From 6aac7efcfb684f9f8abf44077ed5978bfb16f979 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E7=BB=9D=E5=B0=98?= <237809796@qq.com>
Date: Sun, 28 Jun 2026 18:47:21 +0800
Subject: [PATCH] =?UTF-8?q?=E4=BF=AE=E5=A4=8D:=20=E7=9F=A5=E8=AF=86?=
 =?UTF-8?q?=E6=8F=90=E7=82=BC=E5=88=87=E8=AF=BB=20ai=5Fmessages=20?=
 =?UTF-8?q?=E8=A1=A8,=E5=AE=8C=E6=88=90=E6=B6=88=E6=81=AF=E6=BA=AF?=
 =?UTF-8?q?=E6=BA=90=20P2=20=E6=9C=80=E5=90=8E=E4=B8=80=E7=8E=AF?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

消息级溯源 P2 切读早前已覆盖主要读写路径(save_conversation 写 /
switch/export/load_messages 读),但 knowledge_inject 的知识提炼
仍在读旧 ai_conversations.messages JSON 列,与 P2 切读目标不一致。

本次补全:
- extract_knowledge_from_conversation 改为优先读 ai_messages 表
- 表空时 fallback 旧 messages JSON 列(老库兼容,对齐 switch 路径)
- 复用 commands::record_to_message 映射函数(DRY,与其他读路径同源)

至此消息溯源 P2 切读全部完成,所有读写路径统一走 ai_messages 表,
旧 messages JSON 列仅作老库兼容 fallback,不再写入。
---
 docs/待决策.md                                |  6 +++++-
 src-tauri/src/commands/ai/knowledge_inject.rs | 21 +++++++++++++------
 2 files changed, 20 insertions(+), 7 deletions(-)

diff --git a/docs/待决策.md b/docs/待决策.md
index 63751f4..8c41310 100644
--- a/docs/待决策.md
+++ b/docs/待决策.md
@@ -298,7 +298,11 @@
 #### DEC-260628-04 消息级溯源 P2 切读
 - **决策**:✅ b 一次性切读(保留原表数据不删)
 - **记录**:读写路径从 `ai_conversations.messages` 切到 `ai_messages` 表,原列保留
-- **状态**:🟢 已决·待推进
+- **核验(2026-06-28)**:✅ 全部已完成——
+  - 写路径:`save_conversation` 已全量重写 ai_messages
+  - 读路径:`ai_conversation_switch`/`ai_conversation_export`/`remote_bridge::route_load_messages` 优先读 ai_messages,表空时 fallback 旧 JSON(老库兼容)
+  - 最后一处补全:`knowledge_inject::extract_knowledge_from_conversation` 也切到 ai_messages
+- **状态**:✅ 已决·已实施
 
 #### DEC-260628-05 SMELL-P1-6 String→newtype
 - **决策**:✅ b 路线(仅 Rust 内部 newtype)
diff --git a/src-tauri/src/commands/ai/knowledge_inject.rs b/src-tauri/src/commands/ai/knowledge_inject.rs
index 450edd5..7206a94 100644
--- a/src-tauri/src/commands/ai/knowledge_inject.rs
+++ b/src-tauri/src/commands/ai/knowledge_inject.rs
@@ -12,7 +12,7 @@ use df_ai::provider::{ChatMessage, CompletionRequest, LlmProvider, MessageRole};
 use df_ai::router::{
     select_model_id, Modality, TaskRequirements,
 };
-use df_storage::crud::{AiConversationRepo, KnowledgeRepo};
+use df_storage::crud::{AiConversationRepo, AiMessageRepo, KnowledgeRepo};
 use df_storage::db::Database;
 use df_storage::models::{AiProviderRecord, KnowledgeRecord};
 
@@ -566,12 +566,21 @@ async fn extract_knowledge_from_conversation(
         .filter(|t| !t.trim().is_empty())
         .unwrap_or_else(|| "未命名对话".to_string());
 
-    let messages: Vec<ChatMessage> = match serde_json::from_str(&conv.messages) {
-        Ok(v) => v,
-        Err(e) => {
-            tracing::warn!(error = %e, "[ai] 对话消息 JSON 解析失败,降级空消息(知识提炼跳过)");
-            Vec::new()
+    // 消息读取:优先 ai_messages 表(消息拆分存储真相源),表空时 fallback 旧 messages JSON 列(老库兼容)
+    let msg_repo = AiMessageRepo::new(db);
+    let records = msg_repo.list_by_conversation(conv_id).await.unwrap_or_default();
+    let messages: Vec<ChatMessage> = if !records.is_empty() {
+        records.iter().map(crate::commands::ai::commands::record_to_message).collect()
+    } else {
+        // 老库未迁移或坏数据:回退旧 messages JSON 列(与 ai_conversation_switch 同一兼容路径)
+        let has_legacy = conv.messages != "[]" && !conv.messages.is_empty();
+        if has_legacy {
+            tracing::warn!(
+                conv_id,
+                "[KNOWLEDGE-EXTRACT] ai_messages 表为空,回退旧 messages JSON 列(老库兼容)"
+            );
         }
+        serde_json::from_str(&conv.messages).unwrap_or_default()
     };
     // 过滤 user/assistant,取最后 6 条
     let recent: Vec<&ChatMessage> = messages