From @masak

The `return` built-in doesn't take zero arguments, like it can in Perl
5. The attached patch fixes this.

$ cat return_undef.pl
sub f { return undef }
say f()
$ ./perl6 return_undef.pl

$ cat return.pl
sub f { return }
say f()
$ ./perl6 return.pl
too few arguments passed (0) - 1 params expected
current instr.​: 'f' pc 79 (EVAL_13​:35)
called from Sub '_block11' pc 17 (EVAL_13​:10)
called from Sub 'parrot;PCT​::HLLCompiler;eval' pc 806
(src/PCT/HLLCompiler.pir​:481)
called from Sub 'parrot;PCT​::HLLCompiler;evalfiles' pc 1088
(src/PCT/HLLCompiler.pir​:610)
called from Sub 'parrot;PCT​::HLLCompiler;command_line' pc 1267
(src/PCT/HLLCompiler.pir​:699)
called from Sub 'parrot;Perl6​::Compiler;main' pc 13257 (perl6.pir​:172)
perl6(73334) malloc​: *** error for object 0x28909b0​: double free
*** set a breakpoint in malloc_error_break to debug
Segmentation fault

$ patch -p0 < zero-arg-return.patch ; make perl6
[...]

$ ./perl6 return.pl

$

From @masak

Index: src/builtins/control.pir
===================================================================
--- src/builtins/control.pir	(revision 29090)
+++ src/builtins/control.pir	(working copy)
@@ -26,7 +26,13 @@
 .include 'except_types.pasm'
 
 .sub 'return'
-    .param pmc value
+    .param pmc value     :optional
+    .param int has_value :opt_flag
+
+    if has_value goto have_value
+    get_global $P1, "Failure"                                  
+    value = $P1."new"()                                         
+  have_value:
     $P0 = new 'Exception'
     $P0['_type'] = .CONTROL_RETURN
     setattribute $P0, 'payload', value

@pmichaud - Status changed from 'new' to 'open'

From @pmichaud

Applied (with one minor change) in r29143, thanks!

Note however that TimToady on #perl6 speculated [1] that an empty return
should return the Object prototype. I'm not sure the answer was
resolved completely in that thread, however, so we'll go with returning
'undef' for now and wait for a more definitive answer before closing
this ticket.

Thanks!

[1] http://irclog.perlgeek.de/perl6/2008-06-24#i_364168

From @TimToady

On Mon, Jul 07, 2008 at 07​:16​:05PM -0700, Patrick R. Michaud via RT wrote​:
: Note however that TimToady on #perl6 speculated [1] that an empty return
: should return the Object prototype. I'm not sure the answer was
: resolved completely in that thread, however, so we'll go with returning
: 'undef' for now and wait for a more definitive answer before closing
: this ticket.

A C<return> just returns its argument list as a Capture, so it's
probably just returning the "null" Capture, presumably with Object
(undef) in the scalar slot and an empty list in the list slot and
an empty hash in the hash slot. I don't think it's returning
the Capture protoobject itself though, because I think bare return is
returning something defined, so it's just an empty container, much
like the difference between Array and [].

Larry

From @pmichaud

On Mon, Jul 07, 2008 at 08​:18​:36PM -0700, Larry Wall wrote​:

On Mon, Jul 07, 2008 at 07​:16​:05PM -0700, Patrick R. Michaud via RT wrote​:
: Note however that TimToady on #perl6 speculated [1] that an empty return
: should return the Object prototype. I'm not sure the answer was
: resolved completely in that thread, however, so we'll go with returning
: 'undef' for now and wait for a more definitive answer before closing
: this ticket.

A C<return> just returns its argument list as a Capture, so it's
probably just returning the "null" Capture, presumably with Object
(undef) in the scalar slot and an empty list in the list slot and
an empty hash in the hash slot.

By "(undef)" here did you mean simply that it's an undefined
Object (prototype), or that the C<undef> function now returns
Object instead of Failure (S02​:872)?

Thanks,

Pm

From @TimToady

On Mon, Jul 07, 2008 at 11​:43​:56PM -0500, Patrick R. Michaud wrote​:
: On Mon, Jul 07, 2008 at 08​:18​:36PM -0700, Larry Wall wrote​:
: > On Mon, Jul 07, 2008 at 07​:16​:05PM -0700, Patrick R. Michaud via RT wrote​:
: > : Note however that TimToady on #perl6 speculated [1] that an empty return
: > : should return the Object prototype. I'm not sure the answer was
: > : resolved completely in that thread, however, so we'll go with returning
: > : 'undef' for now and wait for a more definitive answer before closing
: > : this ticket.
: >
: > A C<return> just returns its argument list as a Capture, so it's
: > probably just returning the "null" Capture, presumably with Object
: > (undef) in the scalar slot and an empty list in the list slot and
: > an empty hash in the hash slot.
:
: By "(undef)" here did you mean simply that it's an undefined
: Object (prototype), or that the C<undef> function now returns
: Object instead of Failure (S02​:872)?

Hmm, well, on further reflections, it should in all likelihood be
neither of those. I can argue that it should be a special failure type
to try to get an item from a null list​: "Attempt to pull a rabbit out
of an empty hat" or some such. :)

I can also argue that a null list in item context simply promotes to
an empty array just as a list of 2 or more elements does. Given that
"return;" is not the same as "fail;", that's probably the right thing to
do.

It does treat single-item lists as special, but note that .[0] is
a no-op on a non-array. And it's really the .[0] operation that
translates to the pull-a-rabbit-out-of-an-empty-hat failure if you
hand it a null list. Well, it's probably really "Subscript out of
range"...

Which argues that bare return should be treated as returning a list
that happens to be empty (via the Capture that return always returns
to keep the return value contextually lazy).

So

  $x = nothing(); # = []
  @​x = nothing(); # = ()
  %x = nothing(); # = ()
  if nothing() {...} # always false
  $x = nothing()[0] # "Subscript out of range" failure

In other words, I think I'd like to not encourage people to use bare
return to indicate failure.

Larry

From @TimToady

On Mon, Jul 07, 2008 at 10​:09​:37PM -0700, Larry Wall wrote​:
: So
:
: $x = nothing(); # = []
: @​x = nothing(); # = ()
: %x = nothing(); # = ()
: if nothing() {...} # always false
: $x = nothing()[0] # "Subscript out of range" failure

But note that binding the returned capture also fails​:

  $x := nothing();

just as calling foo() without args can't bind to a sig with required
params​:

  sub foo ($x) {...}

Binding to $x is more like looking for .[0], in other words.

Note however that when you call

  foo(nothing())

you're really passing a Capture with 1 positional element, which
contains the empty Capture returned by nothing(), so $x is presumably
bound to [] by promotion of the empty list, just as

  foo(threeitems())

would bind an array of 3 elements.

Larry

From @moritz

A recent change in the test suite made three rakudo tests fail (although
I don't know why, they should be correcter in the new form). All of them
look roughly like this​:

sub bar { return }
ok(!defined(bar()), '... bare return statement returned undef');

(from t/spec/S04-statements/return.t)

I TODO'ed them now (to make spectest_regression clean again), and opened
this ticket instead.

Moritz

--
Moritz Lenz
http://moritz.faui2k3.org/ | http://perl-6.de/

From @moritz

Merging into 58770, because the discussion here is relevant over there
as well, and the original issue (the error message) is resovled.

See also​:
http://www.nntp.perl.org/group/perl.perl6.language/2008/11/msg29884.html
(bare return should now return Nil, which is undef in scalar context).

From @pmichaud

Now fixed in r33898, thanks!

Pm

@pmichaud - Status changed from 'new' to 'resolved'