refactor(lsp): slightly reorganize diagnostics debounce logic (#9796)

This patch doesn't actually fix the bug I was hoping to fix, which is
that `update_diagnostics()` sometimes gets called even when there are
more updates that should be processed first. I did eventually figure out
that this issue is caused by Tokio's cooperative yielding, which
currently can't be disabled.

However overall it makes the debounce code somewhat more readable IMO,
which is why I'm suggesting to land it anyway.

cli/lsp/diagnostics.rs

@@ -24,12 +24,9 @@ use std::sync::Arc;
 use std::thread;
 use tokio::sync::mpsc;
 use tokio::sync::oneshot;
-use tokio::time;
-
-// 150ms between keystrokes is about 45 WPM, so we want something that is longer
-// than that, but not too long to introduce detectable UI delay.  200ms is a
-// decent compromise.
-const DIAGNOSTIC_DEBOUNCE_MS: u64 = 200;
+use tokio::time::sleep;
+use tokio::time::Duration;
+use tokio::time::Instant;
 
 #[derive(Debug, Clone, Hash, PartialEq, Eq)]
 pub enum DiagnosticSource {
@@ -202,34 +199,6 @@ async fn update_diagnostics(
   }
 }
 
-fn handle_request(
-  maybe_request: Option<DiagnosticRequest>,
-  collection: &mut DiagnosticCollection,
-  dirty: &mut bool,
-) -> bool {
-  match maybe_request {
-    Some(request) => {
-      match request {
-        DiagnosticRequest::Get(specifier, source, tx) => {
-          let diagnostics = collection
-            .diagnostics_for(&specifier, &source)
-            .cloned()
-            .collect();
-          if tx.send(diagnostics).is_err() {
-            error!("DiagnosticServer unable to send response on channel.");
-          }
-        }
-        DiagnosticRequest::Invalidate(specifier) => {
-          collection.invalidate(&specifier)
-        }
-        DiagnosticRequest::Update => *dirty = true,
-      }
-      true
-    }
-    _ => false,
-  }
-}
-
 /// A server which calculates diagnostics in its own thread and publishes them
 /// to an LSP client.
 #[derive(Debug)]
@@ -256,36 +225,53 @@ impl DiagnosticsServer {
       let mut collection = DiagnosticCollection::default();
 
       runtime.block_on(async {
-        // Some(snapshot) is the flag we use to determine if something has
-        // changed where we will wait for the timeout of changes or a request
-        // that forces us to update diagnostics
+        // Debounce timer delay. 150ms between keystrokes is about 45 WPM, so we
+        // want something that is longer than that, but not too long to
+        // introduce detectable UI delay; 200ms is a decent compromise.
+        const DELAY: Duration = Duration::from_millis(200);
+        // If the debounce timer isn't active, it will be set to expire "never",
+        // which is actually just 1 year in the future.
+        const NEVER: Duration = Duration::from_secs(365 * 24 * 60 * 60);
+
+        // A flag that is set whenever something has changed that requires the
+        // diagnostics collection to be updated.
         let mut dirty = false;
 
+        let debounce_timer = sleep(NEVER);
+        tokio::pin!(debounce_timer);
+
         loop {
-          let next = rx.recv();
-          tokio::pin!(next);
-
-          let duration = if dirty {
-            time::Duration::from_millis(DIAGNOSTIC_DEBOUNCE_MS)
-          } else {
-            // we need to await an arbitrary silly amount of time, so this is
-            // 1 year in seconds
-            time::Duration::new(31_622_400, 0)
-          };
-
           // "race" the next message off the rx queue or the debounce timer.
-          // if the next message comes off the queue, the next iteration of the
-          // loop will reset the debounce future.  When the debounce future
-          // occurs, the diagnostics will be updated based on the snapshot that
-          // is retrieved, thereby "skipping" all the interim state changes.
+          // The debounce timer gets reset every time a message comes off the
+          // queue. When the debounce timer expires, a snapshot of the most
+          // up-to-date state is used to produce diagnostics.
           tokio::select! {
-            _ = time::sleep(duration) => {
-              if dirty {
+            maybe_request = rx.recv() => {
+              use DiagnosticRequest::*;
+              match maybe_request {
+                None => break, // Request channel closed.
+                Some(Get(specifier, source, tx)) => {
+                  let diagnostics = collection
+                    .diagnostics_for(&specifier, &source)
+                    .cloned()
+                    .collect();
+                  // If this fails, the requestor disappeared; not a problem.
+                  let _ = tx.send(diagnostics);
+                }
+                Some(Invalidate(specifier)) => {
+                  collection.invalidate(&specifier);
+                }
+                Some(Update) => {
+                  dirty = true;
+                  debounce_timer.as_mut().reset(Instant::now() + DELAY);
+                }
+              }
+            }
+            _ = debounce_timer.as_mut(), if dirty => {
               dirty = false;
-                let snapshot = {
-                  // make sure the lock drops
-                  language_server.lock().await.snapshot()
-                };
+              debounce_timer.as_mut().reset(Instant::now() + NEVER);
+
+              let snapshot = language_server.lock().await.snapshot();
               update_diagnostics(
                 &client,
                 &mut collection,
@@ -293,16 +279,6 @@ impl DiagnosticsServer {
                 &ts_server
               ).await;
             }
-              let maybe_request = next.await;
-              if !handle_request(maybe_request, &mut collection, &mut dirty) {
-                break;
-              }
-            }
-            maybe_request = &mut next => {
-              if !handle_request(maybe_request, &mut collection, &mut dirty) {
-                break;
-              }
-            }
           }
         }
       })