Merge branch 'main' of https://github.com/RooVetGit/Roo-Code into upstream-pull-27-feb-2025

Author: Himanshu-Singh-Chauhan

CHANGELOG.md

@@ -1,5 +1,13 @@
 # Roo Code Changelog
 
+## [3.7.6]
+
+- Handle really long text better in the in the ChatRow similar to TaskHeader (thanks @joemanley201!)
+- Support multiple files in drag-and-drop
+- Truncate search_file output to avoid crashing the extension
+- Better OpenRouter error handling (no more "Provider Error")
+- Add slider to control max output tokens for thinking models
+
 ## [3.7.5]
 
 - Fix context window truncation math (see [#1173](https://github.com/RooVetGit/Roo-Code/issues/1173))

package-lock.json

@@ -3,7 +3,7 @@
 	"displayName": "PearAI Roo Code / Cline",
 	"description": "PearAI's integration of Roo Code / Cline, a coding agent.",
 	"publisher": "PearAI",
-	"version": "3.7.5",
+	"version": "3.7.6",
 	"icon": "assets/icons/pear.png",
 	"galleryBanner": {
 		"color": "#617A91",

package.json

@@ -31,7 +31,7 @@ export class AnthropicHandler implements ApiHandler, SingleCompletionHandler {
 		let stream: AnthropicStream<Anthropic.Messages.RawMessageStreamEvent>
 		const cacheControl: CacheControlEphemeral = { type: "ephemeral" }
 		let { id: modelId, info: modelInfo } = this.getModel()
-		const maxTokens = modelInfo.maxTokens || 8192
+		const maxTokens = this.options.modelMaxTokens || modelInfo.maxTokens || 8192
 		let temperature = this.options.modelTemperature ?? ANTHROPIC_DEFAULT_TEMPERATURE
 		let thinking: BetaThinkingConfigParam | undefined = undefined
 
@@ -41,7 +41,15 @@ export class AnthropicHandler implements ApiHandler, SingleCompletionHandler {
 			// `claude-3-7-sonnet-20250219` model with a thinking budget.
 			// We can handle this more elegantly in the future.
 			modelId = "claude-3-7-sonnet-20250219"
-			const budgetTokens = this.options.anthropicThinking ?? Math.max(maxTokens * 0.8, 1024)
+
+			// Clamp the thinking budget to be at most 80% of max tokens and at
+			// least 1024 tokens.
+			const maxBudgetTokens = Math.floor(maxTokens * 0.8)
+			const budgetTokens = Math.max(
+				Math.min(this.options.anthropicThinking ?? maxBudgetTokens, maxBudgetTokens),
+				1024,
+			)
+
 			thinking = { type: "enabled", budget_tokens: budgetTokens }
 			temperature = 1.0
 		}

src/api/providers/anthropic.ts

@@ -108,12 +108,19 @@ export class OpenRouterHandler implements ApiHandler, SingleCompletionHandler {
 			topP = 0.95
 		}
 
+		const maxTokens = this.options.modelMaxTokens || modelInfo.maxTokens
 		let temperature = this.options.modelTemperature ?? defaultTemperature
 		let thinking: BetaThinkingConfigParam | undefined = undefined
 
 		if (modelInfo.thinking) {
-			const maxTokens = modelInfo.maxTokens || 8192
-			const budgetTokens = this.options.anthropicThinking ?? Math.max(maxTokens * 0.8, 1024)
+			// Clamp the thinking budget to be at most 80% of max tokens and at
+			// least 1024 tokens.
+			const maxBudgetTokens = Math.floor((maxTokens || 8192) * 0.8)
+			const budgetTokens = Math.max(
+				Math.min(this.options.anthropicThinking ?? maxBudgetTokens, maxBudgetTokens),
+				1024,
+			)
+
 			thinking = { type: "enabled", budget_tokens: budgetTokens }
 			temperature = 1.0
 		}
@@ -271,7 +278,7 @@ export async function getOpenRouterModels() {
 					modelInfo.supportsPromptCache = true
 					modelInfo.cacheWritesPrice = 3.75
 					modelInfo.cacheReadsPrice = 0.3
-					modelInfo.maxTokens = 16384
+					modelInfo.maxTokens = 64_000
 					break
 				case rawModel.id.startsWith("anthropic/claude-3.5-sonnet-20240620"):
 					modelInfo.supportsPromptCache = true

src/api/providers/openrouter.ts

@@ -87,6 +87,7 @@ export type ClineOptions = {
 
 export class Cline {
 	readonly taskId: string
+	readonly apiConfiguration: ApiConfiguration
 	api: ApiHandler
 	private terminalManager: TerminalManager
 	private urlContentFetcher: UrlContentFetcher
@@ -148,6 +149,7 @@ export class Cline {
 		}
 
 		this.taskId = crypto.randomUUID()
+		this.apiConfiguration = apiConfiguration
 		this.api = buildApiHandler(apiConfiguration)
 		this.terminalManager = new TerminalManager()
 		this.urlContentFetcher = new UrlContentFetcher(provider.context)
@@ -961,13 +963,21 @@ export class Cline {
 				cacheWrites = 0,
 				cacheReads = 0,
 			}: ClineApiReqInfo = JSON.parse(previousRequest)
+
 			const totalTokens = tokensIn + tokensOut + cacheWrites + cacheReads
 
-			const trimmedMessages = truncateConversationIfNeeded(
-				this.apiConversationHistory,
+			const modelInfo = this.api.getModel().info
+			const maxTokens = modelInfo.thinking
+				? this.apiConfiguration.modelMaxTokens || modelInfo.maxTokens
+				: modelInfo.maxTokens
+			const contextWindow = modelInfo.contextWindow
+
+			const trimmedMessages = truncateConversationIfNeeded({
+				messages: this.apiConversationHistory,
 				totalTokens,
-				this.api.getModel().info,
-			)
+				maxTokens,
+				contextWindow,
+			})
 
 			if (trimmedMessages !== this.apiConversationHistory) {
 				await this.overwriteApiConversationHistory(trimmedMessages)
@@ -1010,7 +1020,7 @@ export class Cline {
 		} catch (error) {
 			// note that this api_req_failed ask is unique in that we only present this option if the api hasn't streamed any content yet (ie it fails on the first chunk due), as it would allow them to hit a retry button. However if the api failed mid-stream, it could be in any arbitrary state where some tools may have executed, so that error is handled differently and requires cancelling the task entirely.
 			if (alwaysApproveResubmit) {
-				const errorMsg = error.message ?? "Unknown error"
+				const errorMsg = error.error?.metadata?.raw ?? error.message ?? "Unknown error"
 				const baseDelay = requestDelaySeconds || 5
 				const exponentialDelay = Math.ceil(baseDelay * Math.pow(2, retryAttempt))
 				// Wait for the greater of the exponential delay or the rate limit delay

src/core/Cline.ts

@@ -119,11 +119,21 @@ describe("getMaxTokens", () => {
 		// Max tokens = 100000 - 50000 = 50000
 
 		// Below max tokens - no truncation
-		const result1 = truncateConversationIfNeeded(messages, 49999, modelInfo)
+		const result1 = truncateConversationIfNeeded({
+			messages,
+			totalTokens: 49999,
+			contextWindow: modelInfo.contextWindow,
+			maxTokens: modelInfo.maxTokens,
+		})
 		expect(result1).toEqual(messages)
 
 		// Above max tokens - truncate
-		const result2 = truncateConversationIfNeeded(messages, 50001, modelInfo)
+		const result2 = truncateConversationIfNeeded({
+			messages,
+			totalTokens: 50001,
+			contextWindow: modelInfo.contextWindow,
+			maxTokens: modelInfo.maxTokens,
+		})
 		expect(result2).not.toEqual(messages)
 		expect(result2.length).toBe(3) // Truncated with 0.5 fraction
 	})
@@ -133,11 +143,21 @@ describe("getMaxTokens", () => {
 		// Max tokens = 100000 - (100000 * 0.2) = 80000
 
 		// Below max tokens - no truncation
-		const result1 = truncateConversationIfNeeded(messages, 79999, modelInfo)
+		const result1 = truncateConversationIfNeeded({
+			messages,
+			totalTokens: 79999,
+			contextWindow: modelInfo.contextWindow,
+			maxTokens: modelInfo.maxTokens,
+		})
 		expect(result1).toEqual(messages)
 
 		// Above max tokens - truncate
-		const result2 = truncateConversationIfNeeded(messages, 80001, modelInfo)
+		const result2 = truncateConversationIfNeeded({
+			messages,
+			totalTokens: 80001,
+			contextWindow: modelInfo.contextWindow,
+			maxTokens: modelInfo.maxTokens,
+		})
 		expect(result2).not.toEqual(messages)
 		expect(result2.length).toBe(3) // Truncated with 0.5 fraction
 	})
@@ -147,11 +167,21 @@ describe("getMaxTokens", () => {
 		// Max tokens = 50000 - 10000 = 40000
 
 		// Below max tokens - no truncation
-		const result1 = truncateConversationIfNeeded(messages, 39999, modelInfo)
+		const result1 = truncateConversationIfNeeded({
+			messages,
+			totalTokens: 39999,
+			contextWindow: modelInfo.contextWindow,
+			maxTokens: modelInfo.maxTokens,
+		})
 		expect(result1).toEqual(messages)
 
 		// Above max tokens - truncate
-		const result2 = truncateConversationIfNeeded(messages, 40001, modelInfo)
+		const result2 = truncateConversationIfNeeded({
+			messages,
+			totalTokens: 40001,
+			contextWindow: modelInfo.contextWindow,
+			maxTokens: modelInfo.maxTokens,
+		})
 		expect(result2).not.toEqual(messages)
 		expect(result2.length).toBe(3) // Truncated with 0.5 fraction
 	})
@@ -161,11 +191,21 @@ describe("getMaxTokens", () => {
 		// Max tokens = 200000 - 30000 = 170000
 
 		// Below max tokens - no truncation
-		const result1 = truncateConversationIfNeeded(messages, 169999, modelInfo)
+		const result1 = truncateConversationIfNeeded({
+			messages,
+			totalTokens: 169999,
+			contextWindow: modelInfo.contextWindow,
+			maxTokens: modelInfo.maxTokens,
+		})
 		expect(result1).toEqual(messages)
 
 		// Above max tokens - truncate
-		const result2 = truncateConversationIfNeeded(messages, 170001, modelInfo)
+		const result2 = truncateConversationIfNeeded({
+			messages,
+			totalTokens: 170001,
+			contextWindow: modelInfo.contextWindow,
+			maxTokens: modelInfo.maxTokens,
+		})
 		expect(result2).not.toEqual(messages)
 		expect(result2.length).toBe(3) // Truncated with 0.5 fraction
 	})
@@ -194,7 +234,12 @@ describe("truncateConversationIfNeeded", () => {
 		const maxTokens = 100000 - 30000 // 70000
 		const totalTokens = 69999 // Below threshold
 
-		const result = truncateConversationIfNeeded(messages, totalTokens, modelInfo)
+		const result = truncateConversationIfNeeded({
+			messages,
+			totalTokens,
+			contextWindow: modelInfo.contextWindow,
+			maxTokens: modelInfo.maxTokens,
+		})
 		expect(result).toEqual(messages) // No truncation occurs
 	})
 
@@ -207,7 +252,12 @@ describe("truncateConversationIfNeeded", () => {
 		// With 4 messages after the first, 0.5 fraction means remove 2 messages
 		const expectedResult = [messages[0], messages[3], messages[4]]
 
-		const result = truncateConversationIfNeeded(messages, totalTokens, modelInfo)
+		const result = truncateConversationIfNeeded({
+			messages,
+			totalTokens,
+			contextWindow: modelInfo.contextWindow,
+			maxTokens: modelInfo.maxTokens,
+		})
 		expect(result).toEqual(expectedResult)
 	})
 
@@ -218,14 +268,38 @@ describe("truncateConversationIfNeeded", () => {
 
 		// Test below threshold
 		const belowThreshold = 69999
-		expect(truncateConversationIfNeeded(messages, belowThreshold, modelInfo1)).toEqual(
-			truncateConversationIfNeeded(messages, belowThreshold, modelInfo2),
+		expect(
+			truncateConversationIfNeeded({
+				messages,
+				totalTokens: belowThreshold,
+				contextWindow: modelInfo1.contextWindow,
+				maxTokens: modelInfo1.maxTokens,
+			}),
+		).toEqual(
+			truncateConversationIfNeeded({
+				messages,
+				totalTokens: belowThreshold,
+				contextWindow: modelInfo2.contextWindow,
+				maxTokens: modelInfo2.maxTokens,
+			}),
 		)
 
 		// Test above threshold
 		const aboveThreshold = 70001
-		expect(truncateConversationIfNeeded(messages, aboveThreshold, modelInfo1)).toEqual(
-			truncateConversationIfNeeded(messages, aboveThreshold, modelInfo2),
+		expect(
+			truncateConversationIfNeeded({
+				messages,
+				totalTokens: aboveThreshold,
+				contextWindow: modelInfo1.contextWindow,
+				maxTokens: modelInfo1.maxTokens,
+			}),
+		).toEqual(
+			truncateConversationIfNeeded({
+				messages,
+				totalTokens: aboveThreshold,
+				contextWindow: modelInfo2.contextWindow,
+				maxTokens: modelInfo2.maxTokens,
+			}),
 		)
 	})
 })

src/core/sliding-window/__tests__/sliding-window.test.ts

@@ -1,7 +1,5 @@
 import { Anthropic } from "@anthropic-ai/sdk"
 
-import { ModelInfo } from "../../shared/api"
-
 /**
  * Truncates a conversation by removing a fraction of the messages.
  *
@@ -26,28 +24,29 @@ export function truncateConversation(
 }
 
 /**
- * Conditionally truncates the conversation messages if the total token count exceeds the model's limit.
+ * Conditionally truncates the conversation messages if the total token count
+ * exceeds the model's limit.
  *
  * @param {Anthropic.Messages.MessageParam[]} messages - The conversation messages.
  * @param {number} totalTokens - The total number of tokens in the conversation.
- * @param {ModelInfo} modelInfo - Model metadata including context window size.
+ * @param {number} contextWindow - The context window size.
+ * @param {number} maxTokens - The maximum number of tokens allowed.
  * @returns {Anthropic.Messages.MessageParam[]} The original or truncated conversation messages.
  */
-export function truncateConversationIfNeeded(
-	messages: Anthropic.Messages.MessageParam[],
-	totalTokens: number,
-	modelInfo: ModelInfo,
-): Anthropic.Messages.MessageParam[] {
-	return totalTokens < getMaxTokens(modelInfo) ? messages : truncateConversation(messages, 0.5)
+
+type TruncateOptions = {
+	messages: Anthropic.Messages.MessageParam[]
+	totalTokens: number
+	contextWindow: number
+	maxTokens?: number
 }
 
-/**
- * Calculates the maximum allowed tokens
- *
- * @param {ModelInfo} modelInfo - The model information containing the context window size.
- * @returns {number} The maximum number of tokens allowed
- */
-function getMaxTokens(modelInfo: ModelInfo): number {
-	// The buffer needs to be at least as large as `modelInfo.maxTokens`, or 20% of the context window if for some reason it's not set.
-	return modelInfo.contextWindow - (modelInfo.maxTokens || modelInfo.contextWindow * 0.2)
+export function truncateConversationIfNeeded({
+	messages,
+	totalTokens,
+	contextWindow,
+	maxTokens,
+}: TruncateOptions): Anthropic.Messages.MessageParam[] {
+	const allowedTokens = contextWindow - (maxTokens || contextWindow * 0.2)
+	return totalTokens < allowedTokens ? messages : truncateConversation(messages, 0.5)
 }