Xtext and EMF modeling - Opposite relations parsing - eclipse

Question

I'm currently playing around with XText and EMF and I've come to a dead end.
I have an ecore model that I created using the diagram editor. I don't provide the XML representation; it should be clear from the example. Some of the relations are opposite to each other (like the parent-children relation).
This binding works perfectly fine when I create the instances programatically. Below, I show a test case that is successfully passed.
However, when I parse the model using XText, these opposite relations are not set. I can't find a way to fix this. The relations are strictly one-directional as they appear in the input file. Is there any way to force Xtext to set these? or am I supposed to resolve these manually?
Passing test
WordsFactory factory = WordsFactory.eINSTANCE;
// Prepare a simple dictionary hierarchy
Dictionary d = factory.createDictionary();
Synset s = factory.createAdjectiveSynset();
s.setDescription("A brief statement");
s.setExample("He didn't say a word.");
WordSense ws = factory.createAdjectiveWordSense();
Word w = factory.createWord();
w.setName("word");
ws.setWord(w);
s.getWordSenses().add(ws);
d.getWords().add(w);
d.getSynsets().add(s);
// Now check the bidirectional links
Assert.assertTrue(ws.getSynset() == s);
Assert.assertTrue(w.getSenses().get(0) == ws);
XMI representation of this example
<?xml version="1.0" encoding="ASCII"?>
<words:Dictionary xmi:version="2.0" xmlns:xmi="http://www.omg.org/XMI" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xmlns:words="http://www.example.com/dstahr">
<words senses="//#synsets.0/#wordSenses.0" name="word"/>
<synsets xsi:type="words:AdjectiveSynset" description="A brief statement" example="He didn't say a word.">
<wordSenses xsi:type="words:AdjectiveWordSense" word="//#words.0"/>
</synsets>
</words:Dictionary>
Grammar definition (some unimportant rules removed)
grammar ocs_assignment.dsl.DSL with org.eclipse.xtext.common.Terminals
import "platform:/resource/ocs_assignment.model/model/words.ecore"
import "http://www.eclipse.org/emf/2002/Ecore" as ecore
Dictionary returns Dictionary:
{Dictionary}
'dict' name=EString
('add words' '[' words+=Word* ( "," words+=Word)* ']')?
synsets+=Synset*
;
Synset returns Synset:
AdjectiveSynset | NounSynset | VerbSynset;
WordSense returns WordSense:
AdjectiveWordSense | NounWordSense | VerbWordSense;
Word returns Word:
name=EString;
EString returns ecore::EString:
STRING | ID;
NounWordSense returns NounWordSense:
word=[Word|EString];
NounSynset returns NounSynset:
{NounSynset}
'(N)' name=EString
'{'
'content' '[' (wordSenses+=NounWordSense ( "," wordSenses+=NounWordSense)*)? ']'
'description' description=EString
'example' example=EString
('hyponym' hyponym=[Synset|EString])?
('hypernym' hypernym=[Synset|EString])?
('similarTo' '(' similarTo+=[Synset|EString] ( "," similarTo+=[Synset|EString])* ')' )?
'}';
Parsed file
dict dict
add words test1 test2 test3
(N) test1
{
content [ test1 test2 ]
description "test1"
example "test1"
}
(N) test2
{
content [ test3 ]
description "test2"
example "test2"
hypernym test1
}
XMI representation of the parsed file (missing references for Words)
<?xml version="1.0" encoding="ASCII"?>
<words:Dictionary xmi:version="2.0" xmlns:xmi="http://www.omg.org/XMI" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xmlns:words="http://www.example.com/dstahr">
<words name="test1"/>
<words name="test2"/>
<words name="test3"/>
<synsets xsi:type="words:NounSynset" description="test1" example="test1" name="test1">
<wordSenses xsi:type="words:NounWordSense">
<word href="importedFile1.wdsl#xtextLink_::0.1.0.2.0::1::/0"/>
</wordSenses>
</synsets>
<synsets xsi:type="words:NounSynset" description="test2" example="test2" name="test2">
<hypernym xsi:type="words:AdjectiveSynset" href="importedFile1.wdsl#xtextLink_::0.1.1::1::/21"/>
<wordSenses xsi:type="words:NounWordSense">
<word href="importedFile1.wdsl#xtextLink_::0.1.1.2.0::1::/0"/>
</wordSenses>
</synsets>
</words:Dictionary>

Answer 1

Xtext does not support inverse relations built-in. If you generate the metamodel from your grammar, you would see that the inverse is not set for the corresponding relations. However, if your Ecore model has the inverse relation property is set, Xtext will maintain that.
There are two ways to use such a metamodel:
Create your own Ecore model and reference it.
You could define an IXtext2EcorePostProcessor instance that can update your generated metamodel before it is serialized. For details, see the corresponding blog post of Christian Dietrich.
The first solution is quite easy, but you would lose the auto-generated EMF models. In the second case, you have to write code that updates the Ecore (meta)model before it is serialized that can be really tricky to achieve. We have done the second approach for EMF-IncQuery (for another kind of customization), as at that point the autogeneration was really important, however the customizer became really hard to understand since.