mirror of
https://github.com/RooVetGit/Roo-Code.git
synced 2026-09-07 08:26:51 +00:00
fix: handle undefined content in Gemini assistant messages
- Ensure content is always a string (empty string instead of undefined) for Gemini models - Add thought signature injection for Gemini models with native tool protocol - Add comprehensive tests for Gemini thought signature handling The Gemini API requires assistant message content to be a string, not undefined. When messages only contain tool_calls without text content, convertToOpenAiMessages() sets content to undefined, causing a 400 error.
This commit is contained in:
daniel-lxs
parent
495664c261
commit
162811abb7
2 changed files with 368 additions and 2 deletions
|
|
@ -123,6 +123,26 @@ vitest.mock("../../providers/fetchers/modelCache", () => ({
|
|||
outputPrice: 4,
|
||||
defaultToolProtocol: "native",
|
||||
},
|
||||
"google/gemini-2.5-pro": {
|
||||
maxTokens: 65_536,
|
||||
contextWindow: 1_048_576,
|
||||
supportsImages: true,
|
||||
supportsPromptCache: true,
|
||||
supportsNativeTools: true,
|
||||
inputPrice: 1.25,
|
||||
outputPrice: 5,
|
||||
defaultToolProtocol: "native",
|
||||
},
|
||||
"gemini-2.5-flash": {
|
||||
maxTokens: 65_536,
|
||||
contextWindow: 1_048_576,
|
||||
supportsImages: true,
|
||||
supportsPromptCache: true,
|
||||
supportsNativeTools: true,
|
||||
inputPrice: 0.15,
|
||||
outputPrice: 0.6,
|
||||
defaultToolProtocol: "native",
|
||||
},
|
||||
}
|
||||
}
|
||||
return {}
|
||||
|
|
@ -1066,4 +1086,294 @@ describe("RooHandler", () => {
|
|||
expect(endChunks[0].id).toBe("call_finish_test")
|
||||
})
|
||||
})
|
||||
|
||||
describe("Gemini thought signature handling", () => {
|
||||
it("should inject reasoning_details with reasoning.encrypted for Gemini models with tool calls", async () => {
|
||||
const geminiHandler = new RooHandler({
|
||||
apiModelId: "google/gemini-2.5-pro",
|
||||
})
|
||||
|
||||
// Messages with tool calls in history
|
||||
const messagesWithToolCalls: Anthropic.Messages.MessageParam[] = [
|
||||
{ role: "user", content: "Read the file test.ts" },
|
||||
{
|
||||
role: "assistant",
|
||||
content: [
|
||||
{ type: "text", text: "I'll read the file for you." },
|
||||
{
|
||||
type: "tool_use",
|
||||
id: "call_123",
|
||||
name: "read_file",
|
||||
input: { path: "test.ts" },
|
||||
},
|
||||
],
|
||||
},
|
||||
{
|
||||
role: "user",
|
||||
content: [
|
||||
{
|
||||
type: "tool_result",
|
||||
tool_use_id: "call_123",
|
||||
content: "file content here",
|
||||
},
|
||||
],
|
||||
},
|
||||
{ role: "user", content: "What's in the file?" },
|
||||
]
|
||||
|
||||
const stream = geminiHandler.createMessage(systemPrompt, messagesWithToolCalls)
|
||||
for await (const _chunk of stream) {
|
||||
// Consume stream
|
||||
}
|
||||
|
||||
// Verify the messages sent to the API
|
||||
expect(mockCreate).toHaveBeenCalled()
|
||||
const callArgs = mockCreate.mock.calls[0][0]
|
||||
const messages = callArgs.messages
|
||||
|
||||
// Find the assistant message with tool calls
|
||||
const assistantMessage = messages.find(
|
||||
(m: any) => m.role === "assistant" && m.tool_calls && m.tool_calls.length > 0,
|
||||
)
|
||||
|
||||
expect(assistantMessage).toBeDefined()
|
||||
expect(assistantMessage.reasoning_details).toBeDefined()
|
||||
expect(assistantMessage.reasoning_details).toHaveLength(1)
|
||||
expect(assistantMessage.reasoning_details[0]).toMatchObject({
|
||||
id: "call_123",
|
||||
type: "reasoning.encrypted",
|
||||
data: "skip_thought_signature_validator",
|
||||
index: 0,
|
||||
})
|
||||
})
|
||||
|
||||
it("should inject reasoning_details for Gemini model without google/ prefix", async () => {
|
||||
const geminiHandler = new RooHandler({
|
||||
apiModelId: "gemini-2.5-flash",
|
||||
})
|
||||
|
||||
// Messages with tool calls in history
|
||||
const messagesWithToolCalls: Anthropic.Messages.MessageParam[] = [
|
||||
{ role: "user", content: "List files" },
|
||||
{
|
||||
role: "assistant",
|
||||
content: [
|
||||
{
|
||||
type: "tool_use",
|
||||
id: "call_456",
|
||||
name: "list_files",
|
||||
input: { path: "." },
|
||||
},
|
||||
],
|
||||
},
|
||||
{
|
||||
role: "user",
|
||||
content: [
|
||||
{
|
||||
type: "tool_result",
|
||||
tool_use_id: "call_456",
|
||||
content: "file1.ts\nfile2.ts",
|
||||
},
|
||||
],
|
||||
},
|
||||
]
|
||||
|
||||
const stream = geminiHandler.createMessage(systemPrompt, messagesWithToolCalls)
|
||||
for await (const _chunk of stream) {
|
||||
// Consume stream
|
||||
}
|
||||
|
||||
// Verify the messages sent to the API
|
||||
expect(mockCreate).toHaveBeenCalled()
|
||||
const callArgs = mockCreate.mock.calls[0][0]
|
||||
const messages = callArgs.messages
|
||||
|
||||
// Find the assistant message with tool calls
|
||||
const assistantMessage = messages.find(
|
||||
(m: any) => m.role === "assistant" && m.tool_calls && m.tool_calls.length > 0,
|
||||
)
|
||||
|
||||
expect(assistantMessage).toBeDefined()
|
||||
expect(assistantMessage.reasoning_details).toBeDefined()
|
||||
expect(assistantMessage.reasoning_details[0].type).toBe("reasoning.encrypted")
|
||||
})
|
||||
|
||||
it("should not inject reasoning_details for non-Gemini models", async () => {
|
||||
const nonGeminiHandler = new RooHandler({
|
||||
apiModelId: "anthropic/claude-haiku-4.5",
|
||||
})
|
||||
|
||||
// Messages with tool calls in history
|
||||
const messagesWithToolCalls: Anthropic.Messages.MessageParam[] = [
|
||||
{ role: "user", content: "Read the file test.ts" },
|
||||
{
|
||||
role: "assistant",
|
||||
content: [
|
||||
{
|
||||
type: "tool_use",
|
||||
id: "call_789",
|
||||
name: "read_file",
|
||||
input: { path: "test.ts" },
|
||||
},
|
||||
],
|
||||
},
|
||||
{
|
||||
role: "user",
|
||||
content: [
|
||||
{
|
||||
type: "tool_result",
|
||||
tool_use_id: "call_789",
|
||||
content: "file content here",
|
||||
},
|
||||
],
|
||||
},
|
||||
]
|
||||
|
||||
const stream = nonGeminiHandler.createMessage(systemPrompt, messagesWithToolCalls)
|
||||
for await (const _chunk of stream) {
|
||||
// Consume stream
|
||||
}
|
||||
|
||||
// Verify the messages sent to the API
|
||||
expect(mockCreate).toHaveBeenCalled()
|
||||
const callArgs = mockCreate.mock.calls[0][0]
|
||||
const messages = callArgs.messages
|
||||
|
||||
// Find the assistant message with tool calls
|
||||
const assistantMessage = messages.find(
|
||||
(m: any) => m.role === "assistant" && m.tool_calls && m.tool_calls.length > 0,
|
||||
)
|
||||
|
||||
expect(assistantMessage).toBeDefined()
|
||||
// Should NOT have reasoning_details injected
|
||||
expect(assistantMessage.reasoning_details).toBeUndefined()
|
||||
})
|
||||
|
||||
it("should not inject reasoning_details if already present with encrypted type", async () => {
|
||||
const geminiHandler = new RooHandler({
|
||||
apiModelId: "google/gemini-2.5-pro",
|
||||
})
|
||||
|
||||
// Messages with tool calls that already have reasoning_details
|
||||
const messagesWithExistingDetails: Anthropic.Messages.MessageParam[] = [
|
||||
{ role: "user", content: "Read the file" },
|
||||
{
|
||||
role: "assistant",
|
||||
content: [
|
||||
{
|
||||
type: "tool_use",
|
||||
id: "call_existing",
|
||||
name: "read_file",
|
||||
input: { path: "test.ts" },
|
||||
},
|
||||
],
|
||||
// @ts-expect-error - reasoning_details is not in the Anthropic type
|
||||
reasoning_details: [
|
||||
{
|
||||
type: "reasoning.encrypted",
|
||||
data: "existing_signature",
|
||||
format: "google-gemini-v1",
|
||||
index: 0,
|
||||
},
|
||||
],
|
||||
},
|
||||
{
|
||||
role: "user",
|
||||
content: [
|
||||
{
|
||||
type: "tool_result",
|
||||
tool_use_id: "call_existing",
|
||||
content: "file content",
|
||||
},
|
||||
],
|
||||
},
|
||||
]
|
||||
|
||||
const stream = geminiHandler.createMessage(systemPrompt, messagesWithExistingDetails)
|
||||
for await (const _chunk of stream) {
|
||||
// Consume stream
|
||||
}
|
||||
|
||||
// Verify the messages sent to the API
|
||||
expect(mockCreate).toHaveBeenCalled()
|
||||
const callArgs = mockCreate.mock.calls[0][0]
|
||||
const messages = callArgs.messages
|
||||
|
||||
// Find the assistant message with tool calls
|
||||
const assistantMessage = messages.find(
|
||||
(m: any) => m.role === "assistant" && m.tool_calls && m.tool_calls.length > 0,
|
||||
)
|
||||
|
||||
expect(assistantMessage).toBeDefined()
|
||||
// Should keep existing reasoning_details without duplicating
|
||||
expect(assistantMessage.reasoning_details).toHaveLength(1)
|
||||
expect(assistantMessage.reasoning_details[0].data).toBe("existing_signature")
|
||||
})
|
||||
|
||||
it("should handle multiple tool calls with individual reasoning_details entries", async () => {
|
||||
const geminiHandler = new RooHandler({
|
||||
apiModelId: "google/gemini-2.5-pro",
|
||||
})
|
||||
|
||||
// Messages with multiple tool calls
|
||||
const messagesWithMultipleToolCalls: Anthropic.Messages.MessageParam[] = [
|
||||
{ role: "user", content: "Read both files" },
|
||||
{
|
||||
role: "assistant",
|
||||
content: [
|
||||
{
|
||||
type: "tool_use",
|
||||
id: "call_multi_1",
|
||||
name: "read_file",
|
||||
input: { path: "file1.ts" },
|
||||
},
|
||||
{
|
||||
type: "tool_use",
|
||||
id: "call_multi_2",
|
||||
name: "read_file",
|
||||
input: { path: "file2.ts" },
|
||||
},
|
||||
],
|
||||
},
|
||||
{
|
||||
role: "user",
|
||||
content: [
|
||||
{
|
||||
type: "tool_result",
|
||||
tool_use_id: "call_multi_1",
|
||||
content: "content 1",
|
||||
},
|
||||
{
|
||||
type: "tool_result",
|
||||
tool_use_id: "call_multi_2",
|
||||
content: "content 2",
|
||||
},
|
||||
],
|
||||
},
|
||||
]
|
||||
|
||||
const stream = geminiHandler.createMessage(systemPrompt, messagesWithMultipleToolCalls)
|
||||
for await (const _chunk of stream) {
|
||||
// Consume stream
|
||||
}
|
||||
|
||||
// Verify the messages sent to the API
|
||||
expect(mockCreate).toHaveBeenCalled()
|
||||
const callArgs = mockCreate.mock.calls[0][0]
|
||||
const messages = callArgs.messages
|
||||
|
||||
// Find the assistant message with tool calls
|
||||
const assistantMessage = messages.find(
|
||||
(m: any) => m.role === "assistant" && m.tool_calls && m.tool_calls.length > 0,
|
||||
)
|
||||
|
||||
expect(assistantMessage).toBeDefined()
|
||||
expect(assistantMessage.reasoning_details).toBeDefined()
|
||||
expect(assistantMessage.reasoning_details).toHaveLength(2)
|
||||
expect(assistantMessage.reasoning_details[0].id).toBe("call_multi_1")
|
||||
expect(assistantMessage.reasoning_details[0].index).toBe(0)
|
||||
expect(assistantMessage.reasoning_details[1].id).toBe("call_multi_2")
|
||||
expect(assistantMessage.reasoning_details[1].index).toBe(1)
|
||||
})
|
||||
})
|
||||
})
|
||||
|
|
|
|||
|
|
@ -1,7 +1,7 @@
|
|||
import { Anthropic } from "@anthropic-ai/sdk"
|
||||
import OpenAI from "openai"
|
||||
|
||||
import { rooDefaultModelId, getApiProtocol, type ImageGenerationApiMethod } from "@roo-code/types"
|
||||
import { rooDefaultModelId, getApiProtocol, type ImageGenerationApiMethod, TOOL_PROTOCOL } from "@roo-code/types"
|
||||
import { NativeToolCallParser } from "../../core/assistant-message/NativeToolCallParser"
|
||||
import { CloudService } from "@roo-code/cloud"
|
||||
|
||||
|
|
@ -12,6 +12,7 @@ import { getModelParams } from "../transform/model-params"
|
|||
import { convertToOpenAiMessages } from "../transform/openai-format"
|
||||
import type { RooReasoningParams } from "../transform/reasoning"
|
||||
import { getRooReasoning } from "../transform/reasoning"
|
||||
import { resolveToolProtocol } from "../../utils/resolveToolProtocol"
|
||||
|
||||
import type { ApiHandlerCreateMessageMetadata } from "../index"
|
||||
import { BaseOpenAiCompatibleProvider } from "./base-openai-compatible-provider"
|
||||
|
|
@ -96,11 +97,66 @@ export class RooHandler extends BaseOpenAiCompatibleProvider<string> {
|
|||
const max_tokens = params.maxTokens ?? undefined
|
||||
const temperature = params.temperature ?? this.defaultTemperature
|
||||
|
||||
let openAiMessages = [{ role: "system" as const, content: systemPrompt }, ...convertToOpenAiMessages(messages)]
|
||||
|
||||
// Process reasoning_details when switching models to Gemini for native tool call compatibility
|
||||
// This is the same approach used in OpenRouter - see openrouter.ts lines 246-280
|
||||
const toolProtocol = resolveToolProtocol(this.options, info)
|
||||
const isNativeProtocol = toolProtocol === TOOL_PROTOCOL.NATIVE
|
||||
const isGemini = model.startsWith("google/gemini") || model.includes("gemini")
|
||||
|
||||
// For Gemini models: ensure content is always a string (not undefined)
|
||||
// The Gemini API requires content to be a string, even if empty
|
||||
if (isGemini) {
|
||||
openAiMessages = openAiMessages.map((msg) => {
|
||||
if (msg.role === "assistant" && msg.content === undefined) {
|
||||
return { ...msg, content: "" }
|
||||
}
|
||||
return msg
|
||||
})
|
||||
}
|
||||
|
||||
// For Gemini with native protocol: inject fake reasoning.encrypted blocks for tool calls
|
||||
// This is required when switching from other models to Gemini to satisfy API validation
|
||||
if (isNativeProtocol && isGemini) {
|
||||
openAiMessages = openAiMessages.map((msg) => {
|
||||
if (msg.role === "assistant") {
|
||||
const toolCalls = (msg as any).tool_calls as any[] | undefined
|
||||
const existingDetails = (msg as any).reasoning_details as any[] | undefined
|
||||
|
||||
// Only inject if there are tool calls and no existing encrypted reasoning
|
||||
if (toolCalls && toolCalls.length > 0) {
|
||||
const hasEncrypted = existingDetails?.some((d) => d.type === "reasoning.encrypted") ?? false
|
||||
|
||||
if (!hasEncrypted) {
|
||||
// Filter tool calls that have valid IDs
|
||||
const validToolCalls = toolCalls.filter((tc) => tc.id && typeof tc.id === "string")
|
||||
|
||||
if (validToolCalls.length > 0) {
|
||||
const fakeEncrypted = validToolCalls.map((tc, idx) => ({
|
||||
id: tc.id,
|
||||
type: "reasoning.encrypted",
|
||||
data: "skip_thought_signature_validator",
|
||||
index: (existingDetails?.length ?? 0) + idx,
|
||||
}))
|
||||
|
||||
return {
|
||||
...msg,
|
||||
reasoning_details: [...(existingDetails ?? []), ...fakeEncrypted],
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
return msg
|
||||
})
|
||||
}
|
||||
|
||||
const rooParams: RooChatCompletionParams = {
|
||||
model,
|
||||
max_tokens,
|
||||
temperature,
|
||||
messages: [{ role: "system", content: systemPrompt }, ...convertToOpenAiMessages(messages)],
|
||||
messages: openAiMessages,
|
||||
stream: true,
|
||||
stream_options: { include_usage: true },
|
||||
...(reasoning && { reasoning }),
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue