# New Ticket Created by "Carl Mäsak"
# Please include the string: [perl #75376]
# in the subject line of all future correspondence about this issue.
# <URL: http://rt.perl.org/rt3/Ticket/Display.html?id=75376 >
<masak> rakudo: my class A {}; our class B is A {} # via jnthn++
<p6eval> rakudo 982e8e: OUTPUT«Lexical 'A' not foundcurrent instr.:
'perl6;B;!class_init_11' pc 476 (EVAL_1:212)»
* masak submits rakudobug
The thinking about this issue can be found in the backlog:
00:41 jnthn my class Foo { }; our class Bar is Foo { } # oh noes.
00:42 This explodes now and I don't have a good way to fix it.
00:42 Because right now, my class Foo gets installed into the lexpad
just like we install subs, etc.
00:42 pmichaud protolexpads would help, or we need a way to invoke a
sub such that we keep the lexpad that has already been created for it
00:42 jnthn Whereas our Bar gets set up in the loadinit
00:42 Which happens...before Foo
00:42 pmichaud instead of creating a new one
00:43 anyway, I have two knee-jerk reactions
00:43 (1) making things happen more like BEGIN brings us closer to
the right way of doing things (i.e., I'd think that using phasers is
better than :loadinit)
00:44 (2) I don't see that moving :loadinit to phasers really solves
the problem (but perhaps you've thought of something I haven't there)
00:45 jnthn pmichaud: You may be right on (2). At least it's not the
only thing that will need doing.
00:45 pmichaud: Maybe I should spend the tuits on protolexpad-ish
stuff instead.
00:45 pmichaud the problem is really Parrot's model of lexpads
00:46 jnthn Aye, it's not quite what we want.
00:46 pmichaud from a Parrot perspective, the symbol Foo doesn't get
bound until the scope containing Foo is invoked
00:46 i.e., the lexpad entry for Foo doesn't exist until the outer
scope is invoked
00:47 it's easier to imagine with something like:
00:47 sub xyz() { my class Foo { }; our class Bar is Foo { } }
00:47 jnthn Well, it affects the mainline too
00:48 But yes, that's pretty much it.
00:48 pmichaud the mainline is just another scope, yes.
00:48 and from a Parrot perspective, the lexical symbols in the
mainline don't really exist before the mainline is invoked
00:48 sorear in a perl6 model, lexpads are cloned
00:48 we can do that with a custom lexinfo
00:48 I think
00:49 jnthn sorear: Last time I worked through what it might take, I
think I maybe ended up with us needing a custom Sub too, but it may be
avoidable.
00:49 pmichaud let me look again
00:49 just a sec
00:49 jnthn I'd *really* like to see if we couldn't put this mess to
bed at least somewhat ahead of R* though. It's kinda...embarassing.
00:49 pmichaud Personally, I'm not too concerned if one can't mix
lexical and packaged scoped classes in R*
00:50 maybe I should be... but I'm not.
00:50 jnthn pmichaud: I didn't mean that specific example. I meant the
whole plethora of issues.
00:50 like
00:50 rakudo: my role Foo { method lol { say 42 } }; Foo.lol # fine
00:50 p6eval rakudo edb448: OUTPUT«42»
00:50 jnthn rakudo: our role Foo { method lol { say 42 } }; Foo.lol # not
00:50 p6eval rakudo edb448: OUTPUT«42»
00:50 jnthn ...oh.
00:51 oh, I know what's not
00:51 spinclad would one want a wall to prevent such mixed times, or
delay class Bar till Foo is ready?
00:51 pugssvn r30828 | sorear++ | [viv] Implement basic signature translation
00:51 jnthn rakudo: role Foo { method lol { say 42 } }; class Bar does
Foo { }; Bar.lol
00:51 p6eval rakudo edb448: OUTPUT«Could not find sub &saycurrent
instr.: 'lol' pc 563 (EVAL_1:217)»
00:51 jnthn That one.
00:51 pmichaud technically, Foo can be ready at BEGIN time also, as
long as we have a way to bind it into the lexpad at runtime
00:52 plobsing joined #perl6
00:52 jnthn This happens because when you do a role, it has to go and
clone the role body.
00:52 And thus its methods
00:52 (for type parameters to be captured)
00:52 And if that happens from within a callchain involving loadinit,
we appear to be in a mess.
00:53 pmichaud: Yes, that would be ideal.
00:54 * lue can only hope to aspire to your level of knowledge of
rakudo/perl6 someday.
00:55 jnthn lue: Needs time, minor loss of sanity, beer and hanging
around here soaking up the knowledge of @other. :-)
00:55 jaldhar joined #perl6
00:55 spinclad lue: you need not just hope; you may fully aspire
00:55 pmichaud sorear: custom lexinfo is a good idea... we'd also need
custom lexpads
00:55 jnthn (Or tht's how I did it. :-))
00:56 The only reason I think we may be able to get away with Sub
also being custom is that lexpad when created gets the lexinfo.
00:56 lue [and a well-detailed spec :P]
00:56 jnthn If we can hang the "pernament stuff" in the lexinfo perhaps...
00:56 ...then the lexpad on creation and copy those things into
itself.
00:56 s/and/can/
00:56 pmichaud well, the Sub doesn't actually use the lexinfo directly
-- it just passes the lexinfo to the lexpad when it's created
00:56 jnthn pmichaud: Yes.
00:57 pmichaud so, it'd have to be a custom lexpad to be able to do
something different with the custom lexinfo
00:57 jnthn pmichaud: My worry was more that we'd need something to
trigger the clone of the stuff we already knew.
00:57 pmichaud oh, I think that would just be OurLexPad.new that does
that
00:57 i.e., our custom LexPad would have a custom init_pmc
00:57 jnthn pmichaud: But I suspect between a custom lexinfo and
custom lexpad, we don't need a custom sub.
00:58 pmichaud it would be tricky because the LexPad doesn't learn
about its context until a subsequent step after it's been created
00:58 so we'd have to create the LexPad, and then trigger any cloning
as part of the set_pointer
00:59 i.e., when we attach a context to the LexPad, *then* it could
start populating the context with the items from the LexInfo
00:59 if (!PMC_IS_NULL(sub->lex_info)) {
00:59 Parrot_pcc_set_lex_pad(INTERP, context,
Parrot_pmc_new_init(INTERP,
00:59 Parrot_get_ctx_HLL_type(INTERP, enum_class_LexPad),
00:59 sub->lex_info));
00:59 VTABLE_set_pointer(INTERP, Parrot_pcc_get_lex_pad(INTERP,
context), context);
00:59 }
00:59 sorear lue: try aspiring to my level of understanding
00:59 jnthn pmichaud: ah yes
00:59 But that's do-able.
01:00 If messy.
01:00 pmichaud not sure I'd want to do this for R*, though.
01:00 tough call.
01:00 jnthn Worth prototyping in a branch perhaps?
01:00 pmichaud were you able to move our existing BEGINs to phasers
yet?
01:00 jnthn No, that's high on my todo list.
01:01 pmichaud or does this affect that directly?
01:01 jnthn No, it's not related to that.
01:01 pmichaud I think I'd want to see BEGINs as true phasers first.
01:01 jnthn I brought this up because I just ran into it yet again.
01:01 In a sense, the proto lex pad thing may well be the real fix we
need
01:01 And doing everything else as phasers could come later
01:01 I'm not sure.
01:01 pmichaud I think moving BEGIN to be phasers will tell us a lot,
though.
