brice-morin
commented
7 years ago Yes, I think I have experienced the same issues on the same project :-) I guess the default strategies should be to check on save, or something like this.
Well, the checker is deeply inefficient. It did not really show until we start to have proper ThingML model :-) Each rule basically re-browse the whole configuration and looks for stuff it is interested in. I think a better approach would be:
- only browse the configuration (and everything it contains or refers to) once
- each rule subscribes to model elements of interests (e.g. functions)
- the main "browser" notifies the rules with only the elements they are interested in
- each rule can just check what it is supposed to check, without the need for it to first find the elements of interests.
The checker seem to trigger way too often (or take too much time). Working on this project https://github.com/ffleurey/ThingML-PongTutorial in eclipse is almost impossible because for every time on file is checked, all the thingml files in the workspace are re-parsed and checked. Before the checker was improved to catch more errors, this was not much of a problem. We should:
1/ Make sure that we do not run the checker more than needed (it may be that xtext default behavior is to re-pared and validate too often) and
2/ we should use a profiler on the checker to make sure that there are no big inefficiencies in it.