviously haven't finished the outer block yet, so we have to fake something up. But what? Containers are cloned from a "prototype", and the fake lexpad contains the static container. Thus what's actually happening is that on entry to the scope post-compilation, the Scalar is cloned complete with the value set in the prototype.
15:12 		Of course, binding doesn't have that level of indirection so there's no way it can work.
15:12 		We should perhaps detect it at compile time and say it can't work. Or...we could try diffing the scope after each BEGIN block and trying to somehow propagate bindings.