On 21/09/12 11:07 AM, Martyn Taylor wrote:
On 09/20/2012 08:50 PM, David Lutterkort wrote:
On Thu, 2012-09-20 at 13:20 +0100, Martyn Taylor wrote:
Gents.
I'm having some trouble with getting nested resources working properly
from bespoke XML/JSON in RESTful API. There seems to be some
fundamental problems with the way rails works, which means we have to
write a lot of bespoke code to get it working, possibly even having to
monkey patch Rails itself.
I've written an example rails app that highlights the problem and put a
detailed description in the README you can find it here:
https://github.com/mtaylor/Rails-Nested-Resource-Issues
I've ran into this issue on IME. But it's going to affect us across
all
rails projects with REST APIs, so we could do with discussing this
issue
and coming up with the best possible solution that results in as little
replication as possible.
Please read through the problem and let me know if you have any better
ideas that the solutions I have proposed.
The sample app does a nice job of explaining the problem - my issue with
Rails' REST support though has always been that the XML/JSON
representation is essentially a dump of the DB schema, meaning the XML
changes when the DB schema changes. Not good for a stable API
For Deltacloud (where we use Sinatra, not Rails), we have a similar
problem with the CIMI frontend: the XML/JSON serializations are
determined by the CIMI standard. We've resorted to writing our own
support for serialization/deserialization, complete with a DSL that lets
us describe what they should look like for each class.
For the morbidly curious, the main pieces of code doing this are
schema.rb[1], which is the logic behind parsing/unparsing both JSON and
XML, and base.rb[2], which is the base class for all model classes. And
be warned, there are changes for these classes in the pipeline ...
David
[1]
https://github.com/apache/deltacloud/blob/master/server/lib/cimi/models/schema.rb
[2]
https://github.com/apache/deltacloud/blob/master/server/lib/cimi/models/base.rb
Thanks for this.
Writing our own methods for unmarhalling content was one of the
possible approaches we were looking at. Though it does bypass some of
the nice features of rails such as the abstraction of content-types.
I'm hoping we can catch the problem a bit further down the stack
possibly as Jiri suggested and modify params hash.
CIMI is a good case example of when it's not possible to just accept
rails API preference. I was getting some questions earlier on why
this is necessary.
Martyn
There's also RABL [1] for "controlled serialization". I know Foreman
folks are using it, perhaps Ohad (cc'd) can talk about his team
experiences with it?
-d
[1] https://github.com/nesquena/rabl
