From @masak

<moritz> nom​: class A { eval 'has $.x' }; say A.new(x => 3).x
<p6eval> nom 9c6aed​: OUTPUT«===SORRY!===␤A cannot have attributes at
line 1, near ""␤»
<moritz> nom​: class A { BEGIN eval 'has $.x' }; say A.new(x => 3).x
<p6eval> nom 9c6aed​: OUTPUT«===SORRY!===␤A cannot have attributes at
line 1, near ""␤»
<moritz> why not?
<jnthn> moritz​: Probably that eval doesn't know it's in a class
definition or some such
<jnthn> moritz​: e.g. it's treated like a little bit of mainline
<jnthn> moritz​: I guess you can argue that the BEGIN case there should work
<moritz> jnthn​: but the error message knows that it's in A
<jnthn> moritz​: no, the error message is busted
<jnthn> The first case should certainly not work
<jnthn> The second one...maybe.
<jnthn> So the question is...
<jnthn> In eval, what do we treat as the current package?
<jnthn> masak​: The "A" in the error messsage is not referring to the clas A
<jnthn> Call it B in the code to see :)
<masak> jnthn​: aha, it's an article ;)
<jnthn> nom​: class B { BEGIN eval 'has $.x' }
<p6eval> nom 9c6aed​: OUTPUT«===SORRY!===␤A cannot have attributes at
line 1, near ""␤»
<jnthn> See the double space? :)
<jnthn> It doesn't know what sort of thing it is in :)
<jnthn> nom​: has $.x
<p6eval> nom 9c6aed​: OUTPUT«===SORRY!===␤A cannot have attributes at
line 1, near ""␤»
<jnthn> Right. :)
<jnthn> eval is doing the wrong kind of thing about current package.
* masak submits rakudobug
<jnthn> Yeah, we should probably make that one work.
<jnthn> I'm in the middle of re-doing the repr API at the moment
though, so won't get to that for a day or two.

From @coke

On Sat Oct 29 13​:09​:29 2011, masak wrote​:

<moritz> nom​: class A { eval 'has $.x' }; say A.new(x => 3).x
<p6eval> nom 9c6aed​: OUTPUT«===SORRY!===␤A cannot have attributes at
line 1, near ""␤»
<moritz> nom​: class A { BEGIN eval 'has $.x' }; say A.new(x => 3).x
<p6eval> nom 9c6aed​: OUTPUT«===SORRY!===␤A cannot have attributes at
line 1, near ""␤»
<moritz> why not?
<jnthn> moritz​: Probably that eval doesn't know it's in a class
definition or some such
<jnthn> moritz​: e.g. it's treated like a little bit of mainline
<jnthn> moritz​: I guess you can argue that the BEGIN case there should work
<moritz> jnthn​: but the error message knows that it's in A
<jnthn> moritz​: no, the error message is busted
<jnthn> The first case should certainly not work
<jnthn> The second one...maybe.
<jnthn> So the question is...
<jnthn> In eval, what do we treat as the current package?
<jnthn> masak​: The "A" in the error messsage is not referring to the clas A
<jnthn> Call it B in the code to see :)
<masak> jnthn​: aha, it's an article ;)
<jnthn> nom​: class B { BEGIN eval 'has $.x' }
<p6eval> nom 9c6aed​: OUTPUT«===SORRY!===␤A cannot have attributes at
line 1, near ""␤»
<jnthn> See the double space? :)
<jnthn> It doesn't know what sort of thing it is in :)
<jnthn> nom​: has $.x
<p6eval> nom 9c6aed​: OUTPUT«===SORRY!===␤A cannot have attributes at
line 1, near ""␤»
<jnthn> Right. :)
<jnthn> eval is doing the wrong kind of thing about current package.
* masak submits rakudobug
<jnthn> Yeah, we should probably make that one work.
<jnthn> I'm in the middle of re-doing the repr API at the moment
though, so won't get to that for a day or two.

Behavior has changed a bit.

This now gives an awesome error and can be closed with tests​:

03​:16 < [Coke]> nom​: has $.x
03​:17 <+camelia> rakudo-parrot e8cea1, rakudo-jvm e8cea1, rakudo-moar e8cea1​:
  OUTPUT«===SORRY!=== Error while compiling /tmp/tmpfile␤You
  cannot declare attribute '$.x' here; maybe you'd like a class
  or a role?␤at /tmp/tmpfile​:1␤------> has $.x⏏<EOL>␤ …»

However, the examples from earlier now give errors about non-rakudo objects, or methods we weren't asking about, e.g.​:

03​:16 < [Coke]> nom​: class A { BEGIN EVAL 'has $.x' }; say A.new(x => 3).x
03​:16 <+camelia> rakudo-jvm e8cea1​: OUTPUT«(timeout)»
03​:16 <+camelia> ..rakudo-parrot e8cea1​: OUTPUT«===SORRY!===␤No such method
  'name' for invocant of type 'NQPMu'␤»
03​:16 <+camelia> ..rakudo-moar e8cea1​: OUTPUT«===SORRY!===␤Cannot find method
  'name'␤»

--
Will "Coke" Coleda

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

From @usev6

All of this looks good now (results are identical on Moar, Parrot and JVM)​:

$ perl6 -e 'has $.x'
===SORRY!=== Error while compiling -e
You cannot declare attribute '$.x' here; maybe you'd like a class or a role?
at -e​:1
------> has $.x⏏<EOL>
  expecting any of​:
  scoped declarator
  constraint

$ perl6 -e 'class A { BEGIN EVAL q[has $.x] }; say A.new(x => 3).x'
3

$ perl6-m -e 'class A { EVAL q[has $.x] }; say A.new(x => 3).x'
No such method 'x' for invocant of type 'A'
  in block <unit> at -e​:1

I added tests to S12-attributes/undeclared.t and S12-attributes/class.t with commit Raku/roast@7f8180122c

I'm closing this ticket now.

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