Merge pull request #584 from luojiyin1987/fix/ai-suggestion-cache-key

fix(ai): include semantic request inputs in suggestion cache key

server/CMakeLists.txt

@@ -147,6 +147,8 @@ endif()
 set(SRC_FILES
     "./src/ai/ai_assistant.cpp"
     "./src/ai/ai_assistant.h"
+    "./src/ai/ai_assistant_cache_key.cpp"
+    "./src/ai/ai_assistant_cache_key.h"
     "./src/clipboard/clipboard_history.cpp"
     "./src/clipboard/clipboard_history.h"
     "./src/cloud/cloud_ime.cpp"

server/src/ai/ai_assistant.cpp

@@ -1,4 +1,5 @@
 #include "ai_assistant.h"
+#include "ai_assistant_cache_key.h"
 
 #include <curl/curl.h>
 #include <nlohmann/json.hpp>
@@ -24,13 +25,6 @@ std::chrono::steady_clock::time_point g_last_input;
 AiAssistant::ApplyCallback g_callback;
 std::unordered_map<std::string, std::string> g_candidate_cache;
 
-std::string BuildCacheKey(const AiAssistant::Request &request)
-{
-    nlohmann::json segments = request.pinyin_segments;
-    return request.config.provider + "\n" + request.config.endpoint + "\n" + request.config.model + "\n" +
-           segments.dump();
-}
-
 size_t WriteResponse(char *data, size_t size, size_t count, void *user)
 {
     static_cast<std::string *>(user)->append(data, size * count);
@@ -135,7 +129,7 @@ void WorkerLoop()
             break;
         observed = g_generation.load();
         auto request = g_latest;
-        const auto cached = g_candidate_cache.find(BuildCacheKey(request));
+        const auto cached = g_candidate_cache.find(AiAssistant::detail::BuildSuggestionCacheKey(request));
         if (!request.identity.empty() && cached != g_candidate_cache.end())
         {
             const std::string candidate = cached->second;
@@ -164,7 +158,7 @@ void WorkerLoop()
         {
             {
                 std::lock_guard cache_lock(g_mutex);
-                g_candidate_cache[BuildCacheKey(request)] = candidate;
+                g_candidate_cache[AiAssistant::detail::BuildSuggestionCacheKey(request)] = candidate;
             }
             (void)0;
             g_callback(candidate, request.identity, observed);

server/src/ai/ai_assistant_cache_key.cpp

@@ -0,0 +1,19 @@
+#include "ai_assistant_cache_key.h"
+
+#include <nlohmann/json.hpp>
+
+namespace AiAssistant::detail
+{
+std::string BuildSuggestionCacheKey(const Request &request)
+{
+    return nlohmann::json{{"provider", request.config.provider},
+                          {"endpoint", request.config.endpoint},
+                          {"model", request.config.model},
+                          {"prompt", request.config.prompt},
+                          {"pinyin_segments", request.pinyin_segments},
+                          {"context", request.context},
+                          {"candidate_limit", request.config.candidate_limit}}
+        // context 与 prompt 是自由文本；键在无 try 的工作线程里构建，非法 UTF-8 不能抛出
+        .dump(-1, ' ', false, nlohmann::json::error_handler_t::replace);
+}
+} // namespace AiAssistant::detail

server/src/ai/ai_assistant_cache_key.h

@@ -0,0 +1,10 @@
+#pragma once
+
+#include "ai_assistant.h"
+
+#include <string>
+
+namespace AiAssistant::detail
+{
+std::string BuildSuggestionCacheKey(const Request &request);
+} // namespace AiAssistant::detail

server/tests/CMakeLists.txt

@@ -36,6 +36,8 @@ set(TEST_SOURCE_FILES
     "${CMAKE_CURRENT_SOURCE_DIR}/src/test_input_key_policy.cpp"
     "${CMAKE_CURRENT_SOURCE_DIR}/src/test_candidate_ui_owner.cpp"
     "${CMAKE_CURRENT_SOURCE_DIR}/src/test_async_request_origin.cpp"
+    "${CMAKE_CURRENT_SOURCE_DIR}/src/test_ai_assistant_cache_key.cpp"
+    "${CMAKE_SOURCE_DIR}/src/ai/ai_assistant_cache_key.cpp"
     "${CMAKE_CURRENT_SOURCE_DIR}/src/test_candidate_ui_action_policy.cpp"
     "${CMAKE_CURRENT_SOURCE_DIR}/src/test_candidate_translation_policy.cpp"
     "${CMAKE_CURRENT_SOURCE_DIR}/src/test_candidate_ui_state.cpp"

server/tests/src/test_ai_assistant_cache_key.cpp

@@ -0,0 +1,66 @@
+#include "ai/ai_assistant_cache_key.h"
+#include "tests/includes/test_framework.h"
+
+namespace
+{
+AiAssistant::Request MakeRequest()
+{
+    AiAssistant::Request request;
+    request.pinyin_segments = {"shang", "hai"};
+    request.context = "我准备去";
+    request.config.provider = "deepseek";
+    request.config.endpoint = "https://api.example.com/chat/completions";
+    request.config.model = "example-model";
+    request.config.prompt = "Return the best candidate.";
+    request.config.candidate_limit = 3;
+    return request;
+}
+} // namespace
+
+TEST_CASE(ai_suggestion_cache_hits_for_the_same_semantic_request)
+{
+    const auto request = MakeRequest();
+    const auto same_request = request;
+
+    REQUIRE(AiAssistant::detail::BuildSuggestionCacheKey(request) ==
+            AiAssistant::detail::BuildSuggestionCacheKey(same_request));
+}
+
+TEST_CASE(ai_suggestion_cache_misses_when_context_changes)
+{
+    const auto request = MakeRequest();
+    auto changed_request = request;
+    changed_request.context = "他受到了严重";
+
+    REQUIRE(AiAssistant::detail::BuildSuggestionCacheKey(request) !=
+            AiAssistant::detail::BuildSuggestionCacheKey(changed_request));
+}
+
+TEST_CASE(ai_suggestion_cache_misses_when_prompt_changes)
+{
+    const auto request = MakeRequest();
+    auto changed_request = request;
+    changed_request.config.prompt = "Prefer place names.";
+
+    REQUIRE(AiAssistant::detail::BuildSuggestionCacheKey(request) !=
+            AiAssistant::detail::BuildSuggestionCacheKey(changed_request));
+}
+
+TEST_CASE(ai_suggestion_cache_key_tolerates_invalid_utf8)
+{
+    auto request = MakeRequest();
+    request.context = std::string("\xE6\x88", 2);
+    request.config.prompt = std::string("\xFF", 1);
+
+    REQUIRE(!AiAssistant::detail::BuildSuggestionCacheKey(request).empty());
+}
+
+TEST_CASE(ai_suggestion_cache_misses_when_candidate_limit_changes)
+{
+    const auto request = MakeRequest();
+    auto changed_request = request;
+    changed_request.config.candidate_limit = 1;
+
+    REQUIRE(AiAssistant::detail::BuildSuggestionCacheKey(request) !=
+            AiAssistant::detail::BuildSuggestionCacheKey(changed_request));
+}