On 2013-23-04 1:01, Chris Ainsley wrote:
> I'm making good progress with XText having resolved my lexing issues and
> I've managed to get the content assist working as I expect.
>
> Now I'm trying to tacking the issue of linking and scoping. This thread
> is with regard to linking specifically.
>
> I have a language which doesn't have any traditional cross references.
>
> There are however, many validation rules. Some of these validation rules
> involve loading the contents of DSL files (via eResource) in subfolders,
> recursively, and then applying a validation.
>
You are probably better of letting builders deal with all the content
creating EObjectDescriptions that gets indexed by Xtext. You then search
this index in your linking instead of rescanning the content over and
over. (More below).

Anything that is already an Xtext DSL is automatically added to the
index (and it is just a matter of visibility), and anything that is EMF
is easily also added to the index. If you have some other type of
resources, create a model for it and write your own data to EMF to Xtext
indexing (I do this for Ruby logic parsed a special way).

> Now, as far as I can understand, the validation framework calls a
> validate on each model item in a resource that has changed, and triggers
> a recalculation on items linked to the model items that have changed.
>
> So, my question is, how and where do I explicitly create links between
> model elements in different files without using the cross reference
> grammar in my .xtext file?
>
Take a look at cloudsmith / geppetto @ github. I do not use the regular
linking support provided by EMF Xtext and instead use a different
mechanism (I use an adapter at the referencing end instead of an
EReference). This required quite a lot of work throughout the system,
you can probably borrow quite a lot from my implementation.

Look at the PPLinker and PPResourceLinker classes they adapt my special
linker to the normal LazyLinker. It then does the special linking in a
callback; issues linkage errors and computes various things later needed
in quick fixes (to avoid yet another attempt at resolving a reference at
a later point) etc.

The implementation makes heavy use of the index and does several
optimizations in searching for things.

It is important to record all names that are used, as well as names of
things not found, and names investigated (that if they appear later will
change the resolution). This data is recorded at the resource level and
is used by the Xtext builders to figure out the transitive hull of build
changes / the build order to use.

> A secondary question, is it possible to manually trigger a validation on
> the root model element of a given list of eResources programmatically?
> Is this a good idea?

That can be done. That is what the builder is doing. If this is a good
idea or not depends on why and where you want to do this It is quite
different when you are editing things (you want validation and linking
to take place on the content that is in dirty state in the editors to
give users immediate feedback.

Hope this helps
Regards
- henrik