alternate method dispatch strategies inch closer
It’s been nearly two months since my last update on writing new method resolvers in Perl, and not much has happened. I was very disheartened to learn that while it was easy (for some value of “easy”) to make this work:
This was not as simple:
my $method = 'magic_method'; $object->$method(...);
It’s not all that exciting to explain, so I’ll be brief. In Perl, you can attach “magic” to variables that makes them behave in new ways. This is how the tie built-in works, and it’s very, very powerful and useful. (It’s also very dangerous, if you use it without thinking long and hard, first, so don’t just go crazy with variable magic!)
Method dispatcher replacement worked by attaching magic to the stash, where a class’s symbols are stored. Unfortunately, not all the things that accessed the stash respected magic. Respecting magic can be slower than ignoring magic, so there’s constantly a tension between performance and enchantability in perl5 development.
Florian Ragwitz, who discovered this strategy for method dispatch to begin with, suggested that I stop trying to avoid using his excellent MRO::Define and just use it. It means we’ll need 5.10.1 instead of 5.8.z, but it also means that things work.
Last night I finally got back to my code and pulled MRO::Define back in. All tests pass… except for one. It’s a really cool one, too, demonstrating one of my biggest goals: anonymous, almost-first-class classes and instances.
As you can imagine, there’s a fair bit of magic (or voodoo, in some cases) swirling around this code, and debugging this problem is pretty taxing. Want to give it a go? My code is, for now, in the rjbs/metamethod GitHub repo. It will probably be renamed in the near-ish future.
I look forward for using these new powers for good (and maybe some evil).