Implement a method cache for faster dispatch and allowing super calls

[byroot]

The code is heavily borrowed from ActiveRecord.

It should make the attribute methods faster even though I didn't write any benchmark to back that claim.

The real reason I implemented this, is to be able to use super in attribute methods:

class FooSerializer < ActiveModel::Serializer
  attributes :name

  def name
    super.titleize
  end
end

Regards.

[kurko]

This is interesting. @guilleiguaran @joaomdmoura can you chime in?

[byroot]

Oh I forgot to mention. This technique only works for ruby 2.0+. ActiveRecord up to 4.2 had a fallback. I don't know if it's worth implementing it since 1.9.3 is EOL. But I can do it if it's important.

[bf4]

@byroot We've dropping support for Ruby < 2.0 in master, so this is again viable. Would you be interested in rebasing this off of current master? Let me know if I can help.

[byroot]

I'll try to find some time.

[bf4]

Awesome!

B mobile phone

On Dec 23, 2015, at 3:57 AM, Jean Boussier [email protected] wrote:

I'll try to find some time.

—
Reply to this email directly or view it on GitHub.

[byroot]

So I looked into it, and given the way it's implemented now with AttributeReader etc, this optimization wouldn't help as much. I actually think it would be worse.

The idea was to generate attributes accessor dynamically via class_eval, but to share cache them in a module.

If we were doing that on AttributeReader, it would generate a singleton_class for every attributes, which is quite slow IIRC.

[bf4]

@byroot Oh, but this PR is great! the attributes internal api is still in flux. See e.g. #1370 I hope there's still something we get improve (would need a benchmark, of course, like #233