Don’t implicitly cancel requests when an edit is detected

Sources/BuildServerIntegration/BuildServerManager.swift

@@ -706,7 +706,7 @@ package actor BuildServerManager: QueueBasedMessageHandler {
   package func handle<Request: RequestType>(
     request: Request,
     id: RequestID,
-    reply: @Sendable @escaping (LSPResult<Request.Response>) -> Void
+    reply: @Sendable @escaping (Result<Request.Response, any Error>) -> Void
   ) async {
     let request = RequestAndReply(request, reply: reply)
     switch request {

Sources/BuildServerIntegration/BuiltInBuildServerAdapter.swift

@@ -120,7 +120,7 @@ actor BuiltInBuildServerAdapter: QueueBasedMessageHandler {
   func handle<Request: RequestType>(
     request: Request,
     id: RequestID,
-    reply: @Sendable @escaping (LSPResult<Request.Response>) -> Void
+    reply: @Sendable @escaping (Result<Request.Response, any Error>) -> Void
   ) async {
     let request = RequestAndReply(request, reply: reply)
     await buildServerHooks.preHandleRequest?(request.params)

Sources/SKOptions/SourceKitLSPOptions.swift

@@ -382,7 +382,7 @@ public struct SourceKitLSPOptions: Sendable, Codable, Equatable {
   public var cancelTextDocumentRequestsOnEditAndClose: Bool? = nil
 
   public var cancelTextDocumentRequestsOnEditAndCloseOrDefault: Bool {
-    return cancelTextDocumentRequestsOnEditAndClose ?? true
+    return cancelTextDocumentRequestsOnEditAndClose ?? false
   }
 
   /// Experimental features that are enabled.

Sources/SourceKitLSP/SourceKitLSPServer.swift

@@ -656,10 +656,25 @@ extension SourceKitLSPServer: QueueBasedMessageHandler {
   /// - Important: Should be invoked on `textDocumentTrackingQueue` to ensure that new text document requests are
   ///   registered before a notification that triggers cancellation might come in.
   private func cancelTextDocumentRequests(for uri: DocumentURI, reason: ImplicitTextDocumentRequestCancellationReason) {
-    guard self.options.cancelTextDocumentRequestsOnEditAndCloseOrDefault else {
-      return
-    }
+    let staleRequestSupport = self.capabilityRegistry?.clientCapabilities.general?.staleRequestSupport
     for (requestID, requestMethod) in self.inProgressTextDocumentRequests[uri, default: []] {
+      // Implicitly cancel text document requests if:
+      //  - We have enabled implicit text document request cancellation in the SourceKit options
+      //  - The client has indicated that it does not cancel stale requests. Defaults to `true` because this is an
+      //    option introduced in LSP 3.17 and most clients cancel requests diligently without setting
+      //    `staleRequestSupport.cancel = true`
+      //  - `staleRequestSupport.retryOnContentModified` contains this request method. Documentation for this behavior
+      //    is very limited but it appears that if a request is included in that array, the client (VS Code in
+      //    particular) expects to receive a `ContentModified` response when the LSP server detects an edit, in which
+      //    case it will re-run the request with the new file contents.
+      guard
+        self.options.cancelTextDocumentRequestsOnEditAndCloseOrDefault
+          || !(staleRequestSupport?.cancel ?? true)
+          || staleRequestSupport?.retryOnContentModified.contains(requestMethod) ?? false
+      else {
+        continue
+      }
+
       if reason == .documentChanged && requestMethod == CompletionRequest.method {
         // As the user types, we filter the code completion results. Cancelling the completion request on every
         // keystroke means that we will never build the initial list of completion results for this code
@@ -668,7 +683,7 @@ extension SourceKitLSPServer: QueueBasedMessageHandler {
         continue
       }
       logger.info("Implicitly cancelling request \(requestID)")
-      self.messageHandlingHelper.cancelRequest(id: requestID)
+      self.messageHandlingHelper.cancelRequest(id: requestID, error: .contentModified)
     }
   }
 
@@ -722,7 +737,7 @@ extension SourceKitLSPServer: QueueBasedMessageHandler {
   package func handle<Request: RequestType>(
     request params: Request,
     id: RequestID,
-    reply: @Sendable @escaping (LSPResult<Request.Response>) -> Void
+    reply: @Sendable @escaping (Result<Request.Response, any Error>) -> Void
   ) async {
     defer {
       if let request = params as? any TextDocumentRequest {

Tests/SourceKitLSPTests/SemanticTokensTests.swift

@@ -986,14 +986,22 @@ final class SemanticTokensTests: SourceKitLSPTestCase {
             try? await Task.sleep(for: .seconds(1))
           }
         }
-      })
+      }),
+      capabilities: ClientCapabilities(
+        general: .init(
+          staleRequestSupport: .init(
+            cancel: true,
+            retryOnContentModified: [DocumentSemanticTokensRequest.method]
+          )
+        )
+      )
     )
     let uri = DocumentURI(for: .swift)
     let positions = testClient.openDocument("1️⃣", uri: uri)
 
     let receivedSemanticTokensResponse = self.expectation(description: "Received semantic tokens response")
     testClient.send(DocumentSemanticTokensRequest(textDocument: TextDocumentIdentifier(uri))) { result in
-      XCTAssertEqual(result, .failure(ResponseError.cancelled))
+      XCTAssertEqual(result, .failure(ResponseError(code: .contentModified, message: "content modified")))
       receivedSemanticTokensResponse.fulfill()
     }
     testClient.send(