Active Model - work in progress, questions and thoughts

rubyonrails-core
#1

Hello, World!

I've been doing some hacking on ActiveModel, and I'm pretty satisfied
with the results. Mmost of the validations stuff is finished.
Callbacks and observing is yet to come (though Rick Olson has done
some work on the observers).

Here's one of the tests, which mocks an active record implementation
of AtiveModel: http://pastie.org/185638

It lets you overwrite default validation settings. It detects if the
implementation defines new_record?, if it does, ActiveModel activates
new/existing record differences. If it doesn't, stuff like :on
=> :update is ignored.

The API is pretty clean, and yes, the tests pass ; ) I'm poking around
with error messages now, trying out various approaches. I think the
basic idea of Errors < Hash is DHH's, and I really based my stuff on
his approach, just tweaking it a bit. For instance, instead of
user.errors[:base] << "foo", I've done user.errors << "Foo" (which was
to be honest Pratik's idea). Also, instead of user.errors[:base] for
yielding them, I'm doing user.errors.on_base. It's a very different
kind of error than the attribute-based errors you add with
foo.erros[:bar], so I think it makes sense to have the api different
for those as well.

Certain things makes backwards compability hard, though. I've changed
it so that adding errors on base causes the record to be invalid,
which as far as I know breaks backwards compability (this is the first
time I hack around whith in-deep rails stuff like this, so do arrest
me on this if I'm wrong). Therefore, aiming for ActiveModel being a
none backwards compatible 3.0 thing would make sense. As mentioned, it
would really easen the process of coding it, and makes it less hacky.
And it really affects active record as well, involving a lot of re-
writing, so the 3.0 badge would indeed be deserved for a release like
this.

So, that's it. My work so far is available on github,
http://github.com/leethal/rails/tree/activemodel_cleanup/activemodel.
I hope you like what I've done so far, I've really enjoyed hacking on
this. Comments appreciated =)

#2

I think you're getting a little ahead of yourself. We still haven't
made any real decisions about the direction of ActiveModel. Its pretty
experimental at this point.

Please feel free to continue to play around but don't assume we are
going be taking your direction.

Also, I think we are going go w/ rspec too.

#3

I didn't really see where he was getting ahead of himself. He's done
a lot of hacking on this and he was saying that if you take his
patches/changes then backwards compatibility could be hard. If you
don't, well, who cares? That's the beauty of Git: he can maintain his
own repos of the code, you can maintain yours.

Also, I think perhaps you're getting ahead of yourself, since I don't
see your name on http://rubyonrails.org/core (at least not yet :)).
I'd like to get Rick Olson's take on it (and I think Rick Bradley has
been involved), since I know he has taken a look at this stuff
previous to this ML message.

As for myself, I think this stuff is really cool. I had someone ask
me about something like this (or, well, a little more than
this...closer to DataMapper I guess) at MySQLConf, so it's cool to see
something materializing. I'm going to start toying around with this
and ActiveRelation once it's substantial enough to tinker with. I'm
really excited to see where the model layer in Rails is going...

--Jeremy

#4

I didn't really see where he was getting ahead of himself. He's done
a lot of hacking on this and he was saying that if you take his
patches/changes then backwards compatibility could be hard. If you
don't, well, who cares? That's the beauty of Git: he can maintain his
own repos of the code, you can maintain yours.

This is, indeed, the beauty of distributed source control, and
experimentation with ActiveModel should be encouraged. However it's
missed the 2.1 boat, so lets not get too worried about what is and
isn't yet fit for the core of the framework. We have until (2.1).next
to worry about such things.

Also, I think perhaps you're getting ahead of yourself, since I don't
see your name on http://rubyonrails.org/core (at least not yet :)).
I'd like to get Rick Olson's take on it (and I think Rick Bradley has
been involved), since I know he has taken a look at this stuff
previous to this ML message.

I know rick's been tinkering with ActiveModel in the past, but I'm not
sure how 'in anger' the use has been. Though you're probably in a
better position to know whether or not the entp projects are doing
anything like this.

As for myself, I think this stuff is really cool. I had someone ask
me about something like this (or, well, a little more than
this...closer to DataMapper I guess) at MySQLConf, so it's cool to see
something materializing. I'm going to start toying around with this
and ActiveRelation once it's substantial enough to tinker with. I'm
really excited to see where the model layer in Rails is going...

I'm also looking forward to giving active record a bit of a kicking,
bi-directional associations and an identity map for starters, but I
see ActiveModel as something else entirely. I'm having trouble seeing
it materialise into something useful until we have an alternative
method of persistence to target. ActiveResource got by fine without
it, perhaps we need active document, active flat-file, active
in-memory-thing or something similar to get a feel for what's needed.

#5

Hey peeps,

Glad you find my stuff at least semi-sensible =)

The reason I'm hacking on it, is beacuse after asking about it's
status in #rails-contrib Rick Olson suggested that I'd start hacking
away on it. It was he hat suggested posting this mailing list as
well.

The way I see it, ActiveModel is about tearing apart various libs
inside ActiveRecord and making them stand-alone, so that 1. the active
record code itself gets a lot cleaner, and that 2. people can include
ActiveModel libs in their own code, without having to use active
record. E.g. databaseless models in a rails app that has validations
and callbacks in them.

Rick Olson has been saying all along that this wouldn't get into 2.1,
so that has never been my goal anyway. DHH also said he was okay with
making Active Model a 3.0 thing (in #rails-contrib), so I guess that
makes sense then.

But really, I've just been playing with this, and I'm not a super
experienced Ruby coder (which you can tell from the
read_inheritable_attribute stuff in the validation settings part of my
hacks =P), so I'm not having any high hopes of the core team embracing
my ideas and hacks from day one.