Follow the fix in stream.ts: prompt_tokens (42) includes cached_tokens (5), so input = 42 - 5 - 10 = 27. Use cache_write_tokens (not cacheable_tokens) for cacheWrite, and keep cacheable_tokens:99 in the payload to prove it is no longer mapped to cacheWrite.
227 lines
7.5 KiB
TypeScript
227 lines
7.5 KiB
TypeScript
import type {
|
|
Api,
|
|
AssistantMessage,
|
|
AssistantMessageEvent,
|
|
AssistantMessageEventStream,
|
|
Context,
|
|
Model,
|
|
} from "@earendil-works/pi-ai";
|
|
import { afterEach, describe, expect, it, vi } from "vitest";
|
|
import { streamQoder } from "../stream.js";
|
|
|
|
/**
|
|
* Build a single SSE `data:` line carrying a Qoder envelope:
|
|
* { headers, body: <JSON string>, statusCodeValue, statusCode }
|
|
* The server wraps the OpenAI-style chunk inside `body` as a JSON string.
|
|
*/
|
|
function sseEnvelope(body: object, statusCodeValue = 200, statusCode = "OK"): string {
|
|
return (
|
|
"data:" +
|
|
JSON.stringify({
|
|
headers: { "Content-Type": ["application/json"] },
|
|
body: JSON.stringify(body),
|
|
statusCodeValue,
|
|
statusCode,
|
|
}) +
|
|
"\n\n"
|
|
);
|
|
}
|
|
|
|
const DONE_SSE =
|
|
"data:" +
|
|
JSON.stringify({
|
|
headers: { "Content-Type": ["application/json"] },
|
|
body: "[DONE]",
|
|
statusCodeValue: 200,
|
|
statusCode: "OK",
|
|
}) +
|
|
"\n\n";
|
|
|
|
function chunk(delta: object, extra: object = {}): object {
|
|
return {
|
|
choices: [{ delta, index: 0 }],
|
|
created: 1,
|
|
id: "test-id",
|
|
model: "auto",
|
|
object: "chat.completion.chunk",
|
|
...extra,
|
|
};
|
|
}
|
|
|
|
function finishChunk(finish_reason: string, extra: object = {}): object {
|
|
return {
|
|
choices: [{ finish_reason, index: 0 }],
|
|
created: 1,
|
|
id: "test-id",
|
|
model: "auto",
|
|
object: "chat.completion.chunk",
|
|
usage: { completion_tokens: 1, prompt_tokens: 1, total_tokens: 2 },
|
|
...extra,
|
|
};
|
|
}
|
|
|
|
const SUCCESS_SSE =
|
|
sseEnvelope(chunk({ role: "assistant" })) +
|
|
sseEnvelope(chunk({ reasoning_content: "The user wants OK.", role: "assistant" })) +
|
|
sseEnvelope(chunk({ content: "OK", role: "assistant" })) +
|
|
sseEnvelope(finishChunk("stop")) +
|
|
DONE_SSE;
|
|
|
|
const BLOCKED_SSE = sseEnvelope(
|
|
{ code: "provider_error", message: "Session blocked", request_id: "r", type: "provider_error" },
|
|
406,
|
|
"Not Acceptable",
|
|
);
|
|
|
|
function mockFetch(body: string): typeof fetch {
|
|
const response = new Response(body, {
|
|
status: 200,
|
|
headers: { "content-type": "text/event-stream" },
|
|
});
|
|
return vi.fn(async () => response) as unknown as typeof fetch;
|
|
}
|
|
|
|
function makeModel(): Model<Api> {
|
|
return { id: "ultimate", api: "qoder-api" as Api, provider: "qoder" } as Model<Api>;
|
|
}
|
|
|
|
function makeContext(): Context {
|
|
return {
|
|
systemPrompt: "test",
|
|
messages: [{ role: "user", content: "hi" }],
|
|
tools: [],
|
|
} as unknown as Context;
|
|
}
|
|
|
|
async function consume(stream: AssistantMessageEventStream): Promise<AssistantMessageEvent[]> {
|
|
const events: AssistantMessageEvent[] = [];
|
|
for await (const ev of stream) {
|
|
events.push(ev);
|
|
if (ev.type === "done" || ev.type === "error") break;
|
|
}
|
|
return events;
|
|
}
|
|
|
|
describe("streamQoder", () => {
|
|
const originalFetch = globalThis.fetch;
|
|
afterEach(() => {
|
|
globalThis.fetch = originalFetch;
|
|
vi.restoreAllMocks();
|
|
});
|
|
|
|
it("parses a successful SSE stream into text + stop", async () => {
|
|
globalThis.fetch = mockFetch(SUCCESS_SSE);
|
|
const stream = streamQoder(makeModel(), makeContext(), { apiKey: "fake" });
|
|
const events = await consume(stream);
|
|
|
|
const done = events.find((e) => e.type === "done");
|
|
expect(done, "expected a done event").toBeDefined();
|
|
const msg = (done as { message: AssistantMessage }).message;
|
|
expect(msg.stopReason).toBe("stop");
|
|
const text = msg.content.find((c) => c.type === "text");
|
|
expect(text && "text" in text ? text.text : "").toBe("OK");
|
|
});
|
|
|
|
it("surfaces an upstream 406 'Session blocked' as an error event, not a silent stop", async () => {
|
|
globalThis.fetch = mockFetch(BLOCKED_SSE);
|
|
const stream = streamQoder(makeModel(), makeContext(), { apiKey: "fake" });
|
|
const events = await consume(stream);
|
|
|
|
const err = events.find((e) => e.type === "error");
|
|
expect(err, "expected an error event").toBeDefined();
|
|
const msg = (err as { error: AssistantMessage }).error;
|
|
expect(msg.stopReason).toBe("error");
|
|
expect(msg.errorMessage).toMatch(/Session blocked/);
|
|
expect(msg.errorMessage).toMatch(/406/);
|
|
// Must NOT emit a silent done/stop.
|
|
expect(events.find((e) => e.type === "done")).toBeUndefined();
|
|
});
|
|
|
|
it("preserves finish_reason=length instead of overwriting to stop", async () => {
|
|
const sse =
|
|
sseEnvelope(chunk({ content: "partial", role: "assistant" })) + sseEnvelope(finishChunk("length")) + DONE_SSE;
|
|
globalThis.fetch = mockFetch(sse);
|
|
const stream = streamQoder(makeModel(), makeContext(), { apiKey: "fake" });
|
|
const events = await consume(stream);
|
|
|
|
const done = events.find((e) => e.type === "done");
|
|
const msg = (done as { message: AssistantMessage }).message;
|
|
expect(msg.stopReason).toBe("length");
|
|
});
|
|
|
|
it("captures usage, responseId and responseModel from the finish chunk", async () => {
|
|
const sse =
|
|
sseEnvelope(chunk({ content: "OK", role: "assistant" })) +
|
|
sseEnvelope(
|
|
finishChunk("stop", {
|
|
id: "chatcmpl-abc123",
|
|
model: "qmodel_latest",
|
|
usage: {
|
|
prompt_tokens: 42,
|
|
completion_tokens: 7,
|
|
total_tokens: 49,
|
|
completion_tokens_details: { reasoning_tokens: 3 },
|
|
// prompt_tokens (42) INCLUDES cached_tokens (5) per OpenAI
|
|
// semantics; pi-core expects `input` to exclude them
|
|
// (promptTokens = input + cacheRead + cacheWrite), so input =
|
|
// 42 - 5 - 10 = 27. cacheable_tokens is a capacity metric, not a
|
|
// write count, and must not be mapped to cacheWrite.
|
|
prompt_tokens_details: { cacheable_tokens: 99, cache_write_tokens: 10, cached_tokens: 5 },
|
|
},
|
|
}),
|
|
) +
|
|
DONE_SSE;
|
|
globalThis.fetch = mockFetch(sse);
|
|
const stream = streamQoder(makeModel(), makeContext(), { apiKey: "fake" });
|
|
const events = await consume(stream);
|
|
|
|
const done = events.find((e) => e.type === "done");
|
|
const msg = (done as { message: AssistantMessage }).message;
|
|
expect(msg.responseId).toBe("chatcmpl-abc123");
|
|
expect(msg.responseModel).toBe("qmodel_latest");
|
|
expect(msg.usage.input).toBe(27);
|
|
expect(msg.usage.output).toBe(7);
|
|
expect(msg.usage.totalTokens).toBe(49);
|
|
expect(msg.usage.cacheRead).toBe(5);
|
|
expect(msg.usage.cacheWrite).toBe(10);
|
|
});
|
|
|
|
it("emits a done event with reason=length when finish_reason is length", async () => {
|
|
const sse =
|
|
sseEnvelope(chunk({ content: "partial", role: "assistant" })) + sseEnvelope(finishChunk("length")) + DONE_SSE;
|
|
globalThis.fetch = mockFetch(sse);
|
|
const stream = streamQoder(makeModel(), makeContext(), { apiKey: "fake" });
|
|
const events = await consume(stream);
|
|
|
|
const done = events.find((e) => e.type === "done");
|
|
expect(done, "expected a done event").toBeDefined();
|
|
expect((done as { reason: string }).reason).toBe("length");
|
|
});
|
|
|
|
it("reports a tool_use stop reason when the stream emits tool calls", async () => {
|
|
const sse =
|
|
sseEnvelope(
|
|
chunk({
|
|
tool_calls: [
|
|
{
|
|
index: 0,
|
|
id: "call_1",
|
|
function: { name: "bash", arguments: '{"command":"ls"}' },
|
|
},
|
|
],
|
|
}),
|
|
) +
|
|
sseEnvelope(finishChunk("tool_calls")) +
|
|
DONE_SSE;
|
|
globalThis.fetch = mockFetch(sse);
|
|
const stream = streamQoder(makeModel(), makeContext(), { apiKey: "fake" });
|
|
const events = await consume(stream);
|
|
|
|
const done = events.find((e) => e.type === "done");
|
|
const msg = (done as { message: AssistantMessage }).message;
|
|
expect(msg.stopReason).toBe("toolUse");
|
|
const toolCall = msg.content.find((c) => c.type === "toolCall");
|
|
expect(toolCall).toBeDefined();
|
|
});
|
|
});
|