fix: handle GitHub Copilot gpt-4.1 model ID response bug

src/api/providers/__tests__/vscode-lm.spec.ts

@@ -300,4 +300,152 @@ describe("VsCodeLmHandler", () => {
 			await expect(promise).rejects.toThrow("VSCode LM completion error: Completion failed")
 		})
 	})
+
+	describe("GitHub Copilot gpt-4.1 bug handling", () => {
+		it("should skip chunks that contain only the model ID", async () => {
+			const mockModel = {
+				...mockLanguageModelChat,
+				id: "copilot/gpt-4.1",
+				vendor: "copilot",
+				family: "gpt-4.1",
+			}
+			;(vscode.lm.selectChatModels as Mock).mockResolvedValueOnce([mockModel])
+
+			const systemPrompt = "You are a helpful assistant"
+			const messages: Anthropic.Messages.MessageParam[] = [
+				{
+					role: "user" as const,
+					content: "Hello",
+				},
+			]
+
+			// Mock a response that includes the model ID as a chunk followed by actual content
+			mockModel.sendRequest.mockResolvedValueOnce({
+				stream: (async function* () {
+					yield new vscode.LanguageModelTextPart("copilot/gpt-4.1") // Bug: model ID as response
+					yield new vscode.LanguageModelTextPart("Hello! How can I help you?") // Actual content
+					return
+				})(),
+				text: (async function* () {
+					yield "copilot/gpt-4.1"
+					yield "Hello! How can I help you?"
+					return
+				})(),
+			})
+
+			// Override the default client with our test client
+			handler["client"] = mockModel
+
+			const stream = handler.createMessage(systemPrompt, messages)
+			const chunks = []
+			for await (const chunk of stream) {
+				chunks.push(chunk)
+			}
+
+			// Should only have the actual content and usage, not the model ID
+			expect(chunks).toHaveLength(2)
+			expect(chunks[0]).toEqual({
+				type: "text",
+				text: "Hello! How can I help you?",
+			})
+			expect(chunks[1]).toMatchObject({
+				type: "usage",
+				inputTokens: expect.any(Number),
+				outputTokens: expect.any(Number),
+			})
+		})
+
+		it("should throw error when only model ID is returned", async () => {
+			const mockModel = {
+				...mockLanguageModelChat,
+				id: "copilot/gpt-4.1",
+				vendor: "copilot",
+				family: "gpt-4.1",
+			}
+			;(vscode.lm.selectChatModels as Mock).mockResolvedValueOnce([mockModel])
+
+			const systemPrompt = "You are a helpful assistant"
+			const messages: Anthropic.Messages.MessageParam[] = [
+				{
+					role: "user" as const,
+					content: "Hello",
+				},
+			]
+
+			// Mock a response that only contains the model ID
+			mockModel.sendRequest.mockResolvedValueOnce({
+				stream: (async function* () {
+					yield new vscode.LanguageModelTextPart("copilot/gpt-4.1")
+					return
+				})(),
+				text: (async function* () {
+					yield "copilot/gpt-4.1"
+					return
+				})(),
+			})
+
+			// Override the default client with our test client
+			handler["client"] = mockModel
+
+			const stream = handler.createMessage(systemPrompt, messages)
+
+			// Collect all chunks and expect an error
+			await expect(async () => {
+				const chunks = []
+				for await (const chunk of stream) {
+					chunks.push(chunk)
+				}
+			}).rejects.toThrow(
+				'The VS Code Language Model API returned only the model ID "copilot/gpt-4.1" instead of generating a response',
+			)
+		})
+
+		it("should handle variations of model ID patterns", async () => {
+			const mockModel = {
+				...mockLanguageModelChat,
+				id: "github/gpt-4.1",
+				vendor: "github",
+				family: "gpt-4.1",
+			}
+			;(vscode.lm.selectChatModels as Mock).mockResolvedValueOnce([mockModel])
+
+			const systemPrompt = "You are a helpful assistant"
+			const messages: Anthropic.Messages.MessageParam[] = [
+				{
+					role: "user" as const,
+					content: "Hello",
+				},
+			]
+
+			// Mock a response with different model ID variations
+			mockModel.sendRequest.mockResolvedValueOnce({
+				stream: (async function* () {
+					yield new vscode.LanguageModelTextPart("  GitHub/gpt-4.1  ") // With spaces and different case
+					yield new vscode.LanguageModelTextPart("Actual response content")
+					return
+				})(),
+				text: (async function* () {
+					yield "  GitHub/gpt-4.1  "
+					yield "Actual response content"
+					return
+				})(),
+			})
+
+			// Override the default client with our test client
+			handler["client"] = mockModel
+
+			const stream = handler.createMessage(systemPrompt, messages)
+			const chunks = []
+			for await (const chunk of stream) {
+				chunks.push(chunk)
+			}
+
+			// Should skip the model ID and only return actual content
+			expect(chunks).toHaveLength(2)
+			expect(chunks[0]).toEqual({
+				type: "text",
+				text: "Actual response content",
+			})
+		})
+	})
 })

src/api/providers/vscode-lm.ts

@@ -363,6 +363,8 @@ export class VsCodeLmHandler extends BaseProvider implements SingleCompletionHan
 
 		// Accumulate the text and count at the end of the stream to reduce token counting overhead.
 		let accumulatedText: string = ""
+		let hasValidContent = false
+		let detectedModelIdAsResponse = false
 
 		try {
 			// Create the response stream with minimal required options
@@ -382,12 +384,39 @@ export class VsCodeLmHandler extends BaseProvider implements SingleCompletionHan
 			// Consume the stream and handle both text and tool call chunks
 			for await (const chunk of response.stream) {
 				if (chunk instanceof vscode.LanguageModelTextPart) {
+					// Debug logging for GitHub Copilot gpt-4.1 issue
+					if (client.id && (client.id.includes("gpt-4.1") || client.id.includes("copilot"))) {
+						console.debug("Roo Code <Language Model API>: Processing chunk for model:", client.id, {
+							chunkType: typeof chunk,
+							chunkValue: chunk.value,
+							chunkValueType: typeof chunk.value,
+							chunkValueLength: chunk.value?.length,
+							isModelId: chunk.value === client.id || chunk.value === `${client.vendor}/${client.family}`,
+						})
+					}
+
 					// Validate text part value
 					if (typeof chunk.value !== "string") {
 						console.warn("Roo Code <Language Model API>: Invalid text part value received:", chunk.value)
 						continue
 					}
 
+					// Check if the chunk value is just the model ID (GitHub Copilot gpt-4.1 bug)
+					const modelIdPattern = /^(copilot|github)\/gpt-4\.1$/i
+					const trimmedValue = chunk.value.trim()
+
+					if (modelIdPattern.test(trimmedValue) || trimmedValue === client.id) {
+						console.warn(
+							"Roo Code <Language Model API>: Detected model ID as response content, this appears to be a VS Code LM API bug:",
+							chunk.value,
+						)
+						detectedModelIdAsResponse = true
+						// Skip this chunk as it's not actual content
+						continue
+					}
+
+					// If we get here, we have valid content
+					hasValidContent = true
 					accumulatedText += chunk.value
 					yield {
 						type: "text",
@@ -444,6 +473,13 @@ export class VsCodeLmHandler extends BaseProvider implements SingleCompletionHan
 				}
 			}
 
+			// If we only received the model ID and no valid content, throw an error
+			if (detectedModelIdAsResponse && !hasValidContent) {
+				throw new Error(
+					`Roo Code <Language Model API>: The VS Code Language Model API returned only the model ID "${client.id}" instead of generating a response. This is a known issue with GitHub Copilot's gpt-4.1 model. Please try using a different model or wait for a fix from the VS Code/GitHub Copilot team.`,
+				)
+			}
+
 			// Count tokens in the accumulated text after stream completion
 			const totalOutputTokens: number = await this.internalCountTokens(accumulatedText)