Coding to an interface in RoR?

Hi,

I'm currently evaluating the conversion of a j2ee app to a RoR app.
Without going into much detail, the app lets users compose a meal and
shows
the amounts of nutrients associated to that meal (eg salt,
sodium, ...).

These nutrients are read from a database.
In the original app we use a "service" that abstracts all the database
level code etc.
This service implements an interface that defines the functionality.

That way, we can swap the services when needed, but the interface
stays the same, so no adjustments are needed.
(eg. McDonaldsService instead of 'normal' food, when the app is placed
in a McDonalds to say something)
And that's where my problem lies. How can I achieve the same level of
abstraction in Rails?

In concreto, how can I define my model so that I can change the
original model to another model without changing much?
The only solution I see at the moment is defining a very general
model that fits all the different services,
and put everything in a big table.
However, that way we are losing the power of abstraction we had with
j2ee.

Hi,

In the original app we use a "service" that abstracts all the database
level code etc.

Could you describe this service in more detail? Some of us might
know specific details about the Java stuff you are doing.

This service implements an interface that defines the functionality.

Well Ruby makes the whole java concept of interface disappear.
Look up "duck typing" - if two Ruby classes have identical method
definitions they are interchangeable. If they share subsets of method
definitions, they are interchangeable in contexts where only those
subsets are used.

That way, we can swap the services when needed, but the interface
stays the same, so no adjustments are needed.
(eg. McDonaldsService instead of 'normal' food, when the app is placed
in a McDonalds to say something)
And that's where my problem lies. How can I achieve the same level of
abstraction in Rails?

Well a rails user might consider that an obfuscation, not an abstraction. And
the question of whether that abstraction is desirable comes to mind. It just
strikes me as an unusual way to achieve separation of customer data. You
could just have separate databases for each customer, or maintain a
customer table that has relates to the master food/nutrient lookup table.

In concreto, how can I define my model so that I can change the
original model to another model without changing much?

Are you changing the model or the data? I see nothing in the problem that
would indicate that the food_product, ingredients, nutritional_information
models are essentially different between (e.g.) McDonalds and Ben & Gerrys,
just that each customer would offer a customer specific range of food
products.

The only solution I see at the moment is defining a very general
model that fits all the different services,
and put everything in a big table.

Thats pretty much how a Rails user would do it, but they would define multiple
tables for each model and relate them correctly.

However, that way we are losing the power of abstraction we had with
j2ee.

Do you need that power? Or is that abstraction solving a problem introduced
by a weak previous implementation? Like inadequate distinction between
customer data (objects) and actual model types (classes).

I am sorry if I came across a bit aggressive/blunt but your problem appears
to be the kind of overcomplicated java-centric solution to a simple problem
that I have to deal with far too frequently than I should have to.

It actually sounds like you basically maintain different databases/data models
for each customer with different ORM strategies, and try to treat them as
similar at the 'service' level by wrapping each ORM implementation in a
common interface.

A Rails solution would seek to normalise your DB schema first by doing
appropriate E-R modelling. In the Java/J2ee/Java Web Framework world
it is more common to treat the database as an implementation detail, and
go with code-first approaches (which gets a lot of Java projects into trouble).

regards,
Richard.

Thank you Richard for your long reply!

I've gathered some more information about the problem.

The "service" approach is used to abstract the database totally from
the rest of the application.
The benefit of doing that is that the datamodel can change in a
drastic way, but the rest of the app is unaffected since a service is
used to access the information.

The problem is that in fact the datamodel can change (eg another data
source provider, forced by the government, so no discussion is
possible about the database schema). By changing the datamodel, the
objectmodel however should not change. What does change is the
translation between the data- and objectmodel.

However, in Rails, when changing the datamodel the objectmodel is
changed automatically (which is in most cases a good thing).
But if the model changes (eg other combination of attributes to
achieve the same functionality) the controller will likely have to
change...
So that's what my question is about. Is there any way in Rails to
define an 'interface' to a model, so I can use this interface in the
controllers. When the model changes because the data model changes,
the controllers should not be affected.

Thank you for your time!

Joram

Ruby has this built in. Let's say you have a table to store an item
name (string) and it's price in dollars.

class Item < ActiveRecord::Base
  # columns are name and dollars
end

You originally have the price in dollars, but you change the datamodel
to have the price in pesos:

class Item < ActiveRecord::Base
  # columns are name and pesos
  CONVERSION_VALUE = 10.9378 # 1 Dollar = 10.9378 pesos

  def dollars
    pesos/CONVERSION_VALUE
  end

  def dollars=(value)
    self.pesos = value * CONVERSION_VALUE
  end
end

This gives you the same interface as the model with dollars.

Jeremy

Well since the database is abstracted why do you need active record
models? Why not just use plain ruby objects or the Struct class if
it's just data?

However, if you want some sort of 'interface', I would do as such:

Define a module as your interface, but in this case it's also getting
implementation, not just the 'contract' the interface brings:

Module MyInterface
  def method1
  end

  def method2
  end
end

In your object that requires that interface include the module,

require 'my_interface'

class MyModel
  include MyInterface
end

Thank you Richard for your long reply!

No problem.

The "service" approach is used to abstract the database totally from
the rest of the application.

Essentially below the waterline, everything is custom, you get an
implementation-specific PersistenceService, which returns only
implementation-neutral types (models).

The database ORM is effectively wrapped by whatever you are using.

The benefit of doing that is that the datamodel can change in a
drastic way, but the rest of the app is unaffected since a service is
used to access the information.

Well the implementation of everything below service layer is not
trivial.

The problem is that in fact the datamodel can change (eg another data
source provider, forced by the government, so no discussion is
possible about the database schema). By changing the datamodel, the
objectmodel however should not change. What does change is the
translation between the data- and objectmodel.

I understand. This is a strategy to re-use all code above the service-layer,
and make it pretty much 100% implementation neutral. But this becomes
only implementation neutral from a *type* or *interface* perspective. If the
schema is very different from the original reference (i.e. totally different
table relationships) several problems creep in:
- first you will lose a lot of performance without leaning heavily on Hibernate
style performance features (like in memory object caching).
- you are also probably limiting your ability to perform unusual queries
against the DB, as the implementation of the service layer has to
translate the relationships, which means that query implementations will
be limited
- you are doing a lot of CPU-intensive model translation in the context
of your application

However, in Rails, when changing the datamodel the objectmodel is
changed automatically (which is in most cases a good thing).

Well that would be a *feature* of rails. Though Rails does have features
for working with legacy databases, like defining getters in models and
mapping them to differently named table fields. Though if relationships
change there is not a lot you can do there.

But if the model changes (eg other combination of attributes to
achieve the same functionality) the controller will likely have to
change...
So that's what my question is about. Is there any way in Rails to
define an 'interface' to a model, so I can use this interface in the
controllers. When the model changes because the data model changes,
the controllers should not be affected.

Well you kind of can. This is close to the Java model you describe, but
there is no service, or distinction between service models, and controller
-used models.
1. You can just define arbitrary models with no relationships
2. map the models to the implementation specific table schema, using
a simple vendor specific configuration code and/or file
3. Do Model.find(:foo) at the controller level and tease the relationships
and data conversion together using model helpers (that use vendor
specific libraries or mixins)
4. The models will present a consistent interface to controllers and views
5. It will also perform like an absolute dog, Rails isn't meant for this
kind of in-code model mapping, and you are likely to be pulling back
bigger result sets than necessary

Depending on the size of the source database, you could simply
create a second database (possibly even in memory) and translate
the implementation specific database into the reference database.
Not a perfect solution, keeping the reference database (that the
models look at) up to date is a problem.

There is also the "I hate the people I work with, and in 2 weeks time
I am going travelling the world" approach:
- do implementation specific metaprogramming
- use metaprogramming techniques to dynamically alter your models
& relationships to match the implementation database
- quake in fear at the insanely ambitious use of metaprogramming,
but keep your fingers crossed, as *it just might work*

There is also the other Rails approach:
1 - writing controllers & views is simple, just make a bespoke implementation
for each schema model, and keep the tests intact
2 - refactor controllers and views to use the new models
3 - bill accordingly

regards,
Richard.

Thank you Jeremy, Wes, Apsoto and Richard for your contributions!

You have given me insight into a difficult subject.
I haven't got much to add anymore at the moment, your information has
given me a lot to think about!

Greetings,

Joram