Marshaled data ordering inconsistent

[mla mla]

It seems that when you marshal hashes the binary representations aren't
stable.

This is an issue with PStore because it uses the md5 of the marshaled
data to decide whether it's changed and whether a disk write is needed.

Perl's serialization module, Storable, has a $Storable::canonical option
which, if true, sorts the keys of hashes. Would it be reasonable to add
such an option to Marshal?

I read that Ruby 1.9 will have ordered hashes. So maybe that will make
the issue moot?

Script that illustrates the issue:
http://mla.homeunix.com/tmp/marshal-test.rb

Output that I see:
http://mla.homeunix.com/tmp/marshal-test.txt

ruby 1.8.6 (2007-11-18 patchlevel 5000) [i686-linux]

 

[Robert Klemme]

It seems that when you marshal hashes the binary representations aren't
stable.

This is an issue with PStore because it uses the md5 of the marshaled
data to decide whether it's changed and whether a disk write is needed.

Perl's serialization module, Storable, has a $Storable::canonical option
which, if true, sorts the keys of hashes. Would it be reasonable to add
such an option to Marshal?

Sounds good to me. What happens if keys are not sortable? Fallback
to default, i.e. store unsorted?

I read that Ruby 1.9 will have ordered hashes. So maybe that will make
the issue moot?

Probably not because not all applications need ordered Hashes.
Usually ordering imposes an overhead that you might not want to pay.
So I'd say even in 1.9 this issue is not (fully) gone.

Kind regards

robert