|
| Re: Creating centralised datatypes model [message #629855 is a reply to message #629843] |
Wed, 29 September 2010 18:35   |
 Ed Merks
Messages: 33136
Registered: July 2009 |
Senior Member |
|
|
Alan,
Comments below.
Alan wrote:
> Hi all,
>
> I am looking to create a centralised datatypes model that can be
> shared by different models across the toolsuite we are creating. So
> for example we need the simple types like Bool, Int etc
EcorePackage and XMLTypePackage both provides those already.
> but also the more complex types like arrays.
There are many types of arrays. Things like EByteArray are provided by
EcorePackage and XMLTypePackage...
>
> With the simple types its pretty straightforward as they cannot be
> configured and as such can be shared by everyone who needs a bool.
By configured you mean they are immutable...
> The model is saved in its own resource with the Datatypes container
> having datatypes in it (containment= true) and all referencing objects
> from other models having references to these types (containment=false).
So you want to have objects that contain your values?
> So if two models used a bool they'd both point to the same object (via
> a href) in this central resource.
You'd want, for example, a single EObject to represent a wrapper for the
boolean value true?
>
> So far so good. But when it comes to arrays for example we have an
> issue. These are configured with contained types (arrays of bools,
> ints etc) and also with bounds (0..10, 0..100 etc) so in this case
> having them centrally contained is a little less clear to me. I tried
> to have the create methods on the factory take parameters but this
> caused an issues on loading of resources when the create(ECLass)
> method of the factory is used as I cannot see how this can pass the
> parameters I need to the constructor.
Deserialization generally creates empty instances which are then
populated as the information about them is deserialized...
>
> I could have simple types contained centrally and complex types
> contained by the seperate models but I would need two attrs for this
> then surely ? (simpleDatatype, complexDatatype when I can only have
> one datatype) .
I'm having a hard time trying to figure out what's motivating this. I
can't imagine it's a good thing for there to be a central place that
stores all integer values being used, and all floats, and so on. I
imagine it's generally good to store that information about the value at
the point it's used.
>
> Has anyone built such a model and if so got any advice ?
It's a little hard to imagine why you'd want this.
> Also can EDatatype help me in this ?
How come you can't use reuse EMF existing data types and define new ones
for the types not already supported? What's the purpose of this
centralized place to store wrappers for all values?
>
> thanks,
> Alan.
Ed Merks
Professional Support: https://www.macromodeling.com/
|
|
|
|
| Re: Creating centralised datatypes model [message #630053 is a reply to message #629984] |
Thu, 30 September 2010 14:05 |
Ed Merks
Messages: 33136
Registered: July 2009 |
Senior Member |
|
|
Alan,
Comments below.
Alan wrote:
> Hi Ed,
> thanks for the reply. I think I should maybe state the problem from
> scratch rather than attempt to answer the questions individually. I
> started with the idea of a centralised model but let me take a step
> back and define the issues.
>
> Ok so we are developing an editor for a language.
You're looking at using Xtext?
> Variables in this language have a dataype attribute. These can be
> primitive type such as Bool and Int but can also be User defined so
> the user can define their own extension of primitive types. They can
> also define their own Structs which mean they can combine types in
> another type.
Okay, so there's a type system. The Xtext folks are working on
something they call Xbase to help implement a common core for typed
expression languages.
>
> We also have other seperate tools (non language editors) that have
> objects with types. We need to be able to connect objects from the two
> different tools together and whether they can be connected will be
> determined (to start with at least) on their datatypes.
I see.
> So if both are bool then they are ok for connection for example. Some
> complex types such as Structs will be pre-configured and as such can
> also be shared between the tools (consider it a library of types if
> you will)
So some times are "built-in" while others are user defined in the language.
> As all tools (current and future) have types I thought to create a
> central shared type model. As you point out a lot of what I need is
> already available.
> "How come you can't use reuse EMF existing data types and define new
> ones for the types not already supported?"
> I guess the answer to this is that I can see no reason. Let me ask you
> this - I have a Class Variable and it has an attribute datatype. Using
> the re-use approach what type would datatype be ?
Given that types can be defined by the user, it seems best to treat
built-in ones as much like that as possible. Compare it to XML Schema
for example. You can define simple and complex types in schemas, but
there is a set of built-in simple types. Those are defined in a schema
schema XMLSchema.xsd and those are ubiquitously available in all other
schemas. So from the point of view of a model for XML Schema, there is
just a "special schema" and this ends up acting like a central
instance. Ecore's EcorePackage and XMLTypePackage end up performing
that role for Ecore. It sounds like you want that type of concept for
your language.
Have a look at XText and XBase as well; you might find that 90% of what
you're trying to do has been well thought out already.
>
> Hope I was clear enough on what we need. If not please let me know.
>
> thanks,
> Alan
Ed Merks
Professional Support: https://www.macromodeling.com/
|
|
|
Powered by
FUDForum. Page generated in 0.02404 seconds
] 
Search
Help
Register
Login
Home
