Document diagnostics not cleared on file close when workspace diagnostics enabled

This follows on from #1013, where it was fixed that document diagnostics are cleared when a file closed.

This works fine when a server does not support workspace diagnostics.

There, it was mentioned by @aeschli:

@JohnnyMorganz You should looking into using workspace diagnostics

• Workspace diagnostics are shown for all resources, unless the ones open in an editor
• Document diagnostics are shown for the resources in open editors. Your workspace diagnostics would not contain diagnostics for ignored resources, but the your document diagnostics would.

So I followed these steps. I implemented workspace diagnostics, and my workspace diagnostics do not follow include ignored resources - but document diagnostics do.

However, if I have workspace diagnostics enabled, the document diagnostics do not clear on file close.

I updated my pull diagnostics minimal repro (https://github.com/JohnnyMorganz/pull-diagnostics-bug) with the following diff:

diff --git a/server/src/server.ts b/server/src/server.ts
index 1069d4a..6a8646b 100644
--- a/server/src/server.ts
+++ b/server/src/server.ts
@@ -30,7 +30,7 @@ connection.onInitialize((params: InitializeParams) => {
       textDocumentSync: TextDocumentSyncKind.Incremental,
       diagnosticProvider: {
         interFileDependencies: true,
-        workspaceDiagnostics: false,
+        workspaceDiagnostics: true,
       },
       workspace: {
         workspaceFolders: {
@@ -80,6 +80,12 @@ connection.languages.diagnostics.on((params) => {
   };
 });

+connection.languages.diagnostics.onWorkspace(() => {
+  return {
+    items: [],
+  };
+});
+
 // Make the text document manager listen on the connection
 // for open, change and close text document events
 documents.listen(connection);

[pisv]

@dbaeumer Thank you for the clarification!

[dbaeumer]

The benefit comes with pull on documents. The old push model had no knowledge about which files to priorities. Hence it could easily happen that you saw stale errors for some time. The pull model allows to combine a long running pull with single document pulls hence knowing which once to prioritize.

The solution you implemented is in line with the specification. I will close the issue.