From jswitzer@gmail.com

I ran into a bug where I couldn't refer to 'self' when it logically made
sense I could​:

(15​:33​:15) s1n​: rakudo​: class A { has %.H = (a => -> $x { self.c($x) });
method c($c) { say "got $c" }; method test { %.H<a>.("foo").say; }; };
A.new().test();
(15​:33​:20) p6eval​: rakudo cddb16​: OUTPUT«Lexical 'self' not found␤current
instr.​: 'parrot;A;_block27' pc 243 (EVAL_24​:111)␤»
(15​:33​:54) masak​: s1n​: in fact, 'self' is not defined at the place you're
using it.
(15​:34​:04) masak​: s1n​: only use it from within methods.
(15​:34​:36) moritz_​: but why? `self' in class should just refer to the class
(formerly proto object), no?
(15​:34​:47) masak​: hm.
(15​:34​:51) s1n​: it exists, it should
(15​:34​:56) masak​: rakudo​: class A { self }
(15​:35​:00) p6eval​: rakudo cddb16​: OUTPUT«Lexical 'self' not found␤current
instr.​: 'parrot;A;_block20' pc 94 (EVAL_18​:66)␤»

I am sure there are far simpler test cases. jnth++ agrees that people may be
burned more often than not with the current behavior of self.

-Jason "s1n" Switzer

From @pmichaud

On Sun May 03 13​:51​:11 2009, s1n wrote​:

I ran into a bug where I couldn't refer to 'self' when it logically made
sense I could​:

(15​:33​:15) s1n​: rakudo​: class A { has %.H = (a => -> $x { self.c($x) });

I don't see anything in the spec that indicates that the body of the
class declaration has a lexical 'self', or that it should be pre-set to
the type object (formerly protoobject). AFAICT 'self' is only defined
for method bodies, and it's the invocant of the method.

If that's not the case, we probably need a spec update. In the
meantime, I'm changing the title of this ticket to more accurately
reflect the question at hand.

Also, S12 indicates that C<self> is a function, whereas we currently
implement it as a lexical. If it needs to be a function, we can
probably arrange that. (And perhaps it's a contextual lookup of some sort.)

Thanks,

Pm

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

From @pmichaud

21​:00 <TimToady> pmichaud​: note that for the self ticket, the phrase in
question is a has initializer, which runs in BUILD context, and hence
does have an obvious instance self, so this isn't really about binding
self to the class
21​:04 <pmichaud> TimToady​: okay, thanks.
21​:04 <pmichaud> So we definitely can't leave self as being a plain
lexical, unless we're making use of 'lift' somehow.
21​:05 <pmichaud> or otherwise attach the context to self... which we
have to do anyway. I'll have to think about it a bit more.
21​:05 <jnthn> The thing on the RHS of the = is made a closure already;
we just need to make it an anonymous method.
21​:05 <pmichaud> That works for me.
21​:05 <jnthn> (and remember to pass self in too)
21​:06 <pmichaud> well, the closure has to be invoked on the newly built
object, I suspect.
21​:06 <jnthn> yes
21​:07 <pmichaud> okay, I'll update the ticket.

From @pmichaud

21​:08 <pmichaud> in the general case, though, is 'self' available in the
body of a class declaration?
21​:08 <pmichaud> (as opposed to within initializers, as the ticket case
shows)
21​:08 <jnthn> No.
21​:08 <pmichaud> (ticket updated)
21​:08 <TimToady> I don't see what it buys us that $?CLASS does give us
better
21​:09 <TimToady> or some reasonable fascimile
21​:09 <TimToady> self in the class seems to be confusion waiting to happen
21​:09 <pmichaud> good for me
21​:09 <pmichaud> more ticket updating. :-)

From @jnthn

On Mon May 04 14​:07​:56 2009, pmichaud wrote​:

21​:00 <TimToady> pmichaud​: note that for the self ticket, the phrase in
question is a has initializer, which runs in BUILD context, and hence
does have an obvious instance self, so this isn't really about binding
self to the class
21​:04 <pmichaud> TimToady​: okay, thanks.
21​:04 <pmichaud> So we definitely can't leave self as being a plain
lexical, unless we're making use of 'lift' somehow.
21​:05 <pmichaud> or otherwise attach the context to self... which we
have to do anyway. I'll have to think about it a bit more.
21​:05 <jnthn> The thing on the RHS of the = is made a closure already;
we just need to make it an anonymous method.
21​:05 <pmichaud> That works for me.
21​:05 <jnthn> (and remember to pass self in too)
21​:06 <pmichaud> well, the closure has to be invoked on the newly built
object, I suspect.
21​:06 <jnthn> yes
21​:07 <pmichaud> okay, I'll update the ticket.

Updated Rakudo to work this way in git 6229131. So now these work​:

class Foo { has $.a = 1; has $.b = 2; has $.c = $.a + $.b;}
say Foo.new.c
3

And​:

class Bar { has $.r = rand }
say Bar.new.r for 1..3;
0.470490443377884
0.6400637382559
0.878367274158286

Did some unfudging and additions to S12-attributes/instance.t.

Thanks,

Jonathan

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