A solid, complete implementation of attributes. Luckily, attributes are marked experimental.
Here's the issues I think need to be addressed, in a fuzzily sorted order of importance.
- Namespace scoping:
This is fundamental. No more attributes in UNIVERSAL. It should be as easy to import/export attributes as variables and subroutines. Without it you can't release any general purpose attribute modules on CPAN that other module authors can re-use. They'll only be useful in-house where you can control they won't conflict with each other. Back in 2005 26 of 28 modules (that I found) on CPAN that provided attributes polluted the UNIVERSAL namespace.
- A unified implementation:
The simple FETCH_/MODIFY_*type*_ATTRIBUTES methods implementation is just begging for trouble. It works for systems like Catalyst, but not as a general language feature.
- True attribute arguments:
my $foo : Foo(foo => 'bar'); should work. The parenthesis shouldn't be a single quote yielding a string ("annotation"), but provide arguments to the implementor.
- A working way to store attribute metadata:
After my $foo : Foo;, how do you know that $foo is decorated with Foo?
- Better/working subroutine attributes:
Subroutine attributes should be able to "redefine" the subroutine. (Currently, named subroutines are not fully defined when MODIFY_CODE_ATTRIBUTES is called.) No more broken INIT workarounds. Anonymous subroutine attributes have no chance at all to modify the subroutine (or return another subroutine, wrapping the original). my $foo = sub :Memoize { ... }; cannot be written.
Some two years ago I wrote a note about why I think the current implementation of attributes is broken. I also wrote about a set of modules that would solve some of the issues. While I believe it was a better implementation than the alternatives it didn't overcome the major obstacles as they require changes in the perl source. So I put that work back on the shelf. In real life this is a "redo and do it right" subject.
I still believe there's a lot of potential in attributes, if only implemented properly.
Update: see Namespaces and importing/exporting of attributes for a concrete idea of a possible improvement.
lodin