Perl code would treat the two the as the same, but user-written XS code makes assumptions.

When I read this, I felt an urge to find Perl code that behaves differently for these arrays :)

Turns out that what this API function is doing would require explicit undefs even if the function was written in Perl. This is a subroutine that modifies its input parameters. In Perl you might write it like this:

use Data::Dumper;

my @a;
@a = (undef) x 5;
my @b;
$#b = 4;

sub modify {
    $_ = 5 for @_;
}

modify(@a);
print Dumper \@a;
modify(@b);
print Dumper \@b;
[download]

Here the subroutine "modify" behaves similar to GetVolumeInformation as it modifies the scalars in the array passed to it.

So the arrays @a and @b should be identical. However, modify() will replace @a with five times 5, but it cannot write to @b:

$VAR1 = [
          5,
          5,
          5,
          5,
          5
        ];
Modification of a read-only value attempted at script.pl line 12.
[download]

Update: Since this seems to depend on the version: I am using Perl 5.20.2 on Debian.

I don't see your pure-Perl "Modification of a read-only value..." result on ActiveState 5.8.9 and Strawberry 5.14.4.1 (Update: likewise Strawberries 5.10.1.5 and 5.12.3.0), but there's something else I don't understand:

c:\@Work\Perl\monks>perl -wMstrict -MData::Dumper -le
"sub modify { $_ = 5 for @_; }
 ;;
 my @a;
 @a = (undef) x 5;
 ;;
 modify(@a);
 print Dumper \@a;
 ;;
 my @b;
 $#b = 4;
 ;;
 modify(@b);
 print Dumper \@b;
 ;;
 my @c;
 $#c = 4;
 $c[ $#c ] = undef;
 ;;
 modify(@c);
 print Dumper \@c;
"
$VAR1 = [
          5,
          5,
          5,
          5,
          5
        ];

$VAR1 = [
          undef,
          undef,
          undef,
          undef,
          undef
        ];

$VAR1 = [
          undef,
          undef,
          undef,
          undef,
          5
        ];
[download]

Arrays @b and @c are still unwriteable (or at least unwritten, at least in part), but in a different way.   ?!?

Update: Taking the  @_ argument list out of the equation, I get expected results (thank Larry!) for the problematic cases above (for all Perl versions listed in the first paragraph):

c:\@Work\Perl\monks>perl -wMstrict -MData::Dumper -le
"my @b;
 $#b = 4;
 ;;
 $_ = 5 for @b;
 print Dumper \@b;
 ;;
 my @c;
 $#c = 4;
 $c[ $#c ] = undef;
 ;;
 $_ = 6 for @c;
 print Dumper \@c;
"
$VAR1 = [
          5,
          5,
          5,
          5,
          5
        ];

$VAR1 = [
          6,
          6,
          6,
          6,
          6
        ];
[download]

---

Give a man a fish:  <%-{-{-{-<

You do realise that you aren't passing the array into modify()? You're passing a list. And that list is being aliased.

And prior to the mostly pointless, blanket constanisation of the Perl sources, the (incorrect) attribution of a null C pointer to a scalar would not have been spuriously attributed to a "read-only value".

Two mistakes collide here:

1. The lazy definition of an api that takes a list of pre-existing and defined variables as input in order to provide output.
2. The retroactive application of const'ing at the C-level with callous disregard to the affects at the Perl level.

Upshot: p5p have buggered Perl5 in their attempts to 'be correct'.

---

With the rise and rise of 'Social' network sites: 'Computers are making people easier to use everyday'

Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.

"Science is about questioning the status quo. Questioning authority". I knew I was on the right track :)

In the absence of evidence, opinion is indistinguishable from prejudice.

You do realise that you aren't passing the array into modify()? You're passing a list. And that list is being aliased.

Yes I do :) It's a list of aliases, just like in the subroutine GetVolumeInformation from the original post.

p5p have buggered Perl5 in their attempts to 'be correct'.

I think this is related to some optimization. Some people like to use $#array = ... because it is faster than explicitly filling the array. However this optimization comes with a cost.

Assigning to the length does not put any SVs into the new elements, and this is why you cannot create references or aliases to them. I thought there was something about this in the documentation, but I cannot find it any more (only a related section in perlguts)

I would be very interested in any comments you might have on the, to me, very puzzling behavior seen when iterating over the aliased  @_ argument list with arrays  @b and  @c in the  modify() function in the reply above. I have only been able to confirm this behavior in the older Perl versions noted therein.

---

Give a man a fish:  <%-{-{-{-<