a84edc3267
Reproduces the infinite loop when the model returns finish_reason: "tool-calls" but emits zero actual tool call parts. Test 1 asserts the processor converts finish to "stop" (currently fails). Test 2 verifies valid tool-calls flows are preserved (passes).
230 lines
7.6 KiB
TypeScript
230 lines
7.6 KiB
TypeScript
import { describe, expect, mock, spyOn, test } from "bun:test"
|
|
|
|
mock.module("@/kilo-sessions/remote-sender", () => ({
|
|
RemoteSender: {
|
|
create() {
|
|
return {
|
|
queue() {},
|
|
flush: async () => undefined,
|
|
}
|
|
},
|
|
},
|
|
}))
|
|
|
|
import type { Provider } from "../../src/provider/provider"
|
|
import type { LLM as LLMType } from "../../src/session/llm"
|
|
import type { MessageV2 } from "../../src/session/message-v2"
|
|
import { Log } from "../../src/util/log"
|
|
import { tmpdir } from "../fixture/fixture"
|
|
|
|
Log.init({ print: false })
|
|
|
|
function model(): Provider.Model {
|
|
return {
|
|
id: "gpt-4",
|
|
providerID: "openai",
|
|
name: "GPT-4",
|
|
limit: {
|
|
context: 128000,
|
|
input: 0,
|
|
output: 4096,
|
|
},
|
|
cost: { input: 0, output: 0, cache: { read: 0, write: 0 } },
|
|
capabilities: {
|
|
toolcall: true,
|
|
attachment: false,
|
|
reasoning: false,
|
|
temperature: true,
|
|
input: { text: true, image: false, audio: false, video: false },
|
|
output: { text: true, image: false, audio: false, video: false },
|
|
},
|
|
api: { id: "openai", url: "https://api.openai.com/v1", npm: "@ai-sdk/openai" },
|
|
options: {},
|
|
headers: {},
|
|
} as Provider.Model
|
|
}
|
|
|
|
function stream(events: Array<Record<string, unknown>>): any {
|
|
return {
|
|
fullStream: (async function* () {
|
|
for (const e of events) yield e
|
|
})(),
|
|
}
|
|
}
|
|
|
|
describe("session processor empty tool-calls", () => {
|
|
test("converts finish to stop when model returns tool-calls with no tools", async () => {
|
|
const { Instance } = await import("../../src/project/instance")
|
|
const { LLM } = await import("../../src/session/llm")
|
|
const { Identifier } = await import("../../src/id/id")
|
|
const { MessageV2 } = await import("../../src/session/message-v2")
|
|
|
|
await using tmp = await tmpdir({ git: true })
|
|
|
|
await Instance.provide({
|
|
directory: tmp.path,
|
|
fn: async () => {
|
|
const { Session } = await import("../../src/session")
|
|
const { SessionProcessor } = await import("../../src/session/processor")
|
|
const m = model()
|
|
const session = await Session.create({})
|
|
const user = (await Session.updateMessage({
|
|
id: Identifier.ascending("message"),
|
|
role: "user",
|
|
sessionID: session.id,
|
|
time: { created: Date.now() },
|
|
agent: "code",
|
|
model: { providerID: m.providerID, modelID: m.id },
|
|
tools: {},
|
|
})) as MessageV2.User
|
|
const assistant = (await Session.updateMessage({
|
|
id: Identifier.ascending("message"),
|
|
parentID: user.id,
|
|
role: "assistant",
|
|
mode: "code",
|
|
agent: "code",
|
|
path: { cwd: Instance.directory, root: Instance.worktree },
|
|
cost: 0,
|
|
tokens: { input: 0, output: 0, reasoning: 0, cache: { read: 0, write: 0 } },
|
|
modelID: m.id,
|
|
providerID: m.providerID,
|
|
time: { created: Date.now() },
|
|
sessionID: session.id,
|
|
})) as MessageV2.Assistant
|
|
|
|
// Stream with finish_reason "tool-calls" but zero tool events
|
|
const llm = spyOn(LLM, "stream").mockResolvedValueOnce(
|
|
stream([
|
|
{ type: "start" },
|
|
{ type: "start-step" },
|
|
{
|
|
type: "finish-step",
|
|
finishReason: "tool-calls",
|
|
usage: { inputTokens: 100, completionTokens: 41, totalTokens: 141 },
|
|
providerMetadata: undefined,
|
|
},
|
|
{ type: "finish" },
|
|
]),
|
|
)
|
|
|
|
const processor = SessionProcessor.create({
|
|
assistantMessage: assistant,
|
|
sessionID: session.id,
|
|
model: m,
|
|
abort: AbortSignal.any([]),
|
|
})
|
|
const inp: LLMType.StreamInput = {
|
|
user,
|
|
sessionID: session.id,
|
|
model: m,
|
|
agent: { name: "code", mode: "primary", permission: [], options: {} } as any,
|
|
system: [],
|
|
abort: AbortSignal.any([]),
|
|
messages: [],
|
|
tools: {},
|
|
}
|
|
|
|
try {
|
|
const result = await processor.process(inp)
|
|
// The processor must convert "tool-calls" → "stop" when no tools were emitted
|
|
expect(processor.message.finish).toBe("stop")
|
|
// Verify no tool parts exist on the message
|
|
const parts = await MessageV2.parts(assistant.id)
|
|
const tools = parts.filter((p: any) => p.type === "tool")
|
|
expect(tools.length).toBe(0)
|
|
} finally {
|
|
llm.mockRestore()
|
|
}
|
|
},
|
|
})
|
|
})
|
|
|
|
test("preserves tool-calls finish when tool parts exist", async () => {
|
|
const { Instance } = await import("../../src/project/instance")
|
|
const { LLM } = await import("../../src/session/llm")
|
|
const { Identifier } = await import("../../src/id/id")
|
|
const { MessageV2 } = await import("../../src/session/message-v2")
|
|
|
|
await using tmp = await tmpdir({ git: true })
|
|
|
|
await Instance.provide({
|
|
directory: tmp.path,
|
|
fn: async () => {
|
|
const { Session } = await import("../../src/session")
|
|
const { SessionProcessor } = await import("../../src/session/processor")
|
|
const m = model()
|
|
const session = await Session.create({})
|
|
const user = (await Session.updateMessage({
|
|
id: Identifier.ascending("message"),
|
|
role: "user",
|
|
sessionID: session.id,
|
|
time: { created: Date.now() },
|
|
agent: "code",
|
|
model: { providerID: m.providerID, modelID: m.id },
|
|
tools: {},
|
|
})) as MessageV2.User
|
|
const assistant = (await Session.updateMessage({
|
|
id: Identifier.ascending("message"),
|
|
parentID: user.id,
|
|
role: "assistant",
|
|
mode: "code",
|
|
agent: "code",
|
|
path: { cwd: Instance.directory, root: Instance.worktree },
|
|
cost: 0,
|
|
tokens: { input: 0, output: 0, reasoning: 0, cache: { read: 0, write: 0 } },
|
|
modelID: m.id,
|
|
providerID: m.providerID,
|
|
time: { created: Date.now() },
|
|
sessionID: session.id,
|
|
})) as MessageV2.Assistant
|
|
|
|
// Stream with finish_reason "tool-calls" AND a tool-input-start event
|
|
const llm = spyOn(LLM, "stream").mockResolvedValueOnce(
|
|
stream([
|
|
{ type: "start" },
|
|
{ type: "start-step" },
|
|
{ type: "tool-input-start", id: "call_1", toolName: "test_tool" },
|
|
{
|
|
type: "finish-step",
|
|
finishReason: "tool-calls",
|
|
usage: { inputTokens: 100, completionTokens: 41, totalTokens: 141 },
|
|
providerMetadata: undefined,
|
|
},
|
|
{ type: "finish" },
|
|
]),
|
|
)
|
|
|
|
const processor = SessionProcessor.create({
|
|
assistantMessage: assistant,
|
|
sessionID: session.id,
|
|
model: m,
|
|
abort: AbortSignal.any([]),
|
|
})
|
|
const inp: LLMType.StreamInput = {
|
|
user,
|
|
sessionID: session.id,
|
|
model: m,
|
|
agent: { name: "code", mode: "primary", permission: [], options: {} } as any,
|
|
system: [],
|
|
abort: AbortSignal.any([]),
|
|
messages: [],
|
|
tools: {},
|
|
}
|
|
|
|
try {
|
|
const result = await processor.process(inp)
|
|
// finish must remain "tool-calls" because a tool part WAS created
|
|
expect(processor.message.finish).toBe("tool-calls")
|
|
expect(result).toBe("continue")
|
|
// Verify the tool part exists
|
|
const parts = await MessageV2.parts(assistant.id)
|
|
const tools = parts.filter((p: any) => p.type === "tool")
|
|
expect(tools.length).toBe(1)
|
|
} finally {
|
|
llm.mockRestore()
|
|
}
|
|
},
|
|
})
|
|
})
|
|
})
|