Front page | perl.perl5.porters |
Postings from October 2002
Hmm, seems like an overview could be helpful:
We are talking about using references as hash keys, e.g.
$a = [];
$b = {};
$c = \*main;
$d = \"gunk";
$e = sub { 'foo' };
%h = ($a => 1, $b => 2, $c => 3, $d => 4, $e => 5);
$a->[0] = 'foo';
$b->{foo} = 'bar';
for (keys %h) {
print ref($_), "\n";
}
as provided by the Tie::RefHash module.
But the discussion also started to circulate around related topics: how
to compare two arbitrary objects for equalness and how to produce a
hash key for an arbitrary complex object.
The semantics of Tie::RefHash are well-defined, it was proposed to
create an XS version for speed-up and as a prototype of what could be
later included into the core. I'd like to discuss a different way:
You can already use a reference as a hash key, but it will be converted
into its stringized form, so 'keys %h' returns a list of scalars, not a
list of references. How about adding a special case to hash key
handling: test if the key is a ref and use the stringized form to find
a bucket entry (should already be in place), but store the ref there
(this should be new), not the stringized scalar. As the stringized
form can stay attached to the reference as an alternate form, this
would mean minimal overhead, probably only the look-up inside the
bucket has to be adapted to also include testing for a ref.
I have to admit I haven't taken a look at the source code and am
speculating from what I've read about internal data structures in the
perldocs, so this is just guessing, please forgive me if I am way off.
Regarding the second topic: how to compare complex objects for
equalness should be further discussed, but the proposed way of creating
a serialized version of the object via Storable (maybe with MD5 or
compression) seems OK to me. For blessed objects, a 'cmp' method could
be used (if it is there) to determine equalness (which can take care of
object peculiarities).
But this definitely doesn't have to be included into the core at this
time as it is quite slow, so an external module (Tie::ObjHash?) should
be written first as a proof of concept.
Just my EUR 0.02.
Roland
--
RGiersig@cpan.org
Thread Previous
|
Thread Next