Community
Participate
Working Groups
The Xtext index can impose very heavy costs on Modeling projects; particularly those using UML that provides many version variants of large models. For many projects these costs are completely unnecessary since the Xtext Index is not required. Prior to the index, an Xtext grammar could reference platform:/resource/.... for uninstalled models and there was no need for any further indexing. Now Xtext requires a namespace URI and so needs to provoke this very costly search. Suggest using a .settings file similar to that used by the QVTo project preferences so that projects may declare their own mappings from nsURI to file avoiding the need for global workspace searches for disciplined usage. So: when Xtext/Xtend/... resolves a reference, it first looks for the .settings file If .settings file exists if resolution exists then use it else if resolution exists but is blank/null, start up and use Xtext index else if resolution for "*" exists start up and use Xtext index else generate warning message else start up and use Xtext index This would re-instate the normal Eclipse lazy policy of only incuring heavy costs when needed.
Could you elaborate on the cost that are caused by the indexing? Are you refering to the indexing of *.ecore files? How long does that take for the mentioned packages?
When the Xtext Index 'works' it costs about 10- seconds to index all the UML files on my classpath. Unfortunately the Xtext Index does not work and so sometimes it loops 20 times triggering rebuilds and so a cup of coffee is called for.
Is there an easy way to reproduce the looping?
No. As of M3 some platform build anomalies should go. I was hoping that there would be some .option that could be enabled to identify the triggers for Resource Description Updates so that I could watch those. Running in a nested Eclipse so that I can run an instrumented builder is not much fun when the triggers are difficult to understand and intermittent in behaviour.
(In reply to Ed Willink from comment #2) > When the Xtext Index 'works' it costs about 10- seconds to index all the UML > files on my classpath. Is that *.uml files or the ecore files which define UML?
uml30.ecore is the one that glares at me particularly often from the status message.