refactor(llm): extract testable embedding protocol routing + drop dead fromProviderId

mateclaw-server/src/main/java/vip/mate/llm/embedding/EmbeddingModelFactory.java

@@ -105,6 +105,25 @@ public class EmbeddingModelFactory {
         cache.clear();
     }
 
+    /**
+     * Pick the embedding protocol from the provider's {@code chatModel} column.
+     * Package-private for unit testing.
+     * <p>
+     * dashscope-compat carries 'dashscope' in its providerId but is wired with
+     * OpenAIChatModel + compatible-mode base URL, so substring-matching the
+     * providerId routed it to the native DashScope embedding endpoint and 404'd.
+     * Using the {@code chatModel} column matches {@link ModelProtocol#fromChatModel}
+     * for the chat path — same signal, same case-insensitive trim semantics.
+     */
+    static EmbeddingProtocol resolveEmbeddingProtocol(String chatModel) {
+        if (chatModel == null || chatModel.isBlank()) {
+            return EmbeddingProtocol.OPENAI_EMBEDDING;
+        }
+        return "DashScopeChatModel".equalsIgnoreCase(chatModel.trim())
+                ? EmbeddingProtocol.DASHSCOPE_EMBEDDING
+                : EmbeddingProtocol.OPENAI_EMBEDDING;
+    }
+
     // ==================== 内部实现 ====================
 
     private EmbeddingModel doBuild(ModelConfigEntity modelConfig) {
@@ -114,14 +133,7 @@ public class EmbeddingModelFactory {
                     "Embedding provider '" + modelConfig.getProvider() + "' not found in mate_model_provider");
         }
 
-        // Use chatModel column (same signal as ModelProtocol.fromChatModel) rather than
-        // providerId substring matching. dashscope-compat has "dashscope" in its id but
-        // uses OpenAIChatModel + compatible-mode URL — routing it to DASHSCOPE_EMBEDDING
-        // causes DashScopeApi to construct a native path that returns 404 against the
-        // compat base URL.
-        EmbeddingProtocol protocol = "DashScopeChatModel".equals(provider.getChatModel())
-                ? EmbeddingProtocol.DASHSCOPE_EMBEDDING
-                : EmbeddingProtocol.OPENAI_EMBEDDING;
+        EmbeddingProtocol protocol = resolveEmbeddingProtocol(provider.getChatModel());
         log.info("[EmbeddingFactory] Building embedding model: provider={}, chatModel={}, model={}, protocol={}",
                 provider.getProviderId(), provider.getChatModel(), modelConfig.getModelName(), protocol);
 

mateclaw-server/src/main/java/vip/mate/llm/model/EmbeddingProtocol.java

@@ -6,13 +6,14 @@ package vip.mate.llm.model;
  * 与 {@link ModelProtocol}（Chat 协议）分离——chat 和 embedding 在同一个 Provider 下
  * 可能走不同的请求格式：
  * <ul>
- *   <li>DashScope 的 embedding endpoint 是专用 path（/api/v1/services/embeddings/text-embedding/text-embedding）</li>
+ *   <li>DashScope 的 embedding endpoint 是专用 path
+ *       （/api/v1/services/embeddings/text-embedding/text-embedding）</li>
  *   <li>OpenAI 兼容协议的 embedding 统一走 /v1/embeddings</li>
  * </ul>
  * <p>
- * <b>注意</b>：{@link EmbeddingModelFactory} 现在优先通过 {@code chatModel} 列判断协议
- * （与 {@link ModelProtocol#fromChatModel} 保持一致），{@link #fromProviderId} 已不再使用。
- * 保留该方法仅作参考；不要在新代码中调用它。
+ * Routing happens in {@code EmbeddingModelFactory.resolveEmbeddingProtocol}
+ * via the {@code chatModel} column on {@link ModelProviderEntity}, matching
+ * {@link ModelProtocol#fromChatModel}.
  *
  * @author MateClaw Team
  */
@@ -30,20 +31,4 @@ public enum EmbeddingProtocol {
     public String getId() {
         return id;
     }
-
-    /**
-     * 从 providerId 推断 embedding 协议。
-     * <ul>
-     *   <li>dashscope / 任何包含 "dashscope" / "qwen" / "aliyun" 的 → DASHSCOPE_EMBEDDING</li>
-     *   <li>其他（openai / deepseek / kimi / zhipu / moonshot / 任何 OpenAI 兼容） → OPENAI_EMBEDDING</li>
-     * </ul>
-     */
-    public static EmbeddingProtocol fromProviderId(String providerId) {
-        if (providerId == null) return OPENAI_EMBEDDING;
-        String p = providerId.toLowerCase().trim();
-        if (p.contains("dashscope") || p.contains("qwen") || p.contains("aliyun")) {
-            return DASHSCOPE_EMBEDDING;
-        }
-        return OPENAI_EMBEDDING;
-    }
 }

mateclaw-server/src/test/java/vip/mate/llm/embedding/EmbeddingModelFactoryRoutingTest.java

@@ -0,0 +1,62 @@
+package vip.mate.llm.embedding;
+
+import org.junit.jupiter.api.Test;
+import vip.mate.llm.model.EmbeddingProtocol;
+
+import static org.junit.jupiter.api.Assertions.assertEquals;
+
+/**
+ * Routing tests for {@link EmbeddingModelFactory#resolveEmbeddingProtocol(String)}.
+ * The dashscope-compat regression (#166) lives in the OpenAIChatModel branch.
+ */
+class EmbeddingModelFactoryRoutingTest {
+
+    @Test
+    void dashscopeChatModel_routesToNativeEmbedding() {
+        assertEquals(EmbeddingProtocol.DASHSCOPE_EMBEDDING,
+                EmbeddingModelFactory.resolveEmbeddingProtocol("DashScopeChatModel"));
+    }
+
+    @Test
+    void openaiChatModel_routesToOpenAiCompat() {
+        // dashscope-compat carries chatModel='OpenAIChatModel' — must NOT take the native path
+        assertEquals(EmbeddingProtocol.OPENAI_EMBEDDING,
+                EmbeddingModelFactory.resolveEmbeddingProtocol("OpenAIChatModel"));
+    }
+
+    @Test
+    void anthropicChatModel_routesToOpenAiCompat() {
+        assertEquals(EmbeddingProtocol.OPENAI_EMBEDDING,
+                EmbeddingModelFactory.resolveEmbeddingProtocol("AnthropicChatModel"));
+    }
+
+    @Test
+    void nullChatModel_routesToOpenAiCompat() {
+        assertEquals(EmbeddingProtocol.OPENAI_EMBEDDING,
+                EmbeddingModelFactory.resolveEmbeddingProtocol(null));
+    }
+
+    @Test
+    void blankChatModel_routesToOpenAiCompat() {
+        assertEquals(EmbeddingProtocol.OPENAI_EMBEDDING,
+                EmbeddingModelFactory.resolveEmbeddingProtocol(""));
+        assertEquals(EmbeddingProtocol.OPENAI_EMBEDDING,
+                EmbeddingModelFactory.resolveEmbeddingProtocol("   "));
+    }
+
+    @Test
+    void chatModelMatchIsCaseInsensitiveAndTrimmed() {
+        // Mirror ModelProtocol.fromChatModel's normalization: equalsIgnoreCase + trim.
+        assertEquals(EmbeddingProtocol.DASHSCOPE_EMBEDDING,
+                EmbeddingModelFactory.resolveEmbeddingProtocol("dashscopechatmodel"));
+        assertEquals(EmbeddingProtocol.DASHSCOPE_EMBEDDING,
+                EmbeddingModelFactory.resolveEmbeddingProtocol("  DashScopeChatModel  "));
+    }
+
+    @Test
+    void unknownChatModel_fallsBackToOpenAiCompat() {
+        // Future / custom chatModel strings default to OpenAI-compatible (safe default).
+        assertEquals(EmbeddingProtocol.OPENAI_EMBEDDING,
+                EmbeddingModelFactory.resolveEmbeddingProtocol("CustomGatewayChatModel"));
+    }
+}