[#23] Refresh open editors when compile_commands.json changes

[ddscharfe]

Update open editors when compile_commands.json file of the project of the edited file is added/removed/changed.

This makes use of clangd hot-reload, see https://reviews.llvm.org/D92663.
I thought it should be possible to notify clangd via didChangeWatchedFiles, however clangd receives the command but doesn't seem to update the compilation data base.

[ghentschke]

I thought it should be possible to notify clangd via didChangeWatchedFiles, however clangd receives the command but doesn't seem to update the compilation data base

They mentioned in the commit message of llvm/llvm-project@9899319 :

Slight behavior change: we now only look for build/compile_commands.json
rather than trying every CDB strategy under build subdirectories.

I'am not sure, but it seems that the compile_commands.json file has to be in the subdirectory /build relative to the source code file e.g. src/build/compile_commands.json

[ddscharfe]

Slight behavior change: we now only look for build/compile_commands.json
rather than trying every CDB strategy under build subdirectories.

I'am not sure, but it seems that the compile_commands.json file has to be in the subdirectory /build relative to the source code file e.g. src/build/compile_commands.json

I think this comment just means that the auto refresh only checks the compile_commands.json in the build directory, meaning the toplevel directory. AFAIK you can put compile_commands.json files anywhere in the sub hierarchy but they are not covered by this mechanism.

[ghentschke]

I think this comment just means that the auto refresh only checks the compile_commands.json in the build directory, meaning the toplevel directory.
Have you tried this?

[ddscharfe]

I think this comment just means that the auto refresh only checks the compile_commands.json in the build directory, meaning the toplevel directory.
Have you tried this?

I just did. It doesn't seem to make a difference at all if the json file is at the toplevel or under toplevel/build.

Btw I just tested using org.eclipse.lsp4e 0.15.0 instead of org.eclipse.lsp4e 0.16.0 and with the 0.15.0 version it seems to work without the delay of 5 seconds. However with 0.15.0 I get an exception

INFO: Failed to send notification message.
org.eclipse.lsp4j.jsonrpc.JsonRpcException: java.io.IOException: Stream closed
	at org.eclipse.lsp4j.jsonrpc.json.StreamMessageConsumer.consume(StreamMessageConsumer.java:72)
	at org.eclipse.lsp4e.LanguageServerWrapper.lambda$3(LanguageServerWrapper.java:287)
	at org.eclipse.lsp4j.jsonrpc.RemoteEndpoint.notify(RemoteEndpoint.java:126)
	at org.eclipse.lsp4j.jsonrpc.RemoteEndpoint.sendCancelNotification(RemoteEndpoint.java:180)
	at org.eclipse.lsp4j.jsonrpc.RemoteEndpoint$1.cancel(RemoteEndpoint.java:150)
	at org.eclipse.lsp4e.outline.LSSymbolsContentProvider.refreshTreeContentFromLS(LSSymbolsContentProvider.java:303)
	at org.eclipse.lsp4e.outline.LSSymbolsContentProvider$ReconcilerOutlineUpdater.process(LSSymbolsContentProvider.java:156)
	at org.eclipse.jface.text.reconciler.AbstractReconciler$BackgroundThread.run(AbstractReconciler.java:207)

[ghentschke]

interesting, I'am working with the LSP4E sources in in my workspace. I'll try it

[ghentschke]

Btw I just tested using org.eclipse.lsp4e 0.15.0 instead of org.eclipse.lsp4e 0.16.0 and with the 0.15.0 version it seems to work without the delay of 5 seconds. However with 0.15.0 I get an exception

Do you set schedule the UIJob immediately:

		UIJob.create("Refresh Editors", monitor -> {
			int rangeOffset = textViewer.getTopIndexStartOffset();
			int rangeLength = textViewer.getBottomIndexEndOffset() - rangeOffset;
			editor.getSite().getPage().reuseEditor((IReusableEditor) editor, editor.getEditorInput());
			textViewer.revealRange(rangeOffset, rangeLength);
		}).schedule();

[ghentschke]

@ddscharfe Please rebase your PR to let the CI build run

[ghentschke]

I did the folowing with the LSP4E build form sources:

1. schedule refreshEditor without delay
2. Open IDE
3. Open c/c++ file in LSP editor
4. LSP server gets started
5. Delete compile_commands.json in c++ project
   The LS gets terminated:

I[12:12:20.481] --> reply:textDocument/documentSymbol("4") 0 ms, error: -32602: trying to get AST for non-added document
I[12:12:24.972] <-- shutdown("5")
I[12:12:24.972] --> reply:shutdown("5") 0 ms
I[12:12:24.973] <-- exit
I[12:12:24.973] LSP finished, exiting with status 0

With a 5 sec schedule delay it works. After re-build of the project an a newly created compile_commands.json the editor gets refreshed after 5 sec

[ddscharfe]

Btw I just tested using org.eclipse.lsp4e 0.15.0 instead of org.eclipse.lsp4e 0.16.0 and with the 0.15.0 version it seems to work without the delay of 5 seconds. However with 0.15.0 I get an exception

Do you set schedule the UIJob immediately:

		UIJob.create("Refresh Editors", monitor -> {
			int rangeOffset = textViewer.getTopIndexStartOffset();
			int rangeLength = textViewer.getBottomIndexEndOffset() - rangeOffset;
			editor.getSite().getPage().reuseEditor((IReusableEditor) editor, editor.getEditorInput());
			textViewer.revealRange(rangeOffset, rangeLength);
		}).schedule();

Yes.

[ghentschke]

LGTM. Could you provide test for this feature?

[ddscharfe]

I could add a few integration tests, this would probably make sense. I'd also like to have unit tests, for which using dependency injection and mocking would be helpful. Would you mind using google guice in this project and mockito for the tests?

[ghentschke]

Would you mind using google guice in this project and mockito for the tests?

Mockito is fine. I'am not sure about google guice, since it should be a CDT feature in the future. We should ask @jonahgraham regarding this.

[ddscharfe]

I asked this question to @jonahgraham a few years ago: https://www.eclipse.org/lists/cdt-dev/msg31411.html
So I think having guice as dependency wouldn't be a problem, it's more a design choice, I think. I like to use dependency injection for my projects I think it helps having less coupling between classes, however it does add some kind of complexity.

[ghentschke]

I like to use dependency injection for my projects I think it helps having less coupling between classes, however it does add some kind of complexity.

Yepp, we do this in our projects as well.