Files
agcore/design/pdd/26-step1-phase20-cargo-features-implementation.md
T
徐涛 28ca43ccb2 chore(docs): 将设计文档从 docs 移至 design 目录
将 note、pdd、prd、roadmap 四类文档分别归入 `design/` 下对应子目录中,并新增 `.gitkeep` 占位文件
2026-07-23 05:45:53 +08:00

695 lines
27 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# AG Core v0.3.2 Step 1Phase 20)— Cargo Features 基础设施改造实施方案
## 1. 背景与目标
**背景**agcore 是一个 Rust 编写的智能体核心工具箱,目前约 23,718 行、66 个源文件。v0.3.0 发布后,所有模块在编译时全量捆绑,下游用户无法按需选择模块,即使只使用 LLM 对话也需要编译 sqlite / MCP / agent 引擎等全部依赖。
**目标**:通过 Cargo features 拆分让下游按需选择模块。Step 1 是基础设施变更 —— `Cargo.toml` features 定义 + 依赖 optional 化 + 必要的子模块 cfg 门控,编译通过后打 checkpoint。
**预期效果**
- `default = ["full"]` → v0.3.0 用户零迁移成本
- 最小组合(`document`)零重型外部依赖(仅依赖始终编译的轻量依赖:serde/serde_json/thiserror/async-trait/tracing
- 纯对话组合(`chat + provider-openai`)仅需 ~10 个依赖,不含 sqlite / MCP / engine
## 2. 需求分析
### 2.1 约束条件
| # | 约束 | 说明 |
|---|------|------|
| 1 | `default = ["full"]` | 保持向后兼容,v0.3.0 用户零迁移成本 |
| 2 | `document` feature 零重型外部依赖(仅依赖始终编译的轻量依赖:serde/serde_json/thiserror/async-trait/tracing | 纯 std + 始终编译的轻量依赖(serde/serde_json/thiserror/async-trait/tracing |
| 3 | tokio 从 `["full"]` 拆细 | 已验证全库无 net/fs/signal 使用,拆为 `["rt", "sync", "time", "macros", "process", "io-util"]` |
| 4 | 重型依赖全部 optional | tokio、reqwest、rusqlite、tracing-subscriber、tokio-stream、futures、futures-util、futures-core、bytes、async-stream、tokio-util、time |
| 5 | 始终编译的轻量依赖 | serde、serde_json、thiserror、async-trait、tracing |
### 2.2 关键决策
| # | 决策 | 理由 |
|---|------|------|
| 1 | `tools` feature 必须 `imply tokio` | `src/tools/registry.rs` 使用 `tokio::time::timeout` |
| 2 | `pub mod llm` 门控条件为 `any(feature = "llm-types", feature = "llm")` | `prompt → llm-types` 路径需要 llm 模块编译,但只需 types 子模块 |
| 3 | 测试 dev-dependencies 加 `tokio = { version = "1", features = ["rt", "macros"] }` | 现有 `#[tokio::test]` 需要 tokio runtime |
| 4 | 快捷组合名保持原名(chat/multi/light | 文档中说明各组合包含的 feature 约束 |
| 5 | `init_tracing()` 函数整体用 `#[cfg(feature = "tracing-init")]` 包裹 | 避免 `use tracing_subscriber` 出现在未启用 feature 时编译失败 |
## 3. 方案设计
### 3.1 Features 定义(完整 Cargo.toml `[features]` 草案)
```toml
[features]
default = ["full"]
# === 模块级 features ===
document = []
llm-types = []
prompt = ["llm-types"]
llm = ["llm-types", "tokio", "async-stream", "futures-core", "tokio-stream"]
tools = ["llm-types", "futures", "tokio-util", "tokio"]
tools-mcp = ["tools", "reqwest"]
# memory 模块依赖 llmconversation/vector_store 使用 compact/embedding)、tokioknowledge.rs 使用 Mutex)、timetypes.rs 使用 OffsetDateTime
memory = ["document", "llm", "tokio", "time"]
memory-sqlite = ["memory", "rusqlite", "time"]
agent = ["llm", "tools", "memory", "futures-util"]
engine = ["agent"]
# === Provider features ===
# Provider features — openai/anthropic 额外依赖 bytes(流式解析)和 futures-utilStream 组合)
provider-openai = ["llm", "reqwest", "bytes", "futures-util"]
provider-anthropic = ["llm", "reqwest", "bytes", "futures-util"]
# deepseek/qwen 使用 openai_compat 适配层,不需要 bytes 和 futures-util
provider-deepseek = ["llm", "reqwest"]
provider-qwen = ["llm", "reqwest"]
provider-ollama = ["llm", "reqwest"]
# === 工具 features ===
tracing-init = ["tracing-subscriber"]
# === 快捷组合 ===
full = [
"document", "llm-types", "prompt", "llm",
"tools", "tools-mcp",
"memory", "memory-sqlite",
"agent", "engine",
"provider-openai", "provider-anthropic", "provider-deepseek",
"provider-qwen", "provider-ollama",
"tracing-init",
]
light = ["llm", "provider-openai", "tools", "tools-mcp", "memory", "agent", "engine", "prompt", "document"]
chat = ["agent", "provider-openai"]
multi = ["engine", "provider-openai"]
```
**features 依赖图(简略)**
```
document (零外部依赖)
└── memory (+llm, +tokio, +time) ─── memory-sqlite (+rusqlite, +time)
llm-types (零依赖)
├── prompt
└── llm (+tokio, +async-stream, +futures-core, +tokio-stream)
├── tools (+futures, +tokio-util) ─── tools-mcp (+reqwest)
├── provider-openai / provider-anthropic (+reqwest, +bytes, +futures-util)
├── provider-deepseek / provider-qwen / provider-ollama (+reqwest)
└── agent (+tools, +memory, +futures-util) ─── engine
```
### 3.2 依赖 optional 化方案
**始终编译(5 个,不参与门控)**
```toml
serde = { version = "1", features = ["derive"] }
serde_json = "1"
thiserror = "2"
async-trait = "0.1"
tracing = "0.1"
```
**12 个依赖加 `optional = true`**
| 依赖 | 原声明 | 新声明 |
|------|--------|--------|
| tokio | `{ version = "1", features = ["full"] }` | `{ version = "1", features = ["rt", "sync", "time", "macros", "process", "io-util"], optional = true }` |
| reqwest | `{ version = "0.12", features = ["json", "stream"] }` | `{ version = "0.12", features = ["json", "stream"], optional = true }` |
| rusqlite | `{ version = "0.32", features = ["bundled"] }` | `{ version = "0.32", features = ["bundled"], optional = true }` |
| tracing-subscriber | `{ version = "0.3", features = ["env-filter"] }` | `{ version = "0.3", features = ["env-filter"], optional = true }` |
| tokio-stream | `{ version = "0.1" }` | `{ version = "0.1", optional = true }` |
| futures | `{ version = "0.3" }` | `{ version = "0.3", optional = true }` |
| futures-util | `{ version = "0.3" }` | `{ version = "0.3", optional = true }` |
| futures-core | `{ version = "0.3" }` | `{ version = "0.3", optional = true }` |
| bytes | `{ version = "1" }` | `{ version = "1", optional = true }` |
| async-stream | `{ version = "0.3" }` | `{ version = "0.3", optional = true }` |
| tokio-util | `{ version = "0.7", features = ["rt", "sync"] }` | `{ version = "0.7", features = ["rt", "sync"], optional = true }` |
| time | `{ version = "0.3", features = ["serde", "parsing", "formatting", "macros"] }` | `{ version = "0.3", features = ["serde", "parsing", "formatting", "macros"], optional = true }` |
**dev-dependencies 新增**
```toml
[dev-dependencies]
tokio = { version = "1", features = ["rt", "macros"] }
```
### 3.3 源文件改动清单
共涉及 **8 个文件**(预估 ~100 行改动):`Cargo.toml``src/lib.rs``src/llm.rs``src/llm/cycle.rs``src/tools.rs``src/memory.rs``src/memory/store.rs``src/agent/session.rs`
---
#### 文件 1`Cargo.toml`
**改动 1.1** — 新增 `[features]` 表(约 45 行,插入在 `[package]` 之后、`[dependencies]` 之前)
```diff
+ [features]
+ default = ["full"]
+
+ # === 模块级 features ===
+ document = []
+ llm-types = []
+ prompt = ["llm-types"]
+ llm = ["llm-types", "tokio", "async-stream", "futures-core", "tokio-stream"]
+ tools = ["llm-types", "futures", "tokio-util", "tokio"]
+ tools-mcp = ["tools", "reqwest"]
+ # memory 模块依赖 llmconversation/vector_store 使用 compact/embedding)、tokioknowledge.rs 使用 Mutex)、timetypes.rs 使用 OffsetDateTime
+ memory = ["document", "llm", "tokio", "time"]
+ memory-sqlite = ["memory", "rusqlite", "time"]
+ agent = ["llm", "tools", "memory", "futures-util"]
+ engine = ["agent"]
+
+ # === Provider features ===
+ # Provider features — openai/anthropic 额外依赖 bytes(流式解析)和 futures-utilStream 组合)
+ provider-openai = ["llm", "reqwest", "bytes", "futures-util"]
+ provider-anthropic = ["llm", "reqwest", "bytes", "futures-util"]
+ # deepseek/qwen 使用 openai_compat 适配层,不需要 bytes 和 futures-util
+ provider-deepseek = ["llm", "reqwest"]
+ provider-qwen = ["llm", "reqwest"]
+ provider-ollama = ["llm", "reqwest"]
+
+ # === 工具 features ===
+ tracing-init = ["tracing-subscriber"]
+
+ # === 快捷组合 ===
+ full = [
+ "document", "llm-types", "prompt", "llm",
+ "tools", "tools-mcp",
+ "memory", "memory-sqlite",
+ "agent", "engine",
+ "provider-openai", "provider-anthropic", "provider-deepseek",
+ "provider-qwen", "provider-ollama",
+ "tracing-init",
+ ]
+ light = ["llm", "provider-openai", "tools", "tools-mcp", "memory", "agent", "engine", "prompt", "document"]
+ chat = ["agent", "provider-openai"]
+ multi = ["engine", "provider-openai"]
```
**改动 1.2** — tokio 依赖声明修改
```diff
- tokio = { version = "1", features = ["full"] }
+ tokio = { version = "1", features = ["rt", "sync", "time", "macros", "process", "io-util"], optional = true }
```
**改动 1.3** — 11 个重型依赖逐行加 `optional = true`
```diff
- reqwest = { version = "0.12", features = ["json", "stream"] }
+ reqwest = { version = "0.12", features = ["json", "stream"], optional = true }
- rusqlite = { version = "0.32", features = ["bundled"] }
+ rusqlite = { version = "0.32", features = ["bundled"], optional = true }
- tracing-subscriber = { version = "0.3", features = ["env-filter"] }
+ tracing-subscriber = { version = "0.3", features = ["env-filter"], optional = true }
- tokio-stream = "0.1"
+ tokio-stream = { version = "0.1", optional = true }
- futures = "0.3"
+ futures = { version = "0.3", optional = true }
- futures-util = "0.3"
+ futures-util = { version = "0.3", optional = true }
- futures-core = "0.3"
+ futures-core = { version = "0.3", optional = true }
- bytes = "1"
+ bytes = { version = "1", optional = true }
- async-stream = "0.3"
+ async-stream = { version = "0.3", optional = true }
- tokio-util = { version = "0.7", features = ["rt"] }
+ tokio-util = { version = "0.7", features = ["rt", "sync"], optional = true }
- time = { version = "0.3", features = ["serde", "parsing", "formatting", "macros"] }
+ time = { version = "0.3", features = ["serde", "parsing", "formatting", "macros"], optional = true }
```
**改动 1.4**`[dev-dependencies]` 新增 tokio
```diff
+ [dev-dependencies]
+ tokio = { version = "1", features = ["rt", "macros"] }
```
**说明**:如果原 `Cargo.toml` 已有 `[dev-dependencies]` 则追加该行;若无则新增整个 section。
---
#### 文件 2`src/lib.rs`(当前约 26 行 → 改动后约 40 行)
**当前内容(参考)**
```rust
//! agcore —— 智能体(Agent)核心工具箱。
pub mod llm;
pub mod document;
pub mod prompt;
pub mod tools;
pub mod memory;
pub mod agent;
pub mod engine;
pub use document::Document;
use tracing_subscriber::{EnvFilter, fmt, prelude::*};
static INIT: std::sync::Once = std::sync::Once::new();
pub fn init_tracing() {
INIT.call_once(|| {
let filter = EnvFilter::try_from_default_env()
.unwrap_or_else(|_| EnvFilter::new("agcore=info"));
tracing_subscriber::registry()
.with(fmt::layer())
.with(filter)
.init();
});
}
```
**改动后内容**
```diff
//! agcore —— 智能体(Agent)核心工具箱。
- pub mod llm;
+ #[cfg(any(feature = "llm-types", feature = "llm"))]
+ pub mod llm;
- pub mod document;
+ #[cfg(feature = "document")]
+ pub mod document;
- pub mod prompt;
+ #[cfg(feature = "prompt")]
+ pub mod prompt;
- pub mod tools;
+ #[cfg(feature = "tools")]
+ pub mod tools;
- pub mod memory;
+ #[cfg(feature = "memory")]
+ pub mod memory;
- pub mod agent;
+ #[cfg(feature = "agent")]
+ pub mod agent;
- pub mod engine;
+ #[cfg(feature = "engine")]
+ pub mod engine;
- pub use document::Document;
+ #[cfg(feature = "document")]
+ pub use document::Document;
- use tracing_subscriber::{EnvFilter, fmt, prelude::*};
- static INIT: std::sync::Once = std::sync::Once::new();
- pub fn init_tracing() {
- INIT.call_once(|| {
- let filter = EnvFilter::try_from_default_env()
- .unwrap_or_else(|_| EnvFilter::new("agcore=info"));
- tracing_subscriber::registry()
- .with(fmt::layer())
- .with(filter)
- .init();
- });
- }
+ #[cfg(feature = "tracing-init")]
+ use tracing_subscriber::{EnvFilter, fmt, prelude::*};
+
+ #[cfg(feature = "tracing-init")]
+ static INIT: std::sync::Once = std::sync::Once::new();
+
+ #[cfg(feature = "tracing-init")]
+ pub fn init_tracing() {
+ INIT.call_once(|| {
+ let filter = EnvFilter::try_from_default_env()
+ .unwrap_or_else(|_| EnvFilter::new("agcore=info"));
+ tracing_subscriber::registry()
+ .with(fmt::layer())
+ .with(filter)
+ .init();
+ });
+ }
```
---
#### 文件 3`src/llm.rs`(当前约 12 行 → 改动后约 24 行)
**改动说明**:为每个子模块声明加 feature 门控。`types` 子模块在 `llm-types``llm` 任一 feature 启用时编译(`llm` imply `llm-types`,但 `prompt` 也 depend on `llm-types`);其余子模块(compact/convert/cycle 等)仅在 `llm` feature 启用时编译。
```diff
//! LLM 调用周期 —— 大模型基础调用周期控制。
- pub mod types;
+ #[cfg(feature = "llm-types")]
+ pub mod types;
- pub mod compact;
+ #[cfg(feature = "llm")]
+ pub mod compact;
- pub mod convert;
+ #[cfg(feature = "llm")]
+ pub mod convert;
- pub mod cycle;
+ #[cfg(feature = "llm")]
+ pub mod cycle;
- pub mod embedding;
+ #[cfg(feature = "llm")]
+ pub mod embedding;
- pub mod error;
+ #[cfg(feature = "llm")]
+ pub mod error;
- pub mod hooks;
+ #[cfg(feature = "llm")]
+ pub mod hooks;
- pub mod mock;
+ #[cfg(feature = "llm")]
+ pub mod mock;
- pub mod provider;
+ // provider 模块依赖 reqwest(通过 reqwest::Client),仅在任一 provider feature 启用时编译
+ #[cfg(any(feature = "provider-openai", feature = "provider-anthropic", feature = "provider-deepseek", feature = "provider-qwen", feature = "provider-ollama"))]
+ pub mod provider;
- pub mod stream;
+ #[cfg(feature = "llm")]
+ pub mod stream;
```
---
#### 文件 3b`src/llm/cycle.rs`(新增文件,约 35 行)
**改动说明**`cycle.rs` 中使用 `crate::tools::ToolRegistry`(第 29 行),依赖 `tools` feature。工具相关字段和方法需加 `#[cfg(feature = "tools")]` 门控。
> ⚠️ 这是 Phase 22.7 原计划的门控变更,因为编译阻塞提前到 Step 1 执行。
```diff
//! Cycle —— 多轮对话与工具调用编排。
use async_trait::async_trait;
use futures::StreamExt; // 来自 llm→tokio imply 链
+ #[cfg(feature = "tools")]
use crate::tools::ToolRegistry;
// ... struct / enum 定义 ...
// ===== CycleConfig — 工具相关字段加 cfg 门控 =====
pub struct CycleConfig {
pub max_retries: usize,
pub max_history: usize,
+ #[cfg(feature = "tools")]
pub max_tool_turns: usize,
+ #[cfg(feature = "tools")]
pub tool_timeout_secs: u64,
// ... 其他字段 ...
}
// ===== Cycle — 方法加 cfg 门控 =====
impl Cycle {
/// 仅在有 tools feature 时才有工具调用相关方法
+ #[cfg(feature = "tools")]
pub async fn submit_with_tools(&self, ...) -> Result<...> {
// ...
}
+ /// submit_with_tools_stream 方法同样需要 tools 门控,
+ /// 因参数包含 Arc<ToolRegistry> 而与 submit_with_tools 同理。
+ #[cfg(feature = "tools")]
+ pub async fn submit_with_tools_stream(
+ &self, ... // 方法签名中包含 Arc<ToolRegistry> 参数
+ ) -> Result<...> {
+ // ...
+ }
+ #[cfg(feature = "tools")]
async fn run_tool_loop(&self, ...) -> Result<...> {
// ...
}
}
+ // ===== 顶层函数 — 同样依赖 ToolRegistry =====
+ /// run_tool_loop 函数(顶层函数,非 LlmCycle 方法)同样依赖 Arc<ToolRegistry>
+ /// 参数包含 Arc<ToolRegistry>,需 #[cfg(feature = "tools")]。
+ #[cfg(feature = "tools")]
+ pub async fn run_tool_loop(
+ // ... 函数签名中包含 Arc<ToolRegistry> 参数
+ ) -> Result<...> {
+ // ...
+ }
**说明**`Cycle` 本身的 struct 定义、`submit()` 基础方法、`ResponseStream` 等不依赖 tools 的部分保持无门控,仅在 `llm` feature 下编译即可。
---
#### 文件 4`src/tools.rs`(当前约 13 行 → 改动后约 15 行)
**改动说明**`mcp` 子模块及对应的 `pub use` 仅在 `tools-mcp` feature 启用时编译。其余子模块(base/error/permission/registry)始终在 `tools` feature 下编译。
```diff
//! 工具系统 —— 工具抽象、注册、调用、权限控制与 MCP 集成。
pub mod base;
pub mod error;
- pub mod mcp;
+ #[cfg(feature = "tools-mcp")]
+ pub mod mcp;
pub mod permission;
pub mod registry;
pub use base::{BaseTool, ToolContext, ToolRef};
pub use error::ToolError;
- pub use mcp::{McpClient, McpTransport};
+ #[cfg(feature = "tools-mcp")]
+ pub use mcp::{McpClient, McpTransport};
pub use permission::{Permission, PermissionChecker, PermissionConfig};
pub use registry::{ToolInvocation, ToolRegistry};
```
---
#### 文件 5`src/memory/store.rs`(当前约 62 行 → 改动后约 64 行)
**改动说明**`sqlite_store` 子模块及其 `pub use` 仅在 `memory-sqlite` feature 启用时编译。
```diff
//! MemoryStore 抽象接口与默认实现。
use async_trait::async_trait;
use crate::memory::error::MemoryError;
use crate::memory::types::{MemoryFilter, MemoryItem};
pub mod in_memory;
- pub mod sqlite_store;
+ #[cfg(feature = "memory-sqlite")]
+ pub mod sqlite_store;
pub use in_memory::InMemoryStore;
- pub use sqlite_store::SqliteStore;
+ #[cfg(feature = "memory-sqlite")]
+ pub use sqlite_store::SqliteStore;
```
**说明**`MemoryStore` trait、`EvictionConfig``EvictionPolicy` 等定义保持不变,不需要 cfg 门控。
---
#### 文件 6`src/memory.rs`(当前约 32 行 → 改动后约 34 行)
**改动说明**`SqliteStore` 的重新导出仅在 `memory-sqlite` feature 启用时编译。其余子模块声明和 `pub use` 保持不变(`memory` feature 门控由 `src/lib.rs` 负责)。
```diff
//! 记忆系统 —— 对话消息管理、知识页面存储与关键词检索。
// 所有子模块声明保持不变:
// pub mod conversation;
// pub mod error;
// pub mod graph;
// pub mod knowledge;
// pub mod retriever;
// pub mod store;
// ...
// 高频类型
pub use conversation::{ConversationMemory, ConversationMemoryConfig};
pub use error::MemoryError;
pub use graph::{GraphEntity, GraphRelation, InMemoryGraph, KnowledgeGraph, RelationDirection, ScoredEntity};
pub use knowledge::KnowledgeStore;
pub use retriever::MemoryRetriever;
pub use store::{InMemoryStore, MemoryStore};
+ #[cfg(feature = "memory-sqlite")]
+ pub use store::SqliteStore;
// 其余 pub use 保持不变...
```
---
#### 文件 7`src/agent/session.rs`(新增文件,约 30 行)
**改动说明**`session.rs` 中引用了 `crate::engine::*`SessionMemoryEntry、SessionSnapshot、EngineError),而 `agent` feature 不含 `engine``engine = ["agent"]` 是反向依赖)。需要对 engine 相关导入和方法加门控。
> ⚠️ 阻塞 B3`src/agent/session.rs:28-29` 无条件引用 `crate::engine::*`,在 `agent` feature 下编译时因缺少 engine 而失败。
```diff
//! Session —— Agent 会话管理。
use async_trait::async_trait;
use crate::llm::types::LLMRequest;
use crate::memory::MemoryStore;
+ #[cfg(feature = "engine")]
use crate::engine::snapshot::{SessionMemoryEntry, SessionSnapshot};
+ #[cfg(feature = "engine")]
use crate::engine::EngineError;
// ===== AgentSession — pending_memory_restore 字段 =====
+ /// AgentSession 结构体中的 pending_memory_restore 字段类型来自 engine 模块,
+ /// 需要条件编译。
pub struct AgentSession {
+ // ... 其他字段 ...
+
+ #[cfg(feature = "engine")]
+ pending_memory_restore: Option<HashMap<String, SessionMemoryEntry>>,
+ // ... 其他字段 ...
+ }
impl Session {
/// to_snapshot / from_snapshot / restore_memory 仅在 engine feature 下可用
+ #[cfg(feature = "engine")]
pub fn to_snapshot(&self) -> SessionSnapshot {
// ...
}
+ #[cfg(feature = "engine")]
pub fn from_snapshot(snap: SessionSnapshot) -> Result<Self, EngineError> {
// ...
}
+ #[cfg(feature = "engine")]
async fn restore_memory(&mut self, entries: Vec<SessionMemoryEntry>) -> Result<(), MemoryError> {
// ...
}
}
```
**说明**`Session` 结构体本身以及不依赖 engine 的方法(如 `new()``add_message()``get_history()`)保持无门控,仅在 `agent` feature 下编译即可。
---
## 4. 实施步骤
**3 个 commit** 粒度执行,每个 commit 后编译验证。
### Commit 1Cargo.toml features 定义 + 依赖 optional 化
**涉及文件**:仅 `Cargo.toml`
**操作清单**
1.`[package]` 之后、`[dependencies]` 之前插入 `[features]` 表(16 个 features + 4 个快捷组合,约 45 行)
2. tokio features 从 `["full"]` 改为 `["rt", "sync", "time", "macros", "process", "io-util"]` 并加 `optional = true`
3. reqwest / rusqlite / tracing-subscriber / tokio-stream / futures / futures-util / futures-core / bytes / async-stream / tokio-util / time 共 11 个依赖加 `optional = true`
4.`[dependencies]` 之后新增 `[dev-dependencies]``tokio = { version = "1", features = ["rt", "macros"] }`
**验证**
```bash
cargo build --no-default-features # 不依赖任何 optional crate,应通过
cargo build -F document # 零外部依赖,应通过
```
---
### Commit 2cfg 门控(pub mod + pub use + init_tracing
**涉及文件**`src/lib.rs``src/llm.rs``src/llm/cycle.rs``src/tools.rs``src/memory/store.rs``src/memory.rs``src/agent/session.rs`
**操作清单**
按文件逐一执行:
1. `src/lib.rs` — 7 个 `pub mod``#[cfg(feature = "...")]``Document` pub use 加 cfg、`init_tracing` 整体用 `#[cfg(feature = "tracing-init")]` 包裹
2. `src/llm.rs` — 10 个子模块按 `llm-types` / `llm` / provider 分类门控
3. `src/llm/cycle.rs``ToolRegistry` 导入加 `#[cfg(feature = "tools")]`,工具字段和方法加相同门控
4. `src/tools.rs``pub mod mcp``pub use mcp::*``#[cfg(feature = "tools-mcp")]`
5. `src/memory/store.rs``pub mod sqlite_store``pub use sqlite_store::SqliteStore``#[cfg(feature = "memory-sqlite")]`
6. `src/memory.rs``pub use store::SqliteStore``#[cfg(feature = "memory-sqlite")]`
7. `src/agent/session.rs` — engine 相关导入加 `#[cfg(feature = "engine")]`to_snapshot/from_snapshot/restore_memory 加相同门控
**验证**
```bash
cargo build -F "full" # 全量回归
cargo build -F "prompt" # 验证 llm::types imply 路径
cargo build -F "tools" # 验证 tokio imply 路径
cargo build -F "memory" # 验证记忆模块不含 sqlite
cargo build -F "chat,provider-openai" # 纯对话组合
cargo build -F "chat,provider-openai,tools-mcp" # 带 MCP 对话
```
---
### Commit 3Checkpoint 全量验证
**操作清单**
1. 完整的验证矩阵执行(见第 5 节)
2. `cargo test -F "full"` 确认 427 passed
**验证**
```bash
cargo test -F "full"
cargo build -F "light"
cargo build -F "multi"
```
## 5. 验证标准
### 编译验证矩阵
| 命令 | 验证目标 | 预期结果 |
|------|---------|---------|
| `cargo build --no-default-features` | 空 crate | 编译通过(无模块) |
| `cargo build -F "full"` | 全量回归 | 编译通过,与 v0.3.0 语义一致 |
| `cargo test -F "full"` | 测试回归 | `427 passed` |
| `cargo build -F "document"` | 文档模块独立 | 编译通过,零外部依赖 |
| `cargo build -F "prompt"` | 提示词独立 | 编译通过,`llm::types` imply 路径正确 |
| `cargo build -F "tools"` | 工具独立 | 编译通过,tokio imply 路径正确 |
| `cargo build -F "memory"` | 记忆模块独立编译 | ✅ 不含 sqlite(依赖 B1/B2 修复) |
| `cargo build -F "memory-sqlite"` | 含 SQLite 的记忆模块 | ✅ 含 rusqlite |
| `cargo build -F "agent"` | Agent 独立编译 | ✅ 含 llm+tools+memory,不含 engine(依赖 B1/B3 修复) |
| `cargo build -F "engine"` | Engine 独立编译 | ✅ imply agent → llm+tools+memory |
| `cargo build -F "chat,provider-openai"` | 纯对话组合 | 编译通过,不含 MCP、sqlite |
| `cargo build -F "chat,provider-openai,tools-mcp"` | 带 MCP 对话 | 编译通过,含 reqwest 无 sqlite |
| `cargo build -F "light"` | 生产常用组合 | 编译通过 |
| `cargo build -F "multi"` | 多 provider 组合 | 编译通过 |
### 验证操作指令
每次编译验证后执行(验证编译产物不含意外符号):
```bash
# 确认空 crate 确实没有模块符号
cargo build --no-default-features 2>&1 && echo "OK"
# 确认 document 零外部依赖(无 reqwest/rusqlite 等符号)
cargo build -F "document" 2>&1 && echo "OK"
# 全量构建 + 测试
cargo build -F "full" 2>&1 && cargo test -F "full" 2>&1 | tail -5
```
### 验证通过条件
- 所有 14 条编译验证命令返回 exit code 0
- `cargo test -F "full"` 输出 `427 passed`(与 v0.3.0 基线一致,不要求测试数精确匹配,但必须全部通过且数量合理)
-`unused import` / `unused variable` / `dead code` warning(由 `#[cfg]` 引起的新 warning 需逐一修复)
## 6. 风险评估
| 风险 | 等级 | 缓解措施 |
|------|------|---------|
| **tokio features 拆细遗漏**:某些代码路径用到 net/fs/signal | 低 | 已通过 SA(静态分析)验证全库无相关使用 |
| **`#[tokio::test]` 编译失败**:测试代码无 tokio runtime | 低 | `[dev-dependencies]` 添加 `tokio = { version = "1", features = ["rt", "macros"] }` |
| **下游 transitive tokio features 缩小**:依赖 agcore 的 crate 之前通过 agcore 间接获得 `full` tokio,现在范围缩小 | 中 | Phase 27 README 发布说明中明确告知迁移方案;下游如需完整 tokio 需自行添加 |
| **imply 链未闭合**:某个 feature 依赖了未 imply 的 feature | 低 | 7 种特征组合全部逐条构建验证;features 定义中有交叉引用的全部显式列出 |
| **unused cfg warning**:某些 `#[cfg]` 标记导致编译 warning | 低 | 每个 commit 后检查编译器输出,发现后立即修复 |
| **测试依赖循环**dev-deps 与普通 deps 版本冲突 | 低 | dev-deps 的 tokio 版本与主依赖保持一致 (`version = "1"`,由 cargo 自动选择兼容版本) |
| **memory 跨模块 imply 链** | **高** | memory 模块依赖 llmconversation/vector_store)、tokioknowledge)、timetypes),imply 链必须完整传递 | `memory` feature 定义已包含 `llm``tokio``time`;验证矩阵覆盖 memory 独立编译 |
| **跨模块引用未门控** | **高** | provider.rs 依赖 reqwest、cycle.rs 依赖 ToolRegistry、session.rs 依赖 engineStep 1 必须添加 cfg 门控 | provider 模块 cfg 改为 provider-xxx 条件;cycle.rs 加 tools 门控;session.rs 加 engine 门控 |