datastructures and out of memory errors

[busunsl]

I was playing with XML::Simple recently, a nice module when you have small XML files. Read here for a review.

One of the options of XML::Simple is to insert an array on each hierarchy level, so it can distinguish between attributes and elements.

My first use was without that extra level, so I could access stuff just by dereferencing hashes:

$template->{level1}->{level2}->{element1} = '1234';

It worked well until I realized, that I needed that extra level. The structure changed from something like:

$template = {
              'level1' => {
                            'level2' => {
                                          'element1' => '1234'
                                        }
                          }
            };

to

$template = {
              'level1' => 
                         [ 
                           {
                            'level2' => 
                                        [
                                          {
                                            'element1' => '1234'
                                          }
                                        ]
                           }
                         ]
            };

But I didn't change my program. :-(

print $template->{level1}->{level2}->{element1};

just printed nothing and

$template->{level1}->{level2}->{element1} = '1234';

gave me that wonderful error message, telling me just nothing about what happend:

Out of memory!

So when I access elements of a datastructure in the wrong way I just get undef, but when I assign to it in the wrong way Perl runs out of memory.

I think Perl should report an error, as it does when you try to use a normal hash reference as an array reference.

Is this behaviour documented somewhere?
Is it a bug?

[strat]

I've had a similar problem, it just was a hash of hashes of arrays (about LDIF-Format) unter Perl5.6 (under Win2k, Linux and Solaris).
I decided to use Perl5.005_03, which gave me a better performance, too

Version:           Runtime  Memory-Usage
5.0005_03 (AS522)  ~20 min  ~480 MB
5.61      (AS631)  ~90 min  ~700 MB

When I've got more time, I'll try to extract a simple script with the essential logic and commit a bug-report... because such a difference (especially in runtime) can't be true...

Best regards,
perl -le "s==*F=e=>y~\*martinF~stronat~=>s~[^\w]~~g=>chop,print"

PS: somebody has submitted a similar bug with 5.6.1 to activestate with hashes or hashes or the like...

PS2: I asked Microsoft some months ago when there came an Out-Of-Memory-Error while using about 1.2 GB RAM under Win2k, and they told me something about Memory-Fragmentation, and you could prevent it, if you preallocate the RAM (in Perl with: $#list = 10000; or the like).

[blakem]

Looks like you might have tripped over pseudohashes. This same "out of memory" issue has been mentioned twice in the past past couple weeks.... Check out Odd: Out of memory! error for a good discussion, as well as Re: XML Parsing Out of Memory error.

-Blake

[busunsl]

Yes, pseudohashes might be an explanation.
I didn't think of that.

I never liked them and that is another argument against that misfeature.

Thanks!