Refactor

tests/translator/__snapshots__/golden-request.test.js.snap

@@ -89,7 +89,7 @@ exports[`GOLDEN request: OpenAI → Claude > full body (system/image/tool/tool_r
 }
 `;
 
-exports[`GOLDEN request: OpenAI → Claude > reasoning_effort → thinking budget 1`] = `
+exports[`GOLDEN request: OpenAI → Claude > reasoning_effort → adaptive output_config (claude 4.6+) 1`] = `
 {
   "max_tokens": 64000,
   "messages": [
@@ -104,6 +104,9 @@ exports[`GOLDEN request: OpenAI → Claude > reasoning_effort → thinking budge
     },
   ],
   "model": "claude-opus-4-6",
+  "output_config": {
+    "effort": "high",
+  },
   "stream": true,
   "system": [
     {
@@ -115,10 +118,6 @@ exports[`GOLDEN request: OpenAI → Claude > reasoning_effort → thinking budge
       "type": "text",
     },
   ],
-  "thinking": {
-    "budget_tokens": 16384,
-    "type": "enabled",
-  },
 }
 `;
 

tests/translator/golden-request.test.js

@@ -42,7 +42,7 @@ describe("GOLDEN request: OpenAI → Claude", () => {
     expect(clean(out)).toMatchSnapshot();
   });
 
-  it("reasoning_effort → thinking budget", () => {
+  it("reasoning_effort → adaptive output_config (claude 4.6+)", () => {
     const body = { messages: [{ role: "user", content: "hi" }], reasoning_effort: "high" };
     const out = translateRequest(FORMATS.OPENAI, FORMATS.CLAUDE, "claude-opus-4-6", body, true, { apiKey: "sk-x" }, "anthropic");
     expect(clean(out)).toMatchSnapshot();

tests/translator/real/thinking.real.test.js

@@ -0,0 +1,122 @@
+// REAL integration test for thinking normalization: sends a reasoning prompt with
+// reasoning_effort to every thinking-capable provider that has an active credential,
+// then asserts the upstream accepted it (no 400) and emitted reasoning output.
+// Gated by RUN_REAL=1 so the default `vitest run` never touches the network.
+//
+//   RUN_REAL=1 npx vitest run -c tests/vitest.config.js "tests/translator/real/thinking"
+//   RUN_REAL=1 REAL_PROVIDERS=claude,glm,deepseek npx vitest run -c tests/vitest.config.js "tests/translator/real/thinking"
+//
+import { describe, it, expect } from "vitest";
+import { getProviderCredentials } from "../../../src/sse/services/auth.js";
+import { checkAndRefreshToken } from "../../../src/sse/services/tokenRefresh.js";
+import { handleChatCore } from "../../../open-sse/handlers/chatCore.js";
+import { getModelsByProviderId } from "../../../open-sse/config/providerModels.js";
+import { getCapabilitiesForModel } from "../../../open-sse/providers/capabilities.js";
+
+const RUN_REAL = process.env.RUN_REAL === "1";
+const MAX_TOKENS = 512;
+const TIMEOUT_MS = 120000;
+const EFFORT = process.env.THINK_EFFORT || "high";
+const PROVIDER_FILTER = (process.env.REAL_PROVIDERS || "")
+  .split(",").map((s) => s.trim()).filter(Boolean);
+
+// First plain llm model that the capability registry marks as reasoning-capable.
+function firstReasoningModel(providerId) {
+  const models = getModelsByProviderId(providerId);
+  for (const m of models) {
+    if ((m.type || "llm") !== "llm") continue;
+    if (/embedding|image|tts|whisper|rerank|vision-model/i.test(m.id)) continue;
+    if (getCapabilitiesForModel(providerId, m.id).reasoning) return m.id;
+  }
+  return null;
+}
+
+async function drainSSE(response) {
+  if (!response?.body) return "";
+  const reader = response.body.getReader();
+  const decoder = new TextDecoder();
+  let out = "";
+  while (true) {
+    const { done, value } = await reader.read();
+    if (done) break;
+    out += decoder.decode(value, { stream: true });
+  }
+  return out;
+}
+
+describe.skipIf(!RUN_REAL).concurrent("REAL thinking normalization", () => {
+  it("has active providers in DB", () => {
+    expect(targetProviders().length).toBeGreaterThan(0);
+  });
+
+  for (const providerId of (RUN_REAL ? targetProviders() : [])) {
+    it.concurrent(
+      `${providerId}: accepts reasoning_effort=${EFFORT} and reasons`,
+      async () => {
+        const model = firstReasoningModel(providerId);
+        if (!model) {
+          console.warn(`[skip] ${providerId}: no reasoning-capable model`);
+          return expect(true).toBe(true);
+        }
+
+        const credentials = await getProviderCredentials(providerId, new Set(), model);
+        if (!credentials || credentials.allRateLimited) {
+          console.warn(`[skip] ${providerId}: no usable credential`);
+          return expect(true).toBe(true);
+        }
+
+        const refreshed = await checkAndRefreshToken(providerId, credentials);
+        const result = await handleChatCore({
+          body: {
+            model: `${providerId}/${model}`,
+            stream: true,
+            max_tokens: MAX_TOKENS,
+            reasoning_effort: EFFORT,
+            messages: [{ role: "user", content: "Think step by step, then answer: what is 17 * 23?" }],
+          },
+          modelInfo: { provider: providerId, model },
+          credentials: refreshed,
+          connectionId: credentials.connectionId,
+        });
+
+        if (!result.success) {
+          // 400 = our thinking payload was rejected → real bug. Other codes = credential/quota.
+          const credIssue = [401, 402, 403, 429].includes(Number(result.status));
+          if (credIssue) {
+            console.warn(`[skip] ${providerId}: ${result.status} (credential/quota)`);
+            return expect(true).toBe(true);
+          }
+          console.error(`[REJECT] ${providerId}/${model} ${result.status}:`, JSON.stringify(result.error)?.slice(0, 600));
+          throw new Error(`${providerId}/${model} thinking REJECTED: ${result.status}`);
+        }
+
+        const raw = await drainSSE(result.response);
+        expect(raw.length, `${providerId}: empty response`).toBeGreaterThan(0);
+        const hasReasoning = /reasoning_content|"thinking"|reasoning_details|<think/.test(raw);
+        // Log so the operator can eyeball which providers actually streamed reasoning.
+        console.log(`[ok] ${providerId}/${model} reasoning=${hasReasoning}`);
+        expect(/data:|"delta"|"content"|finish_reason/.test(raw), `${providerId}: not SSE`).toBe(true);
+      },
+      TIMEOUT_MS
+    );
+  }
+});
+
+function targetProviders() {
+  try {
+    const Database = require("better-sqlite3");
+    const os = require("os");
+    const path = require("path");
+    const dbPath = process.env.DATA_DIR
+      ? path.join(process.env.DATA_DIR, "db", "data.sqlite")
+      : path.join(os.homedir(), ".9router", "db", "data.sqlite");
+    const db = new Database(dbPath, { readonly: true });
+    const rows = db.prepare("SELECT DISTINCT provider FROM providerConnections WHERE isActive = 1").all();
+    db.close();
+    let list = rows.map((r) => r.provider).sort();
+    if (PROVIDER_FILTER.length) list = list.filter((p) => PROVIDER_FILTER.includes(p));
+    return list;
+  } catch {
+    return [];
+  }
+}

tests/translator/real/vision-capability-survey.real.test.js

@@ -25,6 +25,11 @@ const PROVIDER_FILTER = (process.env.REAL_PROVIDERS || "")
 const PNG_DATA_URI = "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAABAAAAAQCAIAAACQkWg2AAAAFklEQVR4nGO4I2JDEmIY1TCqYfhqAAAeBCwQ8YdREQAAAABJRU5ErkJggg==";
 // Tiny silent WAV (44-byte header, no samples) — enough to probe audioInput acceptance.
 const WAV_B64 = "UklGRiQAAABXQVZFZm10IBAAAAABAAEARKwAAIhYAQACABAAZGF0YQAAAAA=";
+// Stable public image URL (probe remote-URL handling vs base64).
+const IMAGE_REMOTE_URL = "https://www.google.com/images/branding/googlelogo/1x/googlelogo_color_272x92dp.png";
+// Tiny valid PDF (probe file/document handling).
+const PDF_B64 = "JVBERi0xLjEKMSAwIG9iajw8L1R5cGUvQ2F0YWxvZy9QYWdlcyAyIDAgUj4+ZW5kb2JqCjIgMCBvYmo8PC9UeXBlL1BhZ2VzL0tpZHNbMyAwIFJdL0NvdW50IDE+PmVuZG9iagozIDAgb2JqPDwvVHlwZS9QYWdlL1BhcmVudCAyIDAgUi9NZWRpYUJveFswIDAgMjAwIDIwMF0+PmVuZG9iagp4cmVmCjAgNAowMDAwMDAwMDAwIDY1NTM1IGYgCjAwMDAwMDAwMDkgMDAwMDAgbiAKMDAwMDAwMDA1MiAwMDAwMCBuIAowMDAwMDAwMTAxIDAwMDAwIG4gCnRyYWlsZXI8PC9TaXplIDQvUm9vdCAxIDAgUj4+CnN0YXJ0eHJlZgoxNjYKJSVFT0Y=";
+const PDF_DATA_URI = `data:application/pdf;base64,${PDF_B64}`;
 
 // Capability probes: each sends one modality-specific content block. cap = capability flag tested.
 const CAPABILITY_PROBES = {
@@ -42,6 +47,20 @@ const CAPABILITY_PROBES = {
       { type: "input_audio", input_audio: { data: WAV_B64, format: "wav" } },
     ],
   },
+  imageUrl: {
+    cap: "vision",
+    content: [
+      { type: "text", text: "What is in this image? One word." },
+      { type: "image_url", image_url: { url: IMAGE_REMOTE_URL } },
+    ],
+  },
+  file: {
+    cap: "pdf",
+    content: [
+      { type: "text", text: "Summarize this document. One word." },
+      { type: "file", file: { filename: "doc.pdf", file_data: PDF_DATA_URI } },
+    ],
+  },
 };
 // Set PROBES=vision,audio (default vision only to limit live quota).
 const ACTIVE_PROBES = (process.env.PROBES || "vision").split(",").map((s) => s.trim()).filter(Boolean);

tests/translator/thinking-unified.test.js

@@ -0,0 +1,131 @@
+// Unit tests for unified thinking normalization (thinkingUnified.js).
+// Covers extract, suffix parse, and per-provider apply per MATRIX (.docs/thinking/plan.md).
+import { describe, it, expect } from "vitest";
+import {
+  parseSuffix,
+  extractThinking,
+  applyThinking,
+} from "../../open-sse/translator/concerns/thinkingUnified.js";
+import { extractReasoningText } from "../../open-sse/translator/concerns/reasoning.js";
+
+const apply = (targetFormat, model, body, provider) => {
+  const b = JSON.parse(JSON.stringify(body));
+  applyThinking(targetFormat, model, b, provider);
+  return b;
+};
+
+describe("parseSuffix", () => {
+  it("parses level suffix", () => {
+    expect(parseSuffix("gpt-5(high)")).toEqual({ cleanModel: "gpt-5", override: { mode: "level", level: "high" } });
+  });
+  it("parses numeric budget suffix", () => {
+    expect(parseSuffix("model(8192)")).toEqual({ cleanModel: "model", override: { mode: "budget", budget: 8192 } });
+  });
+  it("parses auto / none", () => {
+    expect(parseSuffix("m(auto)").override).toEqual({ mode: "auto" });
+    expect(parseSuffix("m(none)").override).toEqual({ mode: "none" });
+  });
+  it("no suffix → passthrough", () => {
+    expect(parseSuffix("claude-opus-4.7")).toEqual({ cleanModel: "claude-opus-4.7", override: null });
+  });
+});
+
+describe("extractThinking", () => {
+  it("claude enabled+budget", () => {
+    expect(extractThinking({ thinking: { type: "enabled", budget_tokens: 4096 } })).toEqual({ mode: "budget", budget: 4096 });
+  });
+  it("claude disabled", () => {
+    expect(extractThinking({ thinking: { type: "disabled" } })).toEqual({ mode: "none" });
+  });
+  it("openai reasoning_effort", () => {
+    expect(extractThinking({ reasoning_effort: "high" })).toEqual({ mode: "level", level: "high" });
+  });
+  it("responses reasoning.effort none", () => {
+    expect(extractThinking({ reasoning: { effort: "none" } })).toEqual({ mode: "none" });
+  });
+  it("gemini thinkingBudget 0 → none", () => {
+    expect(extractThinking({ thinkingConfig: { thinkingBudget: 0 } })).toEqual({ mode: "none" });
+  });
+  it("qwen enable_thinking false", () => {
+    expect(extractThinking({ enable_thinking: false })).toEqual({ mode: "none" });
+  });
+  it("no intent → null", () => {
+    expect(extractThinking({ messages: [] })).toBeNull();
+  });
+});
+
+describe("applyThinking per provider format", () => {
+  it("claude 4.6+ → adaptive output_config (no budget_tokens)", () => {
+    const out = apply("claude", "claude-opus-4.7", { reasoning_effort: "high" }, "claude");
+    expect(out.output_config).toEqual({ effort: "high" });
+    expect(out.thinking).toBeUndefined();
+  });
+  it("claude haiku → enabled+budget", () => {
+    const out = apply("claude", "claude-haiku-4.5", { reasoning_effort: "high" }, "claude");
+    expect(out.thinking).toEqual({ type: "enabled", budget_tokens: 24576 });
+  });
+  it("gemini-3 → thinkingLevel", () => {
+    const out = apply("gemini", "gemini-3-pro", { reasoning_effort: "medium" }, "gemini");
+    expect(out.generationConfig.thinkingConfig.thinkingLevel).toBe("medium");
+  });
+  it("gemini-2.5 → thinkingBudget", () => {
+    const out = apply("gemini", "gemini-2.5-flash", { reasoning_effort: "high" }, "gemini");
+    expect(out.generationConfig.thinkingConfig.thinkingBudget).toBe(24576);
+    expect(out.generationConfig.thinkingConfig.thinkingLevel).toBeUndefined();
+  });
+  it("GLM off → enable_thinking:false (not thinking.disabled)", () => {
+    const out = apply("openai", "glm-4.6", { reasoning_effort: "none" }, "glm");
+    expect(out.enable_thinking).toBe(false);
+    expect(out.thinking).toBeUndefined();
+  });
+  it("Qwen on → enable_thinking + thinking_budget", () => {
+    const out = apply("openai", "qwen3-max", { reasoning_effort: "medium" }, "qwen");
+    expect(out.enable_thinking).toBe(true);
+    expect(out.thinking_budget).toBe(8192);
+  });
+  it("QwQ cannot disable → clamp minimal", () => {
+    const out = apply("openai", "qwq-32b", { reasoning_effort: "none" }, "qwen");
+    expect(out.enable_thinking).toBe(true);
+  });
+  it("DeepSeek → enabled + reasoning_effort high (low→high)", () => {
+    const out = apply("openai", "deepseek-v4-pro", { reasoning_effort: "low" }, "deepseek");
+    expect(out.thinking).toEqual({ type: "enabled" });
+    expect(out.reasoning_effort).toBe("high");
+  });
+  it("Kimi on → reasoning_effort", () => {
+    const out = apply("openai", "kimi-k2.6", { reasoning_effort: "high" }, "kimi");
+    expect(out.reasoning_effort).toBe("high");
+  });
+  it("MiniMax M3 → adaptive", () => {
+    const out = apply("claude", "MiniMax-M3", { reasoning_effort: "high" }, "minimax");
+    expect(out.thinking).toEqual({ type: "adaptive" });
+  });
+  it("non-reasoning model → strips thinking", () => {
+    const out = apply("openai", "gpt-4o", { reasoning_effort: "high" }, "openai");
+    expect(out.reasoning_effort).toBeUndefined();
+  });
+  it("aggregator (siliconflow) GLM model → forced openai reasoning_effort", () => {
+    const out = apply("openai", "zai-org/GLM-5", { reasoning_effort: "high" }, "siliconflow");
+    expect(out.reasoning_effort).toBe("high");
+    expect(out.enable_thinking).toBeUndefined();
+  });
+  it("suffix overrides body", () => {
+    const out = apply("openai", "gpt-5(low)", { reasoning_effort: "high" }, "openai");
+    expect(out.reasoning_effort).toBe("low");
+  });
+});
+
+describe("extractReasoningText (response shapes)", () => {
+  it("reasoning_content (GLM/Qwen/DeepSeek)", () => {
+    expect(extractReasoningText({ reasoning_content: "abc" })).toBe("abc");
+  });
+  it("reasoning fallback", () => {
+    expect(extractReasoningText({ reasoning: "xyz" })).toBe("xyz");
+  });
+  it("reasoning_details[] (MiniMax split)", () => {
+    expect(extractReasoningText({ reasoning_details: [{ text: "a" }, { content: "b" }, "c"] })).toBe("abc");
+  });
+  it("no reasoning → empty", () => {
+    expect(extractReasoningText({ content: "hello" })).toBe("");
+  });
+});