Eclipse Community Forums: TMF (Xtext) » refining generated model

Help

Home

Home » Modeling » TMF (Xtext) » refining generated model

Show: Today's Messages :: Show Polls :: Message Navigator

refining generated model [message #761095]

Mon, 05 December 2011 22:07

Vlad Dumitrescu

Messages: 431
Registered: July 2009
Location: Gothenburg

Senior Member

Hi,

I think that is easiest to write a grammar that creates a rather generic parse tree and would like to refine the generated model to provide the more specific features. For example,

Module: forms+=Forms*;
Form: Attribute | Function;

where I'd like to have access to the attributes and functions separately. I can't write

Module: (attributes+=Attribute | functions+=Function)*;

because this loses the order and the serialized form puts all attributes or functions at the start.

Can I somehow refine the model directly, or do I have to resort to helper classes? Would it work better if I was to write my code in Xtend?

(This is quite a lot to take in, and then there's the use of Guice too that makes things less familiar, so please excuse me if all these questions are naive. I do try to find the answers myself first, but most examples are for simple languages.)

best regards,
Vlad

Report message to a moderator

Re: refining generated model [message #761157 is a reply to message #761095]

Tue, 06 December 2011 01:52

Henrik Lindberg

Messages: 2509
Registered: July 2009

Senior Member

The approach I took was to use the generated model until I was finished
exploring what my language would be like. I then copied the model to a
separate project and turned off generation in favor of using an external
model. I then refined the model using EMF tools.

After some point in the dev cycle you really do not want changes in the
grammar to have an effect on the model.

Regards
- henrik

On 2011-05-12 23:07, Vlad Dumitrescu wrote:
> Hi,
>
> I think that is easiest to write a grammar that creates a rather generic
> parse tree and would like to refine the generated model to provide the
> more specific features. For example, Module: forms+=Forms*;
> Form: Attribute | Function;
> where I'd like to have access to the attributes and functions
> separately. I can't write
> Module: (attributes+=Attribute | functions+=Function)*;
> because this loses the order and the serialized form puts all attributes
> or functions at the start.
>
> Can I somehow refine the model directly, or do I have to resort to
> helper classes? Would it work better if I was to write my code in Xtend?
>
> (This is quite a lot to take in, and then there's the use of Guice too
> that makes things less familiar, so please excuse me if all these
> questions are naive. I do try to find the answers myself first, but most
> examples are for simple languages.)
>
> best regards,
> Vlad
>
>

Report message to a moderator

Re: refining generated model [message #761259 is a reply to message #761157]

Tue, 06 December 2011 07:48

Vlad Dumitrescu

Messages: 431
Registered: July 2009
Location: Gothenburg

Senior Member

Thank you Henrik!

This is a good idea, definitely to be used when the grammar is stable. However, if I understand correctly, it means that until then I still have to find a way to encapsulate the model refinements (because I don't want to rewrite all the IDE support). I think I might be able to use Xtend and its extension methods.

best regards,
Vlad

Report message to a moderator

Re: refining generated model [message #761262 is a reply to message #761157]

Tue, 06 December 2011 07:50

Meinte Boersma

Messages: 434
Registered: July 2009
Location: Leiden, Netherlands

Senior Member

I tend to create model helper classes/extensions which provide "derived model properties" quite early on in development. Thanks to Xtend2's extension mechanism these tie in beautifully to any Xtend2 code (e.g., templates) that I'm using. You can have Guice inject those helper classes by just annotating the field of the client with '@Inject' - no need to define a binding anywhere if the helper class is its own implementation.

Fixing the meta model is useful when the meta model doesn't change anymore (or in a very controlled fashion) and it is part of the public API. In the beginning of dev, it's typically only a hindrance because you want to be able to move quickly in exploring what your language should look like.

Xtext blogs: executable models...again? | workshop material | custom scoping with Xtend

Report message to a moderator

Re: refining generated model [message #761266 is a reply to message #761262]

Tue, 06 December 2011 07:55

Vlad Dumitrescu

Messages: 431
Registered: July 2009
Location: Gothenburg

Senior Member

Thanks Meinte!

Yes, I see. Using Guice is still something new for me and it solves a lot of issues that otherwise require quite a lot of plumbing.

My language is actually stable, the problems that I have are related to trying to get as close as possible to an existing model and also to handling a few constructs that seem difficult to parse with a LL grammar (the original parser is a recursive-descent one).

best regards,
Vlad

Report message to a moderator

Re: refining generated model [message #761300 is a reply to message #761266]

Tue, 06 December 2011 08:58

Meinte Boersma

Messages: 434
Registered: July 2009
Location: Leiden, Netherlands

Senior Member

Xtext (ANTLR) actually generates a recursive-descent parser, so in theory those constructs shouldn't be harder to parse with an LL(*)-grammar. Mapping a recursive-descent parser implementation to an LL(*)-grammar isn't wholly trivial, though Wink

Xtext blogs: executable models...again? | workshop material | custom scoping with Xtend

Report message to a moderator

Re: refining generated model [message #761879 is a reply to message #761300]

Wed, 07 December 2011 08:15

Vlad Dumitrescu

Messages: 431
Registered: July 2009
Location: Gothenburg

Senior Member

Yes, I noticed that! I got it almost working, but there are a couple of constructs that are ambiguous and I'm not sure what to do about them. I can tweak the grammar, but then the parse tree will be very weird... Well, if it was easy, everybody could do it!

regards,
Vlad

Report message to a moderator

Re: refining generated model [message #762125 is a reply to message #761879]

Wed, 07 December 2011 15:32

Meinte Boersma

Messages: 434
Registered: July 2009
Location: Leiden, Netherlands

Senior Member

In case you haven't done already, you should dive into the subject "(un)assigned actions" and "syntactic predicates" which might very well help you with re-writing the AST to something useful and getting rid of ambiguities, resp.

Xtext blogs: executable models...again? | workshop material | custom scoping with Xtend

Report message to a moderator