feat: add diskann on arm

.github/workflows/03-macos-linux-build.yml

@@ -203,7 +203,6 @@ jobs:
           cd "$GITHUB_WORKSPACE"
           python -m pytest python/tests/test_collection_diskann.py -v
         shell: bash
-
       # Verify installing libaio does not affect existing non-DiskAnn tests.
       - name: Run HNSW Tests (w/ libaio)
         if: matrix.platform == 'linux-x64'

CMakeLists.txt

@@ -122,14 +122,17 @@ else()
 endif()
 message(STATUS "RABITQ_ARCH_FLAG: ${RABITQ_ARCH_FLAG}")
 
-# DiskAnn support (Linux x86_64 only; libaio loaded at runtime via dlopen)
-if(CMAKE_SYSTEM_NAME STREQUAL "Linux" AND CMAKE_SYSTEM_PROCESSOR MATCHES "x86_64|i686|i386" AND NOT ANDROID AND NOT IOS)
+# DiskAnn support:
+#   - Linux (x86_64, i686, i386, aarch64, arm64) with libaio (loaded via dlopen)
+#   - macOS (x86_64, ARM64/Apple Silicon) with kqueue
+if((CMAKE_SYSTEM_NAME STREQUAL "Linux" AND CMAKE_SYSTEM_PROCESSOR MATCHES "x86_64|i686|i386|aarch64|arm64" AND NOT ANDROID AND NOT IOS)
+    OR (CMAKE_SYSTEM_NAME STREQUAL "Darwin" AND NOT IOS))
   set(DISKANN_SUPPORTED ON)
   add_definitions(-DDISKANN_SUPPORTED=1)
 else()
   set(DISKANN_SUPPORTED OFF)
   add_definitions(-DDISKANN_SUPPORTED=0)
-  message(STATUS "DiskAnn support disabled - only supported on Linux x86_64")
+  message(STATUS "DiskAnn support disabled - supported on Linux (x86_64/ARM64 with libaio) and macOS (with kqueue)")
 endif()
 message(STATUS "DISKANN_SUPPORTED: ${DISKANN_SUPPORTED}")
 

README.md

@@ -35,14 +35,14 @@
 **Zvec** is an open-source, in-process vector database — lightweight, lightning-fast, and designed to embed directly into applications. Battle-tested within Alibaba Group, it delivers production-grade, low-latency and scalable similarity search with minimal setup.
 
 > [!Important]
-> 🚀 **v0.5.0 (June 12, 2026)**
+> 🚀 **v0.6.0 (July 20, 2026)**
 >
-> - **Full-Text Search (FTS)**: Native full-text search — attach an FTS index to any string field and query it with natural-language or structured expressions, no external search engine required.
-> - **Hybrid Retrieval**: Combine full-text, dense-vector, and sparse-vector retrieval in a single query call, with scalar filtering and reranking.
-> - **DiskANN Index**: New on-disk index that keeps the bulk of the index on disk, drastically cutting memory usage for large-scale datasets.
-> - **Ecosystem & Platforms**: New official [Go](https://github.com/zvec-ai/zvec-go) / [Rust](https://github.com/zvec-ai/zvec-rust) SDKs, the [Zvec Studio](https://github.com/zvec-ai/zvec-studio) visual tool, and RISC-V support.
+> - **Group-By Search**: Retrieve top-K results per group instead of globally (group-by deduplication) across Flat, HNSW, HNSW-RaBitQ, and sparse indexes.
+> - **Random Rotation Quantization**: Optional random rotation for INT8/INT4 quantization distributes variance evenly across dimensions, significantly boosting recall.
+> - **Enhanced Full-Text Search**: Upgraded FTS pipeline with a Unicode UAX #29 standard tokenizer, UTF-8 / ASCII folding, and a Snowball-based stemmer supporting 34+ languages.
+> - **Faster & More Robust**: Block-max skip speeds up FTS conjunction queries by 22–38%, plus a new DiskANN C API and numerous stability fixes.
 >
-> 👉 [Read the Release Notes](https://github.com/alibaba/zvec/releases/tag/v0.5.0) | [View Roadmap 📍](https://github.com/alibaba/zvec/issues/309)
+> 👉 [Read the Release Notes](https://github.com/alibaba/zvec/releases/tag/v0.6.0) | [View Roadmap 📍](https://github.com/alibaba/zvec/issues/309)
 
 ## 💫 Features
 
@@ -62,7 +62,7 @@ Zvec offers official SDKs across multiple languages:
 - **[Python](https://pypi.org/project/zvec/)**: `pip install zvec` (requires 64-bit Python 3.10–3.14)
 - **[Node.js](https://www.npmjs.com/package/@zvec/zvec)**: `npm install @zvec/zvec`
 - **[Go](https://github.com/zvec-ai/zvec-go)**: High-performance Go bindings.
-- **[Rust](https://github.com/zvec-ai/zvec-rust)**: High-performance Rust bindings.
+- **[Rust](https://crates.io/crates/zvec-rust)**: `cargo add zvec-rust`
 - **[Dart/Flutter](https://pub.dev/packages/zvec)**: `flutter pub add zvec`
 
 Prefer a visual tool? Try **[Zvec Studio](https://github.com/zvec-ai/zvec-studio)** to browse data and debug queries — no code required.
@@ -123,7 +123,7 @@ For detailed benchmark methodology, configurations, and complete results, please
 
 | 💬 DingTalk | 📱 WeChat | 🎮 Discord | X (Twitter) |
 | :---: | :---: | :---: | :---: |
-| <img src="https://zvec.oss-cn-hongkong.aliyuncs.com/qrcode/dingding.png" width="150" alt="DingTalk QR Code"/> | <img src="https://zvec.oss-cn-hongkong.aliyuncs.com/qrcode/wechat.png" width="150" alt="WeChat QR Code"/> | [![Discord](https://img.shields.io/badge/Discord-Join%20Server-5865F2?style=for-the-badge&logo=discord&logoColor=white)](https://discord.gg/rKddFBBu9z) | [![X (formerly Twitter) Follow](https://img.shields.io/twitter/follow/ZvecAI)](<https://x.com/ZvecAI>) |
+| <img src="https://zvec.oss-cn-hongkong.aliyuncs.com/qrcode/dingding.png" width="150" alt="DingTalk QR Code"/> | <img src="https://zvec.oss-cn-hongkong.aliyuncs.com/qrcode/wechat.png?v1" width="150" alt="WeChat QR Code"/> | [![Discord](https://img.shields.io/badge/Discord-Join%20Server-5865F2?style=for-the-badge&logo=discord&logoColor=white)](https://discord.gg/rKddFBBu9z) | [![X (formerly Twitter) Follow](https://img.shields.io/twitter/follow/ZvecAI)](<https://x.com/ZvecAI>) |
 | Scan to join | Scan to join | Click to join | Click to follow |
 
 </div>

README_CN.md

@@ -35,14 +35,14 @@
 **Zvec** 是一款开源的嵌入式(进程内)向量数据库 — 轻量、极速，可直接嵌入应用程序。以极简的配置提供生产级、低延迟、可扩展的向量检索能力。
 
 > [!IMPORTANT]
-> 🚀  **v0.5.0（2026 年 6 月 12 日）**
+> 🚀  **v0.6.0（2026 年 7 月 20 日）**
 >
-> - **全文检索（FTS）**：原生全文检索能力——可为任意字符串字段挂载 FTS 索引，使用自然语言或结构化表达式检索，无需外接搜索引擎。
-> - **混合检索**：在单次查询调用中融合全文、稠密向量和稀疏向量检索，并支持标量过滤与重排。
-> - **DiskANN 索引**：全新磁盘索引，将索引主体存于磁盘，大幅降低大规模数据集的内存占用。
-> - **生态与平台**：全新官方 [Go](https://github.com/zvec-ai/zvec-go) / [Rust](https://github.com/zvec-ai/zvec-rust) SDK、可视化工具 [Zvec Studio](https://github.com/zvec-ai/zvec-studio)，以及 RISC-V 架构支持。
+> - **分组检索（Group-By）**：支持按分组去重检索，返回每个分组的 Top-K 结果而非全局 Top-K，覆盖 Flat、HNSW、HNSW-RaBitQ 与稀疏索引。
+> - **随机旋转量化**：为 INT8/INT4 量化新增可选的随机旋转能力，将方差均匀分布到各维度，显著提升召回率。
+> - **全文检索增强**：升级 FTS 文本分析管线，新增基于 Unicode UAX #29 的标准分词器、UTF-8 / ASCII 折叠，以及基于 Snowball、支持 34+ 种语言的词干提取过滤器。
+> - **更快更稳健**：Block-max 跳跃优化使 FTS 合取查询提速 22–38%，同时新增 DiskANN C API，以及大量稳定性修复。
 >
-> 👉 [查看更新日志](https://github.com/alibaba/zvec/releases/tag/v0.5.0) | [查看路线图 📍](https://github.com/alibaba/zvec/issues/309)
+> 👉 [查看更新日志](https://github.com/alibaba/zvec/releases/tag/v0.6.0) | [查看路线图 📍](https://github.com/alibaba/zvec/issues/309)
 
 ## 💫 核心特性
 
@@ -62,7 +62,7 @@ Zvec 提供多语言官方 SDK：
 - **[Python](https://pypi.org/project/zvec/)**：`pip install zvec`（需 64 位 Python 3.10–3.14）
 - **[Node.js](https://www.npmjs.com/package/@zvec/zvec)**：`npm install @zvec/zvec`
 - **[Go](https://github.com/zvec-ai/zvec-go)**：高性能的 Go 绑定。
-- **[Rust](https://github.com/zvec-ai/zvec-rust)**：高性能的 Rust 绑定。
+- **[Rust](https://crates.io/crates/zvec-rust)**：`cargo add zvec-rust`
 - **[Dart/Flutter](https://pub.dev/packages/zvec)**：`flutter pub add zvec`
 
 想要图形界面？试试 **[Zvec Studio](https://github.com/zvec-ai/zvec-studio)**，零代码浏览数据与调试查询。
@@ -121,7 +121,7 @@ Zvec 提供极致的速度和效率，能够轻松应对高要求的生产环境
 
 | 💬 钉钉群 | 📱 微信群 | 🎮 Discord | X (Twitter) |
 | :---: | :---: | :---: | :---: |
-| <img src="https://zvec.oss-cn-hongkong.aliyuncs.com/qrcode/dingding.png" width="150" alt="钉钉二维码"/> | <img src="https://zvec.oss-cn-hongkong.aliyuncs.com/qrcode/wechat.png" width="150" alt="微信二维码"/> | [![Discord](https://img.shields.io/badge/Discord-Join%20Server-5865F2?style=for-the-badge&logo=discord&logoColor=white)](https://discord.gg/rKddFBBu9z) | [![X (formerly Twitter) Follow](https://img.shields.io/twitter/follow/ZvecAI)](<https://x.com/ZvecAI>) |
+| <img src="https://zvec.oss-cn-hongkong.aliyuncs.com/qrcode/dingding.png" width="150" alt="钉钉二维码"/> | <img src="https://zvec.oss-cn-hongkong.aliyuncs.com/qrcode/wechat.png?v1" width="150" alt="微信二维码"/> | [![Discord](https://img.shields.io/badge/Discord-Join%20Server-5865F2?style=for-the-badge&logo=discord&logoColor=white)](https://discord.gg/rKddFBBu9z) | [![X (formerly Twitter) Follow](https://img.shields.io/twitter/follow/ZvecAI)](<https://x.com/ZvecAI>) |
 | 扫码加入 | 扫码加入 | 点击加入 | 点击关注 |
 
 </div>

examples/c++/CMakeLists.txt

@@ -15,6 +15,7 @@ endif()
 get_filename_component(ZVEC_ROOT_DIR "${CMAKE_CURRENT_LIST_DIR}/../.." ABSOLUTE)
 
 set(ZVEC_INCLUDE_DIR ${ZVEC_ROOT_DIR}/src/include)
+set(ZVEC_SRC_DIR ${ZVEC_ROOT_DIR}/src)
 set(ZVEC_LIB_DIR ${ZVEC_ROOT_DIR}/${HOST_BUILD_DIR}/lib)
 
 include_directories(${ZVEC_INCLUDE_DIR})

python/tests/detail/fixture_helper.py

@@ -2,12 +2,10 @@
 import logging
 import platform
 
-DISKANN_SUPPORTED = platform.system() == "Linux" and platform.machine() in (
-    "x86_64",
-    "AMD64",
-    "i686",
-    "i386",
-)
+DISKANN_SUPPORTED = (
+    platform.system() == "Linux"
+    and platform.machine() in ("x86_64", "AMD64", "i686", "i386", "aarch64", "arm64")
+) or platform.system() == "Darwin"
 
 from typing import Any, Generator
 from zvec.typing import DataType, StatusCode, MetricType, QuantizeType
@@ -27,9 +25,8 @@ def _ensure_diskann_runtime_or_reason() -> str | None:
     _DISKANN_PRELOAD_DONE = True
 
     if not DISKANN_SUPPORTED:
-        _DISKANN_PRELOAD_REASON = "DiskAnn only supported on Linux x86_64"
+        _DISKANN_PRELOAD_REASON = "DiskAnn is supported on Linux (x86_64/ARM64 with libaio) and macOS (kqueue)"
         return _DISKANN_PRELOAD_REASON
-
     _DISKANN_PRELOAD_REASON = None
     return None
 

python/tests/test_collection_diskann.py

@@ -17,8 +17,8 @@
 
 Two platform-level prerequisites are enforced at module import time:
 
-1. DiskAnn is currently built only for Linux x86_64 — other platforms are
-   skipped wholesale.
+1. DiskAnn is built for Linux (x86_64/ARM64 with libaio) and macOS (with kqueue) —
+   other platforms are skipped wholesale.
 2. libaio is loaded eagerly (via dlopen) inside DiskAnnBuilder::init() /
    DiskAnnStreamer::init(). If libaio is missing, DiskAnn falls back to
    synchronous pread() — the tests still run but with degraded performance.
@@ -39,8 +39,14 @@
 # Platform gating (must happen BEFORE we touch zvec).
 # --------------------------------------------------------------------------- #
 pytestmark = pytest.mark.skipif(
-    not (sys.platform == "linux" and platform.machine() in ("x86_64", "AMD64")),
-    reason="DiskAnn plugin is only supported on Linux x86_64",
+    not (
+        (
+            sys.platform == "linux"
+            and platform.machine() in ("x86_64", "AMD64", "aarch64", "arm64")
+        )
+        or sys.platform == "darwin"
+    ),
+    reason="DiskAnn is supported on Linux (x86_64/ARM64 with libaio) and macOS (kqueue)",
 )
 
 import zvec  # noqa: E402

src/binding/python/model/common/python_config.cc

@@ -242,4 +242,4 @@ void ZVecPyConfig::Initialize(pybind11::module_ &m) {
 }
 
 
-}  // namespace zvec
+}  // namespace zvec

src/binding/python/model/python_collection.cc

@@ -313,7 +313,16 @@ void ZVecPyCollection::bind_dql_methods(
           "given vector column. One of 'mmap', 'buffer_pool', 'contiguous'. "
           "Raises KeyError if no HNSW index exists on the column, or "
           "ValueError if the column's index is not an HNSW index. Intended "
-          "for introspection and testing only; not part of the stable API.");
+          "for introspection and testing only; not part of the stable API.")
+      .def(
+          "_debug_io_backend_type",
+          [](const Collection &self) {
+            const auto result = self.DebugGetIoBackendType();
+            return unwrap_expected(result);
+          },
+          "Debug-only: returns the I/O backend type used by DiskAnn. "
+          "One of 'libaio', 'pread'. Intended for introspection and "
+          "testing only; not part of the stable API.");
 }
 
 }  // namespace zvec

src/core/algorithm/CMakeLists.txt

@@ -17,7 +17,7 @@ else()
   # Empty stub library for unsupported platforms
   file(WRITE ${CMAKE_CURRENT_BINARY_DIR}/diskann_stub.cc
     "// Stub implementation for unsupported platforms\n"
-    "// DiskAnn only supports Linux x86_64\n"
+    "// DiskAnn supports Linux (x86_64/ARM64 with libaio) and macOS (kqueue)\n"
     "namespace zvec { namespace core { /* empty namespace for compatibility */ } }\n"
   )
 

src/core/algorithm/diskann/CMakeLists.txt

@@ -14,7 +14,7 @@ file(GLOB_RECURSE ALL_SRCS *.cc *.c)
 # ${CMAKE_DL_LIBS} for dlopen/dlsym/dlclose.
 set(CORE_KNN_DISKANN_LIBS core_framework core_knn_cluster)
 
-if(CMAKE_SYSTEM_NAME STREQUAL "Linux" AND CMAKE_SYSTEM_PROCESSOR MATCHES "x86_64|i686|i386")
+if(CMAKE_SYSTEM_NAME STREQUAL "Linux" AND CMAKE_SYSTEM_PROCESSOR MATCHES "x86_64|i686|i386|aarch64|arm64")
     list(APPEND CORE_KNN_DISKANN_LIBS ${CMAKE_DL_LIBS})
 endif()
 
@@ -31,4 +31,4 @@ cc_library(
     INCS . ${PROJECT_ROOT_DIR}/src/core ${PROJECT_ROOT_DIR}/src/core/algorithm
     LDFLAGS "${CORE_KNN_DISKANN_LDFLAGS}"
     VERSION "${PROXIMA_ZVEC_VERSION}"
-)
+)

src/core/algorithm/diskann/diskann_context.h

@@ -349,7 +349,13 @@ class DiskAnnContext : public IndexContext,
   uint32_t group_num_{0};
   std::map<std::string, TopkHeap> group_topk_heaps_{};
 
-  IOContext io_ctx_{0};
+  IOContext io_ctx_{
+#if defined(__APPLE__) || defined(__MACH__)
+      -1
+#else
+      0
+#endif
+  };
   SearchStats query_stats_;
 
   float *pq_table_dist_buffer_{nullptr};

src/core/algorithm/diskann/diskann_entity.h

@@ -72,17 +72,6 @@ struct DiskAnnMetaHeader {
     clear();
   }
 
-  DiskAnnMetaHeader(const DiskAnnMetaHeader &header) {
-    memcpy(this, &header, sizeof(header));
-  }
-
-  DiskAnnMetaHeader &operator=(const DiskAnnMetaHeader &header) {
-    if (this != &header) {
-      memcpy(this, &header, sizeof(header));
-    }
-    return *this;
-  }
-
   void reset() {
     doc_cnt = 0U;
   }
@@ -104,15 +93,6 @@ struct DiskAnnPqMeta {
     clear();
   }
 
-  DiskAnnPqMeta(const DiskAnnPqMeta &meta) {
-    memcpy(this, &meta, sizeof(meta));
-  }
-
-  DiskAnnPqMeta &operator=(const DiskAnnPqMeta &meta) {
-    memcpy(this, &meta, sizeof(meta));
-    return *this;
-  }
-
   void clear() {
     memset(this, 0, sizeof(DiskAnnPqMeta));
   }