NET 10：Oxigraph 的 FFI 桥接实践

架构概览┌──────────────────────────────────────────────────────────┐ │ .NET 应用层 │ │ ┌──────────┐ ┌────────────────┐ ┌──────────────────┐ │ │ │ Model │ │ Store │ │ CustomFunctions │ │ │ │ (record) │ │ (SPARQLI/O) │ │ (C#→Rust 回调) │ │ │ └────┬─────┘ └───────┬────────┘ └────────┬─────────┘ │ │ │ │ │ │ │ ┌────┴────────────────┴────────────────────┴─────────┐ │ │ │ Interop 层 (FFI/Json) │ │ │ │ NativeMethods.g.cs │ SafeHandles │ FFIHelper │ │ │ │ StreamInterop.cs │ Term JSON Converters │ │ │ └──────────────────────┬──────────────────────────────┘ │ └─────────────────────────┼─────────────────────────────────┘ │ P/Invoke JSON 字符串 ┌─────────────────────────┼─────────────────────────────────┐ │ oxigraph_dotnet.dll │ (Rust cdylib) │ │ ┌──────────────────────┴──────────────────────────────┐ │ │ │ ffi.rs (~2774 行) │ │ │ │ ├─ Store CRUD/Match/SPARQL │ │ │ │ ├─ File/Stream/Callback I/O │ │ │ │ ├─ Lazy Query Iterator │ │ │ │ ├─ Chunked Bulk Loader │ │ │ │ ├─ QueryResults Serialization │ │ │ │ ├─ Dataset (In-Memory) Canonicalization │ │ │ │ └─ Custom Functions / Aggregate │ │ │ ├─ stream_ffi.rs: CallbackReader / CallbackWriter │ │ │ ├─ error.rs: {ok:...} / {error:...} 协议 │ │ │ └─ model_ffi.rs: JSON↔Quad 序列化 │ │ │ ┌─────────────────────────────────────────────────────┐ │ │ │ oxigraph oxrdf (Rust crate) RocksDB │ │ │ └─────────────────────────────────────────────────────┘ │ └──────────────────────────────────────────────────────────┘核心设计原则零复制复杂数据类型全量 JSON 序列化跨 FFI 边界。Rust 侧所有函数签名遵循统一模式(*const c_char) → *mut c_char输入和输出都是 JSON 字符串。二、工具链2.1 Rust 侧cdylibdotnet/src/oxigraph-dotnet/Cargo.toml[package] name oxigraph-dotnet version 0.5.7 edition 2024 [lib] crate-type [cdylib] # ← 编译为动态链接库 [dependencies] oxigraph { path ../../../lib/oxigraph } oxrdf { path ../../../lib/oxrdf, features [serde] } serde { version 1, features [derive] } serde_json 1 [profile.release] opt-level z # 体积优化减小 .dll 尺寸 lto true # 链接时优化产物Windows:oxigraph_dotnet.dllLinux:liboxigraph_dotnet.somacOS:liboxigraph_dotnet.dylib2.2 .NET 侧LibraryImport生成源.NET 10 使用源生成器LibraryImport取代旧的DllImport进行 P/Invoke// NativeMethods.g.cs — 自动生成的 FFI 绑定 internal static partial class OxigraphNative { private const string LibName oxigraph_dotnet; [LibraryImport(LibName, EntryPoint oxigraph_store_open, StringMarshalling StringMarshalling.Utf8)] internal static partial IntPtr store_open(string path); [LibraryImport(LibName, EntryPoint oxigraph_store_query, StringMarshalling StringMarshalling.Utf8)] internal static partial IntPtr store_query(IntPtr handle, string queryJson); [LibraryImport(LibName, EntryPoint oxigraph_register_custom_function, StringMarshalling StringMarshalling.Utf8)] internal static partial IntPtr register_custom_function( string name, IntPtr callback); // ... 共 60 个 FFI 函数 }关键差异LibraryImport是源生成器模式编译时生成封送代码避免了DllImport的运行时 IL 生成和 AOT 兼容问题。2.3 构建流水线build_package.py编排整个构建过程# 步骤 1: 编译 Rust cdylib cargo build --release -p oxigraph-dotnet --features rocksdb # 步骤 2: 编译 C# 项目 dotnet build dotnet/ -c Release # 步骤 3: 将 .dll 拷贝到测试输出目录 # 步骤 4: 运行 xUnit 测试 dotnet test dotnet/tests/Oxigraph.Tests -c Release三、数据协议JSON over FFI3.1 统一响应格式所有 Rust FFI 函数返回*mut c_charJSON 字符串包含两种可能的格式{ok: result} // 或 {error: {kind: store|parse|invalid_argument, message: ...}}C# 侧FFIHelper统一处理// 带返回值的调用 internal static T CallT(FuncIntPtr ffiCall) where T : class { IntPtr jsonPtr ffiCall(); string json ReadAndFree(jsonPtr); // Marshal.PtrToStringUTF8 free_string ThrowIfError(json); // 检查 {error:...} using var doc JsonDocument.Parse(json); return JsonSerializer.DeserializeT( doc.RootElement.GetProperty(ok).GetRawText())!; } // 无返回值的调用 internal static void CallVoid(FuncIntPtr ffiCall) { IntPtr jsonPtr ffiCall(); string json ReadAndFree(jsonPtr); ThrowIfError(json); }3.2 错误映射Rust 错误类型自动映射为 C# 异常private static Exception MapError(string kind, string message) { return kind switch { store new StoreException(message), parse new ParseException(message), invalid_argument new ArgumentException(message), _ new OxigraphException($[{kind}] {message}), }; }3.3 不透明句柄Opaque HandlesRust 对象不直接暴露给 C#而是通过指针传递// Rust 侧Store 包装为 BoxUnsafeCellStore返回指针的 u64 表示 fn store_to_handle(store: Store) - *mut c_char { let boxed Box::new(UnsafeCell::new(store)); let ptr Box::into_raw(boxed); let handle_value ptr as u64; // 返回 {ok:{handle:12345678}} ok_json(handle_value) }// C# 侧SafeHandle 封装原始指针确保析构 internal sealed class StoreSafeHandle : SafeHandleZeroOrMinusOneIsInvalid { protected override bool ReleaseHandle() { OxigraphNative.store_destroy(handle); // Rust 侧 drop(Box::from_raw(ptr)) return true; } }所有句柄类型Rust 类型C# SafeHandle用途StoreHandle*mut UnsafeCellStoreStoreSafeHandleStore 读写DatasetHandle*mut UnsafeCellDatasetDatasetSafeHandleDatasetQuadIterHandle*mut UnsafeCellReaderQuadParserQuadIterSafeHandle懒解析迭代器QueryResultsHandle*mut QueryResultsWrapperQueryResultsSafeHandle流式查询结果四、Stream 回调机制对于 .NET Stream 的读写LoadFromStream/DumpToStream无法简单地把 Stream传过去因为 Rust 不认识 .NET 的Stream类型。解决方案C 风格回调。4.1 Rust 侧实现Read/Writetraitpub type ReadFn unsafe extern C fn( context: *mut c_void, buf: *mut u8, buf_size: i32 ) - i32; pub struct CallbackReader { context: *mut c_void, callback: ReadFn, } impl Read for CallbackReader { fn read(mut self, buf: mut [u8]) - io::Resultusize { let result unsafe { (self.callback)(self.context, buf.as_mut_ptr(), buf.len() as i32) }; match result { n if n 0 Ok(n as usize), 0 Ok(0), // EOF -1 Err(...), // Error } } }CallbackWriter对Writetrait 实现类似。4.2 C# 侧GCHandle 固定的委托[UnmanagedFunctionPointer(CallingConvention.Cdecl)] internal delegate int ReadCallback(IntPtr context, IntPtr buffer, int bufferSize); internal sealed class ReadContext : IDisposable { private readonly Stream _stream; private readonly byte[] _buffer; private readonly GCHandle _gcHandle; // ← 防止 GC 回收委托 public readonly IntPtr ContextPtr; public readonly ReadCallback Callback; public ReadContext(Stream stream) { _stream stream; Callback ReadImpl; _gcHandle GCHandle.Alloc(Callback); // 固定委托 ContextPtr (IntPtr)_gcHandle; // 作为 context 传递 } private int ReadImpl(IntPtr context, IntPtr buffer, int bufferSize) { try { int read _stream.Read(_buffer, 0, Math.Min(bufferSize, _buffer.Length)); Marshal.Copy(_buffer, 0, buffer, read); return read; } catch { return -1; } } public void Dispose() { _gcHandle.Free(); } }调用链C# Stream.Read()→ 拷贝到 Rust buffer → RustReadtrait →RdfParser.for_reader()五、自定义函数回调桥SPARQL 自定义函数需要 Rust 在执行查询时回调到 C#。这是最复杂的 FFI 场景Rust 调用 C#C# 再把结果返回给 Rust。5.1 简单自定义函数// Rust 侧函数指针类型 type CustomFnCallback unsafe extern C fn(args_json: *const c_char) - *mut c_char; // 全局注册表 static CUSTOM_FUNCTIONS: LazyLockMutexHashMapString, CustomFnCallback ...; #[unsafe(no_mangle)] pub extern C fn oxigraph_register_custom_function( name: *const c_char, callback: CustomFnCallback, // C# 函数的指针 ) - *mut c_char { CUSTOM_FUNCTIONS.lock().unwrap().insert(name_str, callback); ok_json(registered) }Rust 收到callback后在 SPARQL 评估时调用它Rust SPARQL evaluator → 遇到 my:func(?x) → 查找 CUSTOM_FUNCTIONS[http://example.com/myFunc] → 调用 callback(args_json) // ← 跨 FFI 边界回调到 C# → 收到返回的 result_json → 继续 SPARQL 评估C# 侧// BridgeDelegate接收 JSON 参数数组返回 JSON Term private delegate IntPtr BridgeDelegate(IntPtr argsJsonPtr); private static IntPtr BridgeImpl(IntPtr argsJsonPtr) { var json Marshal.PtrToStringUTF8(argsJsonPtr); // json: [http://example.com/myFunc, {type:literal,value:hello}] using var doc JsonDocument.Parse(json); var name doc.RootElement[0].GetString(); // 解析剩余元素为 ITerm[] var terms ...; // 调用 .NET 函数 var result _functions[name](terms); // 序列化结果返回给 Rust return Marshal.StringToHGlobalAnsi(JsonSerializer.Serialize(result)); } // 固定 BridgeDelegate防止 GC 回收 private static readonly GCHandle _gcHandle GCHandle.Alloc(_bridge); private static readonly IntPtr _bridgePtr Marshal.GetFunctionPointerForDelegate(_bridge);5.2 自定义聚合函数聚合函数需要更复杂的 4 回调协议Rust 调用: new_fn() → 返回 ctx handleGCHandle 包装的 C# 对象 acc_fn(ctx, term_json) → 每行数据调用一次 finish_fn(ctx) → 返回聚合结果 free_fn(ctx) → 释放 C# 对象// Rust 适配器实现 oxigraph::sparql::AggregateFunctionAccumulator struct CallbackAggregateAccumulator { ctx: *mut c_void, acc_fn: AggregateAccCallback, finish_fn: AggregateFinishCallback, free_fn: AggregateFreeCallback, } impl AggregateFunctionAccumulator for CallbackAggregateAccumulator { fn accumulate(mut self, element: Term) { let json serde_json::to_string(element).unwrap(); let c_str CString::new(json).unwrap(); unsafe { (self.acc_fn)(self.ctx, c_str.as_ptr()) }; } fn finish(mut self) - OptionTerm { let ptr unsafe { (self.finish_fn)(self.ctx) }; // 解析 C# 返回的 JSON Term 或 null if ptr.is_null() { return None; } let json unsafe { c_str_to_str(ptr) }; serde_json::from_str(json).unwrap_or(None) } }六、RDF 数据类型的 JSON 序列化Rust 的 serde 格式直接映射到 C# 的System.Text.Json。两端使用相同的tagged-enum JSON 模式// NamedNode {type: uri, value: http://example.com/s} // BlankNode {type: bnode, value: b1_abc123} // Literal (plain) {type: literal, value: hello} // Literal (language-tagged) {type: literal, value: bonjour, language: fr} // Literal (typed) {type: literal, value: 42, datatype: {type:uri,value:http://www.w3.org/2001/XMLSchema#integer}} // Triple (RDF-star) {type: triple, subject: {type:uri,value:http://s}, predicate: {type:uri,value:http://p}, object: {type:uri,value:http://o}} // DefaultGraph {type: default}C# 侧自定义 Converterpublic class TermConverter : JsonConverterITerm { public override ITerm? Read(ref Utf8JsonReader reader, ...) { using var doc JsonDocument.ParseValue(ref reader); var kind doc.RootElement.GetProperty(type).GetString(); return kind switch { uri new NamedNode(doc.RootElement.GetProperty(value).GetString()!), bnode new BlankNode(doc.RootElement.GetProperty(value).GetString()!), literal /* 解析 value/language/datatype/direction */, triple JsonSerializer.DeserializeTriple(root.GetRawText())!, _ throw new JsonException($Unknown term type: {kind}) };