On Wed, 2012-09-05 at 15:15 +0200, [email protected] wrote:
> From: Michal Fojtik <[email protected]>
>
> * This patch also moves this to CIMI::Model::Base instead
> of schema to make sure the updated schema is the model
> schema.
>
> Signed-off-by: Michal fojtik <[email protected]>
> diff --git a/server/lib/cimi/models/base.rb b/server/lib/cimi/models/base.rb
> index b375ce0..b627b82 100644
> --- a/server/lib/cimi/models/base.rb
> +++ b/server/lib/cimi/models/base.rb
> @@ -94,6 +94,20 @@ class CIMI::Model::Base
> # attribute, we also define a getter and a setter to access/change the
> # value for that attribute
> class << self
> +
> + def <<(model)
> + clone_base_schema unless base_schema_cloned?
> + member_name = model.name.split("::").last
> + if ::Struct.const_defined?("CIMI_#{member_name}")
> + puts "Removing struct"
> + ::Struct.send(:remove_const, "CIMI_#{member_name}")
> + end
> + member_symbol = member_name.underscore.pluralize.to_sym
> + members = CIMI::Model::Schema::Array.new(member_symbol)
> + members.struct.schema.attributes = model.schema.attributes
> + base_schema.attributes << members
> + end
I have to admit that I don't understand this code at all; how does the
assignment members.struct.schema.attributes = model.schema.attributes
make things safer ? Also, why can't the same be done in the Schema
class, since we only manipulate attributes of the Schema class in the
code above ?
In general, we should be able to get rid of all the *Collection classes
- that was part of the whole collection exercise in CIMI, to make it
possible to use the same code for all collections.
In terms of the schema DSL, I would strive towards something that lets
us describe machines like so:
class CIMI::Model::Machine < CIMI::Model::Base
# Set up plumbing for the machine collection, i.e., code
# to serialize/deserialize a collection of machines
acts_as_root_collection
text :state
text :cpu
...
# Expand to a collection with count, operations, etc.
collection :disks, :class => CIMI::Model::Disk
end
I think we will need to have collections as a subclass of
Schema::Attribute at some point, so that we can control expansion of
collections via $expand during the serialization phase. Basically,
collections have two serializations: the unexpanded one with just a
href, and the expanded one with all the details. Which means that at
some point the to_xml/to_json methods will have to take parameters to
control expansion.
I don't think we should address $expand complications with these
patches, but that's where this code is headed.
David
