opencode-qwen-cli-auth 2.2.5 → 2.2.7

package/dist/index.js

@@ -15,10 +15,12 @@ import { PROVIDER_ID, AUTH_LABELS, DEVICE_FLOW, PORTAL_HEADERS } from "./lib/con
 import { logError, logInfo, logWarn, LOGGING_ENABLED } from "./lib/logger.js";
 const CHAT_REQUEST_TIMEOUT_MS = 30000;
 const CHAT_MAX_RETRIES = 0;
+const CHAT_MAX_TOKENS_CAP = 2048;
 const MAX_CONSECUTIVE_POLL_FAILURES = 3;
 const QUOTA_DEGRADE_MAX_TOKENS = 1024;
-const CLI_FALLBACK_TIMEOUT_MS = 45000;
+const CLI_FALLBACK_TIMEOUT_MS = 8000;
 const CLI_FALLBACK_MAX_BUFFER_CHARS = 1024 * 1024;
+const ENABLE_CLI_FALLBACK = process.env.OPENCODE_QWEN_ENABLE_CLI_FALLBACK === "1";
 const PLUGIN_USER_AGENT = "opencode-qwen-cli-auth/2.2.1";
 const CLIENT_ONLY_BODY_FIELDS = new Set([
     "providerID",
@@ -178,6 +180,14 @@ function sanitizeOutgoingPayload(payload) {
         delete sanitized.stream_options;
         changed = true;
     }
+    if (typeof sanitized.max_tokens === "number" && sanitized.max_tokens > CHAT_MAX_TOKENS_CAP) {
+        sanitized.max_tokens = CHAT_MAX_TOKENS_CAP;
+        changed = true;
+    }
+    if (typeof sanitized.max_completion_tokens === "number" && sanitized.max_completion_tokens > CHAT_MAX_TOKENS_CAP) {
+        sanitized.max_completion_tokens = CHAT_MAX_TOKENS_CAP;
+        changed = true;
+    }
     return changed ? sanitized : payload;
 }
 function createQuotaDegradedPayload(payload) {
@@ -309,7 +319,10 @@ function extractQwenCliText(events) {
     }
     return null;
 }
-function makeQwenCliCompletionResponse(model, content, context) {
+function createSseResponseChunk(data) {
+    return `data: ${JSON.stringify(data)}\n\n`;
+}
+function makeQwenCliCompletionResponse(model, content, context, streamMode) {
     if (LOGGING_ENABLED) {
         logInfo("Qwen CLI fallback returned completion", {
             request_id: context.requestId,
@@ -317,6 +330,51 @@ function makeQwenCliCompletionResponse(model, content, context) {
             modelID: model,
         });
     }
+    if (streamMode) {
+        const completionId = `chatcmpl-${randomUUID()}`;
+        const created = Math.floor(Date.now() / 1000);
+        const encoder = new TextEncoder();
+        const stream = new ReadableStream({
+            start(controller) {
+                controller.enqueue(encoder.encode(createSseResponseChunk({
+                    id: completionId,
+                    object: "chat.completion.chunk",
+                    created,
+                    model,
+                    choices: [
+                        {
+                            index: 0,
+                            delta: { role: "assistant", content },
+                            finish_reason: null,
+                        },
+                    ],
+                })));
+                controller.enqueue(encoder.encode(createSseResponseChunk({
+                    id: completionId,
+                    object: "chat.completion.chunk",
+                    created,
+                    model,
+                    choices: [
+                        {
+                            index: 0,
+                            delta: {},
+                            finish_reason: "stop",
+                        },
+                    ],
+                })));
+                controller.enqueue(encoder.encode("data: [DONE]\n\n"));
+                controller.close();
+            },
+        });
+        return new Response(stream, {
+            status: 200,
+            headers: {
+                "content-type": "text/event-stream; charset=utf-8",
+                "cache-control": "no-cache",
+                "x-qwen-cli-fallback": "1",
+            },
+        });
+    }
     const body = {
         id: `chatcmpl-${randomUUID()}`,
         object: "chat.completion",
@@ -346,8 +404,9 @@ function makeQwenCliCompletionResponse(model, content, context) {
         },
     });
 }
-async function runQwenCliFallback(payload, context) {
+async function runQwenCliFallback(payload, context, abortSignal) {
     const model = typeof payload?.model === "string" && payload.model.length > 0 ? payload.model : "coder-model";
+    const streamMode = payload?.stream === true;
     const prompt = buildQwenCliPrompt(payload);
     const args = [prompt, "-o", "json", "--max-session-turns", "1", "--model", model];
     if (LOGGING_ENABLED) {
@@ -363,6 +422,8 @@ async function runQwenCliFallback(payload, context) {
         let stdout = "";
         let stderr = "";
         let timer = null;
+        let child = undefined;
+        let abortHandler = undefined;
         const useShell = shouldUseShell(QWEN_CLI_COMMAND);
         const finalize = (result) => {
             if (settled) {
@@ -372,9 +433,18 @@ async function runQwenCliFallback(payload, context) {
             if (timer) {
                 clearTimeout(timer);
             }
+            if (abortSignal && abortHandler) {
+                abortSignal.removeEventListener("abort", abortHandler);
+            }
             resolve(result);
         };
-        let child;
+        if (abortSignal?.aborted) {
+            finalize({
+                ok: false,
+                reason: "cli_aborted",
+            });
+            return;
+        }
         try {
             child = spawn(QWEN_CLI_COMMAND, args, {
                 shell: useShell,
@@ -389,6 +459,20 @@ async function runQwenCliFallback(payload, context) {
             });
             return;
         }
+        if (abortSignal) {
+            abortHandler = () => {
+                try {
+                    child?.kill();
+                }
+                catch (_killError) {
+                }
+                finalize({
+                    ok: false,
+                    reason: "cli_aborted",
+                });
+            };
+            abortSignal.addEventListener("abort", abortHandler, { once: true });
+        }
         timer = setTimeout(() => {
             try {
                 child.kill();
@@ -418,7 +502,7 @@ async function runQwenCliFallback(payload, context) {
             if (content) {
                 finalize({
                     ok: true,
-                    response: makeQwenCliCompletionResponse(model, content, context),
+                    response: makeQwenCliCompletionResponse(model, content, context, streamMode),
                 });
                 return;
             }
@@ -470,6 +554,7 @@ async function sendWithTimeout(input, requestInit) {
 }
 async function failFastFetch(input, init) {
     const requestInit = init ? { ...init } : {};
+    const sourceSignal = requestInit.signal;
     const rawPayload = parseJsonRequestBody(requestInit);
     const sessionID = typeof rawPayload?.sessionID === "string" ? rawPayload.sessionID : undefined;
     let payload = rawPayload;
@@ -532,10 +617,34 @@ async function failFastFetch(input, init) {
                         return response;
                     }
                     const fallbackBody = await response.text().catch(() => "");
-                    const cliFallback = await runQwenCliFallback(payload, context);
+                    if (ENABLE_CLI_FALLBACK) {
+                        const cliFallback = await runQwenCliFallback(payload, context, sourceSignal);
+                        if (cliFallback.ok) {
+                            return cliFallback.response;
+                        }
+                        if (cliFallback.reason === "cli_aborted") {
+                            return makeFailFastErrorResponse(400, "request_aborted", "Qwen request was aborted");
+                        }
+                        if (LOGGING_ENABLED) {
+                            logWarn("Qwen CLI fallback failed", {
+                                request_id: context.requestId,
+                                sessionID: context.sessionID,
+                                modelID: context.modelID,
+                                reason: cliFallback.reason,
+                                stderr: cliFallback.stderr,
+                            });
+                        }
+                    }
+                    return makeQuotaFailFastResponse(fallbackBody, response.headers, context);
+                }
+                if (ENABLE_CLI_FALLBACK) {
+                    const cliFallback = await runQwenCliFallback(payload, context, sourceSignal);
                     if (cliFallback.ok) {
                         return cliFallback.response;
                     }
+                    if (cliFallback.reason === "cli_aborted") {
+                        return makeFailFastErrorResponse(400, "request_aborted", "Qwen request was aborted");
+                    }
                     if (LOGGING_ENABLED) {
                         logWarn("Qwen CLI fallback failed", {
                             request_id: context.requestId,
@@ -545,20 +654,6 @@ async function failFastFetch(input, init) {
                             stderr: cliFallback.stderr,
                         });
                     }
-                    return makeQuotaFailFastResponse(fallbackBody, response.headers, context);
-                }
-                const cliFallback = await runQwenCliFallback(payload, context);
-                if (cliFallback.ok) {
-                    return cliFallback.response;
-                }
-                if (LOGGING_ENABLED) {
-                    logWarn("Qwen CLI fallback failed", {
-                        request_id: context.requestId,
-                        sessionID: context.sessionID,
-                        modelID: context.modelID,
-                        reason: cliFallback.reason,
-                        stderr: cliFallback.stderr,
-                    });
                 }
             }
             return makeQuotaFailFastResponse(firstBody, response.headers, context);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "opencode-qwen-cli-auth",
-  "version": "2.2.5",
+  "version": "2.2.7",
   "description": "Qwen OAuth authentication plugin for opencode - use your Qwen account instead of API keys",
   "main": "./dist/index.js",
   "types": "./dist/index.d.ts",