Bugzilla – Full Text Bug Listing
| Summary: | [xtext][builder] AbstractBuilderState does not load the persisted contents in all cases | ||
|---|---|---|---|
| Product: | [Modeling] TMF | Reporter: | Sebastian Zarnekow <sebastian.zarnekow> |
| Component: | Xtext | Assignee: | Project Inbox <tmf.xtext-inbox> |
| Status: | REOPENED --- | QA Contact: | |
| Severity: | major | ||
| Priority: | P3 | CC: | mark.g.j.christiaens |
| Version: | 2.0.0 | ||
| Target Milestone: | --- | ||
| Hardware: | PC | ||
| OS: | Mac OS X - Carbon (unsup.) | ||
| Whiteboard: | |||
|
Description
Sebastian Zarnekow
Pushed to master. Just wanted to report that I applied your fix (injected the patched ClusteringBuilderState) and it does seem to resolve the issue during startup. Apparently though, that's only half of the story. I have pretty much the same issue during cleaning of a project: - I have an editor open that is properly linked - I do a project clean - The editor gets a lot of error markers (I guess that's ok; while doing a clean build I don't expect cross references to be resolved) - Once the clean build is done, the error markers remain and I can't follow the cross references (F3). - When I close and open that editor, things are fine again. - I also have a second editor that's open but not visible (another tab). Once the cleaning is done, that editor seems to be fine. Very strange. I disabled the IResourceScopeCache just to make sure that I wasn't looking at a caching issue but that didn't help. Currently, I kind of suspect that after the clean build, a refresh of the editor's state is omitted but I'm not familiar with that part of the code. I'll get back when I know more. Hi Mark, thanks for the follow up. A build will trigger a IResourceDescription.Event that contains a list of deltas. This event finally ends up in DirtyStateEditorSupport#isReparseRequired(XtextResource, Event) where IResourceDescription.Manager#isAffected(Delta, IResourceDescription) is used to decide whether the editor has to be re-validated or not. Could you please verify that these methods are actually called in your case? Did you do any customizing in the implementation of the IResourceDescription.Manager? I'm using the DefaultResourceDescriptionManager. When I request a clean build, the isReparseRequired method is invoked twice for the resource corresponding to the open editor. In both cases, it returns false. Note BTW that the editor window itself already has lost its semantic highlighting and already contains annotations of broken cross reference when entering the isReparseRequired method for the first time. So there must be another trigger to refreshing the editor. I suspect that this is the problem: When I request a clean build, a DirtyStateEditorStateJob is created and runs. That job will eventually reparse the content of my editor. While that job is running, the builder starts and iterates over all resources to perform the clean build. It too will reparse the content of my editor (well, I guess the content of the underlying file of the editor). Unless, those two happen in complete isolation and the update to the ResourceSet occurs atomically, I think this is a race. To confirm that this is happening I have the following to stack traces that were simultaneously visible in my debugger: Thread [Worker-39] (Suspended (breakpoint at line 27 in AbstractParser)) VhdlParser(AbstractParser).parse(Reader) line: 27 LazyLinkingResource(XtextResource).doLoad(InputStream, Map<?,?>) line: 151 LazyLinkingResource.doLoad(InputStream, Map<?,?>) line: 69 LazyLinkingResource(XtextResource).reparse(String) line: 173 DirtyStateEditorSupport$UpdateEditorStateJob$2.process(XtextResource) line: 143 DirtyStateEditorSupport$UpdateEditorStateJob$2.process(Object) line: 1 DirtyStateEditorSupport$UpdateEditorStateJob$2(IUnitOfWork$Void<T>).exec(T) line: 36 XtextDocument$XtextDocumentLocker(AbstractReadWriteAcces<P>).modify(IUnitOfWork<T,P>) line: 49 XtextDocument$XtextDocumentLocker.modify(IUnitOfWork<T,XtextResource>) line: 189 XtextDocument.internalModify(IUnitOfWork<T,XtextResource>) line: 98 DirtyStateEditorSupport$UpdateEditorStateJob.run(IProgressMonitor) line: 131 Worker.run() line: 54 The LazyLinkingResource is org.eclipse.xtext.linking.lazy.LazyLinkingResource@d21532ea uri='platform:/resource/demo/top2.vhd' Thread [Worker-46] (Suspended (breakpoint at line 27 in AbstractParser)) owns: VhdlClusteringBuilderState (id=1870) VhdlParser(AbstractParser).parse(Reader) line: 27 LazyLinkingResource(XtextResource).doLoad(InputStream, Map<?,?>) line: 151 LazyLinkingResource.doLoad(InputStream, Map<?,?>) line: 69 LazyLinkingResource(ResourceImpl).load(InputStream, Map<?,?>) line: 1494 LazyLinkingResource(ResourceImpl).load(Map<?,?>) line: 1282 XtextResourceSet(ResourceSetImpl).demandLoad(Resource) line: 255 XtextResourceSet(ResourceSetImpl).demandLoadHelper(Resource) line: 270 XtextResourceSet(ResourceSetImpl).getResource(URI, boolean) line: 397 VhdlClusteringBuilderState(ClusteringBuilderState).writeNewResourceDescriptions(BuildData, IResourceDescriptions, CurrentDescriptions, IProgressMonitor) line: 238 VhdlClusteringBuilderState(ClusteringBuilderState).doUpdate(BuildData, ResourceDescriptionsData, IProgressMonitor) line: 93 VhdlClusteringBuilderState(AbstractBuilderState).update(BuildData, IProgressMonitor) line: 107 XtextBuilder.doBuild(ToBeBuilt, IProgressMonitor, IXtextBuilderParticipant$BuildType) line: 158 XtextBuilder.fullBuild(IProgressMonitor, boolean) line: 182 XtextBuilder.build(int, Map, IProgressMonitor) line: 85 BuildManager$2.run() line: 629 SafeRunner.run(ISafeRunnable) line: 42 BuildManager.basicBuild(int, IncrementalProjectBuilder, Map, MultiStatus, IProgressMonitor) line: 172 BuildManager.basicBuild(IProject, int, ICommand[], MultiStatus, IProgressMonitor) line: 203 BuildManager$1.run() line: 255 SafeRunner.run(ISafeRunnable) line: 42 BuildManager.basicBuild(IProject, int, MultiStatus, IProgressMonitor) line: 258 BuildManager.basicBuildLoop(IProject[], IProject[], int, MultiStatus, IProgressMonitor) line: 311 BuildManager.build(int, IProgressMonitor) line: 343 AutoBuildJob.doBuild(IProgressMonitor) line: 144 AutoBuildJob.run(IProgressMonitor) line: 242 Worker.run() line: 54 The LazyLinkingResource is org.eclipse.xtext.linking.lazy.LazyLinkingResource@dc84320f uri='platform:/resource/demo/top2.vhd' I must say that it's unclear to me why the UpdateEditorStateJob is actually created. It's triggered because org.eclipse.xtext.ui.editor.DirtyStateEditorSupport.UpdateEditorStateJob.run(IProgressMonitor).affectedResources is not empty. Don't know why because I'm working with a completely clean project. Whatever the reason for this, the UpdateEditorStateJob and the Builder probably need more synchronization. Hi Mark, I guess you are right. The scheduling rule that is used for the update editor job may not be sufficient. We'll investigate further for SR1. Regards, Sebastian We've noticed that something similar seems to be happening when a project is first created. Our new project wizard creates a project with a single file that's immediately opened. That file is syntactically colored but not semantically. Since, the project starts building immediately, I suspect that a similar race is present between the builder and updating the editor. I'll dig a bit deeper to see whether I can find a workaround since this is quite a critical problem for us. Hi Mark, to verify your assumption, you could override DirtyStateEditorSupport.createUpdateEditorJob() and the IWorkspace.getRuleFactory().buildRule() as the scheduling rule. (In reply to comment #8) > Hi Mark, > > to verify your assumption, you could override > DirtyStateEditorSupport.createUpdateEditorJob() and the > IWorkspace.getRuleFactory().buildRule() as the scheduling rule. I've added this scheduling constraint. It's _slightly_ better. Previously, when I had an open editor and did a clean build, the editor would lose its semantic highlighting. Now, it does not. What's still broken is when I perform _two_ clean builds back to back while an editor is open. So, I start a clean build and while that is running I start another one (that's of course waiting for the first). Right when the second one starts, the editor again loses its semantic highlighting. I suspect that semantic highlighting sees an empty global scope right before the second clean build but I'm still investigating. I would expect that there is a time-window during which the global index is empty or incomplete and in the process of being refilled. During that period, the highlighter can't follow cross file references. During that window, I see two sensible strategies: - Let the highlighter run but re-trigger it once the global scope is rebuilt. - Block the highlighter from running until the global scope is rebuilt. I think the first strategy is probably what Xtext is trying to do but I'm not sure. If so, there's probably a refresh trigger missing somewhere. That refresh would re-highlight the broken editor. I've been following the execution of the refresh of the editor after the second clean. I see that a UpdateEditorStateJob is (re-)scheduled. It doesn't do anything though. I think the cause is that affectedResources is empty. Digging deeper, the cause seems to be that collectAffectedResources filters out the current resource:
...
List<Resource> resources = resourceSet.getResources();
for(int i = 0; i< resources.size(); i++) {
Resource res = resources.get(i);
if (res != resource && res != null) {
URI uri = res.getURI();
...
The resource of the open editor is present in the event.getDeltas () but this filter blocks it from ending up in the result List.
I don't know why this filter is present. I would be inclined to remove it?
The resource itself should not be contained in the list of affectedResources because these will be unloaded. You should make sure that isReparseRequired returns true, instead. I noticed that removing that filter is not a good idea :)
I now added the following code which triggers a refresh of the editor after the second consecutive clean build. Seems to do the trick. (I also have the code in place that uses the builder's scheduling rule for the editor update).
In DirtyStateEditorSupport:
@Override
protected boolean isReparseRequired(XtextResource resource, Event event) {
return super.isReparseRequired(resource, event) || isCleaned (resource, event);
}
private boolean isCleaned(XtextResource resource, Event event) {
ImmutableList<Delta> deltas = event.getDeltas();
ResourceSet resourceSet = resource.getResourceSet();
URIConverter converter = resourceSet.getURIConverter();
for (Delta delta : deltas) {
if (delta.getOld() == null) {
final URI newUri = delta.getNew().getURI();
final URI normalizeNewUri = converter.normalize(newUri);
if (normalizeNewUri.equals(resource.getURI())) {
return true;
}
}
}
return false;
}
Not 2.1 |