Merge PR #470: Windows build support for runanywhere-commons (rebase of #383)

Adds first-class Windows (MSVC/VS 2022) build support: POSIX compat shim,
CMake presets, bundled zlib, ONNX Runtime Windows fetch, .bat build
scripts, plus cross-platform bug fixes surfaced during review.

Original author: @peilinok (via sanchitmonga22 revival)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: sanchitmonga22

sdk/runanywhere-commons/.gitattributes

@@ -0,0 +1,9 @@
+# Windows batch files MUST use CRLF line endings. cmd.exe's parser can
+# misbehave at 512-byte boundaries with LF-only line endings (label parsing
+# breaks, labels with delayed expansion fail intermittently). Keep these
+# files as CRLF even on non-Windows checkouts.
+*.bat text eol=crlf
+*.cmd text eol=crlf
+
+# Shell scripts must stay LF.
+*.sh text eol=lf

sdk/runanywhere-commons/CMakeLists.txt

@@ -46,6 +46,10 @@ option(RAC_BUILD_SERVER "Build OpenAI-compatible HTTP server (runanywhere-server
 # C++ CONFIGURATION
 # =============================================================================
 
+# C++20 is required because MSVC does not implement C++17 designated initializers
+# (e.g. `struct S s = {.field = value};`) — the codebase uses them in several
+# places (service vtable registration, config structs). All supported compilers
+# (GCC 10+, Clang 11+, MSVC 19.28+ / VS 2019 16.8+) support C++20.
 set(CMAKE_CXX_STANDARD 20)
 set(CMAKE_CXX_STANDARD_REQUIRED ON)
 set(CMAKE_CXX_EXTENSIONS OFF)
@@ -90,6 +94,9 @@ elseif(APPLE)
 elseif(UNIX)
     set(RAC_PLATFORM_LINUX TRUE)
     set(RAC_PLATFORM_NAME "Linux")
+elseif(WIN32)
+    set(RAC_PLATFORM_WINDOWS TRUE)
+    set(RAC_PLATFORM_NAME "Windows")
 else()
     set(RAC_PLATFORM_NAME "Unknown")
 endif()
@@ -125,6 +132,57 @@ if(NOT DEFINED LIBARCHIVE_VERSION)
     set(LIBARCHIVE_VERSION "3.8.1")
 endif()
 
+# -----------------------------------------------------------------------------
+# Zlib: Bundle from source when system zlib is not available.
+# Windows (MSVC) and some cross-compilation targets don't ship zlib.
+# We try system first; if not found, build from source so libarchive gets it.
+# -----------------------------------------------------------------------------
+find_package(ZLIB QUIET)
+if(NOT ZLIB_FOUND)
+    message(STATUS "System zlib not found — bundling from source...")
+    FetchContent_Declare(
+        zlib_src
+        GIT_REPOSITORY https://github.com/madler/zlib.git
+        GIT_TAG        v1.3.1
+        GIT_SHALLOW    TRUE
+    )
+    # Prevent zlib from installing or building examples
+    set(ZLIB_BUILD_EXAMPLES OFF CACHE BOOL "" FORCE)
+    set(SKIP_INSTALL_ALL ON CACHE BOOL "" FORCE)
+    set(SKIP_INSTALL_LIBRARIES ON CACHE BOOL "" FORCE)
+    set(SKIP_INSTALL_HEADERS ON CACHE BOOL "" FORCE)
+    set(SKIP_INSTALL_FILES ON CACHE BOOL "" FORCE)
+    # Save and restore BUILD_SHARED_LIBS
+    set(_SAVED_BSL_ZLIB ${BUILD_SHARED_LIBS})
+    set(BUILD_SHARED_LIBS OFF)
+    FetchContent_MakeAvailable(zlib_src)
+    set(BUILD_SHARED_LIBS ${_SAVED_BSL_ZLIB})
+
+    # Set cache variables so libarchive's find_package(ZLIB) picks up our build.
+    # For multi-config generators (Visual Studio), we point to the static lib
+    # output. The actual path is resolved at build time via target_link_libraries.
+    set(ZLIB_INCLUDE_DIR "${zlib_src_SOURCE_DIR};${zlib_src_BINARY_DIR}" CACHE PATH "" FORCE)
+    set(ZLIB_INCLUDE_DIRS "${zlib_src_SOURCE_DIR};${zlib_src_BINARY_DIR}" CACHE PATH "" FORCE)
+    # Use the CMake target name directly - this works because we link manually below
+    set(ZLIB_LIBRARY zlibstatic CACHE STRING "" FORCE)
+    set(ZLIB_LIBRARIES zlibstatic CACHE STRING "" FORCE)
+    set(ZLIB_FOUND TRUE CACHE BOOL "" FORCE)
+    # Create an imported ZLIB::ZLIB target that points to our zlibstatic
+    if(NOT TARGET ZLIB::ZLIB)
+        add_library(ZLIB::ZLIB ALIAS zlibstatic)
+    endif()
+    # On MSVC, set zlibstatic output to a known location so the linker can find it
+    if(MSVC AND TARGET zlibstatic)
+        set_target_properties(zlibstatic PROPERTIES
+            ARCHIVE_OUTPUT_DIRECTORY_RELEASE "${CMAKE_BINARY_DIR}/lib"
+            ARCHIVE_OUTPUT_DIRECTORY_DEBUG "${CMAKE_BINARY_DIR}/lib"
+        )
+    endif()
+    message(STATUS "Bundled zlib ready (v1.3.1)")
+else()
+    message(STATUS "Using system zlib: ${ZLIB_LIBRARIES}")
+endif()
+
 # -----------------------------------------------------------------------------
 # BZip2: Bundle from source for cross-compilation targets
 # Android NDK and Emscripten don't ship libbz2. macOS/iOS have it in the SDK.
@@ -156,6 +214,13 @@ if(NOT BZIP2_FOUND)
     set(BZIP2_INCLUDE_DIR "${bzip2_src_SOURCE_DIR}" CACHE PATH "" FORCE)
     set(BZIP2_LIBRARIES bz2_bundled CACHE STRING "" FORCE)
     set(BZIP2_FOUND TRUE CACHE BOOL "" FORCE)
+    # On MSVC, set output to a known location so the linker can find it
+    if(MSVC AND TARGET bz2_bundled)
+        set_target_properties(bz2_bundled PROPERTIES
+            ARCHIVE_OUTPUT_DIRECTORY_RELEASE "${CMAKE_BINARY_DIR}/lib"
+            ARCHIVE_OUTPUT_DIRECTORY_DEBUG "${CMAKE_BINARY_DIR}/lib"
+        )
+    endif()
     message(STATUS "Bundled BZip2 ready (v1.0.8)")
 else()
     message(STATUS "Using system BZip2: ${BZIP2_LIBRARIES}")
@@ -182,7 +247,16 @@ set(ENABLE_LIBXML2 OFF CACHE BOOL "" FORCE)
 set(ENABLE_EXPAT OFF CACHE BOOL "" FORCE)
 set(ENABLE_PCREPOSIX OFF CACHE BOOL "" FORCE)
 set(ENABLE_PCRE2POSIX OFF CACHE BOOL "" FORCE)
-set(ENABLE_LIBGCC OFF CACHE BOOL "" FORCE)
+# libarchive's ENABLE_LIBGCC gates a `find_library(GCC_LIBRARY gcc)` / `-lgcc`
+# link. Needed on MSVC for the bundled runtime; a no-op on Apple/Clang (no
+# libgcc); on Linux+GCC keep it OFF so we don't add an explicit -lgcc to the
+# link line (it's pulled in implicitly and adding it can cause duplicate-symbol
+# warnings in static configurations).
+if(MSVC)
+    set(ENABLE_LIBGCC ON CACHE BOOL "" FORCE)
+else()
+    set(ENABLE_LIBGCC OFF CACHE BOOL "" FORCE)
+endif()
 set(ENABLE_CNG OFF CACHE BOOL "" FORCE)
 set(ENABLE_TAR OFF CACHE BOOL "" FORCE)      # Don't build bsdtar binary
 set(ENABLE_CPIO OFF CACHE BOOL "" FORCE)     # Don't build bsdcpio binary
@@ -203,6 +277,32 @@ if(TARGET bz2_bundled AND TARGET archive_static)
     target_link_libraries(archive_static bz2_bundled)
 endif()
 
+# Link bundled zlib to libarchive if we built it from source
+if(TARGET zlibstatic AND TARGET archive_static)
+    target_link_libraries(archive_static zlibstatic)
+    target_include_directories(archive_static PRIVATE ${zlib_src_SOURCE_DIR} ${zlib_src_BINARY_DIR})
+endif()
+
+# On MSVC multi-config generators, libarchive references "zlibstatic.lib" and
+# "bz2_bundled.lib" by bare filename. The linker can't find them because they
+# are built into <config>/ subdirectories. Fix: redirect their output to a
+# single known directory and add it to the global linker search path.
+if(MSVC)
+    set(RAC_BUNDLED_LIB_DIR "${CMAKE_BINARY_DIR}/_bundled_libs")
+    foreach(_cfg Release Debug RelWithDebInfo MinSizeRel)
+        string(TOUPPER "${_cfg}" _CFG)
+        if(TARGET zlibstatic)
+            set_target_properties(zlibstatic PROPERTIES
+                ARCHIVE_OUTPUT_DIRECTORY_${_CFG} "${RAC_BUNDLED_LIB_DIR}")
+        endif()
+        if(TARGET bz2_bundled)
+            set_target_properties(bz2_bundled PROPERTIES
+                ARCHIVE_OUTPUT_DIRECTORY_${_CFG} "${RAC_BUNDLED_LIB_DIR}")
+        endif()
+    endforeach()
+    link_directories("${RAC_BUNDLED_LIB_DIR}")
+endif()
+
 message(STATUS "libarchive ready (v${LIBARCHIVE_VERSION})")
 
 # =============================================================================
@@ -234,6 +334,22 @@ if(CMAKE_CXX_COMPILER_ID MATCHES "Clang|GNU")
         add_compile_options(-fvisibility=hidden)
         add_compile_options(-ffunction-sections -fdata-sections)
     endif()
+elseif(MSVC)
+    # /W3 = reasonable warning level, /utf-8 = source/execution charset
+    # /Zc:__cplusplus = report correct __cplusplus value
+    # /EHsc = standard C++ exception handling
+    add_compile_options(/W3 /utf-8 /Zc:__cplusplus /EHsc)
+    # Suppress common harmless warnings in third-party code
+    add_compile_options(/wd4244 /wd4267 /wd4996)
+    add_compile_definitions(_CRT_SECURE_NO_WARNINGS NOMINMAX WIN32_LEAN_AND_MEAN)
+    # Use generator expressions instead of CMAKE_BUILD_TYPE so Release flags
+    # apply correctly with multi-config generators (Visual Studio), where
+    # CMAKE_BUILD_TYPE is empty at configure time and the config is selected
+    # at build time via --config Release.
+    add_compile_options(
+        $<$<CONFIG:Release>:/O2>
+        $<$<CONFIG:Release>:/DNDEBUG>
+    )
 endif()
 
 # =============================================================================
@@ -352,8 +468,11 @@ set(RAC_FEATURES_SOURCES
     src/features/embeddings/embeddings_component.cpp
     # Voice Agent
     src/features/voice_agent/voice_agent.cpp
-    # Result memory management
-    src/features/result_free.cpp
+    # Result memory management (weak symbol fallbacks)
+    # On MSVC, each service .cpp already provides strong definitions,
+    # so result_free.cpp (which uses __attribute__((weak)) on GCC/Clang)
+    # must be excluded to avoid LNK2005 duplicate symbol errors.
+    $<$<NOT:$<CXX_COMPILER_ID:MSVC>>:src/features/result_free.cpp>
 )
 
 # Platform services (Apple Foundation Models + System TTS + CoreML Diffusion)
@@ -421,7 +540,20 @@ if(RAC_BUILD_SHARED)
 endif()
 
 # libarchive - native archive extraction
-target_link_libraries(rac_commons PRIVATE archive_static)
+# PUBLIC on Windows because MSVC static libs don't propagate transitive deps
+if(WIN32)
+    target_link_libraries(rac_commons PUBLIC archive_static)
+    # Tell libarchive headers we're using the static library (not DLL)
+    target_compile_definitions(rac_commons PRIVATE LIBARCHIVE_STATIC)
+    if(TARGET zlibstatic)
+        target_link_libraries(rac_commons PUBLIC zlibstatic)
+    endif()
+    if(TARGET bz2_bundled)
+        target_link_libraries(rac_commons PUBLIC bz2_bundled)
+    endif()
+else()
+    target_link_libraries(rac_commons PRIVATE archive_static)
+endif()
 target_include_directories(rac_commons PRIVATE ${libarchive_SOURCE_DIR}/libarchive ${libarchive_BINARY_DIR})
 
 # Platform-specific linking
@@ -441,6 +573,12 @@ if(RAC_PLATFORM_ANDROID)
     target_link_libraries(rac_commons PUBLIC log)
 endif()
 
+if(RAC_PLATFORM_WINDOWS)
+    target_compile_definitions(rac_commons PRIVATE RAC_PLATFORM_WINDOWS=1)
+    # ws2_32 = Winsock, shlwapi = path utilities
+    target_link_libraries(rac_commons PUBLIC ws2_32 shlwapi)
+endif()
+
 set_target_properties(rac_commons PROPERTIES
     CXX_STANDARD 20
     CXX_STANDARD_REQUIRED ON
@@ -588,6 +726,9 @@ message(STATUS "  Services:     LLM, STT, TTS, VAD interfaces")
 if(APPLE AND RAC_BUILD_PLATFORM)
     message(STATUS "  Platform:     Apple Foundation Models, System TTS")
 endif()
+if(RAC_PLATFORM_WINDOWS)
+    message(STATUS "  Platform:     Windows (MSVC)")
+endif()
 if(RAC_BUILD_BACKENDS)
     message(STATUS "  Backends:     LlamaCPP=${RAC_BACKEND_LLAMACPP}, ONNX=${RAC_BACKEND_ONNX}, WhisperCPP=${RAC_BACKEND_WHISPERCPP}, WhisperKitCoreML=${RAC_BACKEND_WHISPERKIT_COREML}")
     if(RAC_BACKEND_RAG)

sdk/runanywhere-commons/CMakePresets.json

@@ -27,6 +27,42 @@
         "CMAKE_BUILD_TYPE": "Debug",
         "CMAKE_EXPORT_COMPILE_COMMANDS": "ON"
       }
+    },
+    {
+      "name": "windows-msvc",
+      "displayName": "Windows MSVC (all backends)",
+      "description": "Windows build with MSVC, all backends enabled",
+      "binaryDir": "${sourceDir}/build/windows",
+      "generator": "Visual Studio 17 2022",
+      "architecture": {
+        "value": "x64",
+        "strategy": "set"
+      },
+      "cacheVariables": {
+        "RAC_BUILD_BACKENDS": "ON",
+        "RAC_BACKEND_LLAMACPP": "ON",
+        "RAC_BACKEND_ONNX": "ON",
+        "RAC_BACKEND_RAG": "OFF",
+        "RAC_BACKEND_WHISPERCPP": "OFF",
+        "RAC_BUILD_PLATFORM": "OFF",
+        "RAC_BUILD_TESTS": "ON"
+      }
+    },
+    {
+      "name": "windows-msvc-core",
+      "displayName": "Windows MSVC (core only)",
+      "description": "Windows build with MSVC, core library only (no backends). Build config is selected at build time via --config Release (Visual Studio generators are multi-config and ignore CMAKE_BUILD_TYPE).",
+      "binaryDir": "${sourceDir}/build/windows-core",
+      "generator": "Visual Studio 17 2022",
+      "architecture": {
+        "value": "x64",
+        "strategy": "set"
+      },
+      "cacheVariables": {
+        "RAC_BUILD_BACKENDS": "OFF",
+        "RAC_BUILD_PLATFORM": "OFF",
+        "RAC_BUILD_TESTS": "ON"
+      }
     }
   ]
 }

sdk/runanywhere-commons/VERSIONS

@@ -49,6 +49,9 @@ ONNX_VERSION_MACOS=1.23.2
 # Linux version (can use latest)
 ONNX_VERSION_LINUX=1.23.2
 
+# Windows version (can use latest)
+ONNX_VERSION_WINDOWS=1.23.2
+
 # =============================================================================
 # Sherpa-ONNX (via runanywhere-core)
 # =============================================================================
@@ -69,6 +72,9 @@ SHERPA_ONNX_VERSION_MACOS=1.12.18
 # Linux version
 SHERPA_ONNX_VERSION_LINUX=1.12.23
 
+# Windows version
+SHERPA_ONNX_VERSION_WINDOWS=1.12.23
+
 # =============================================================================
 # llama.cpp (LLM inference)
 # =============================================================================

sdk/runanywhere-commons/cmake/FetchONNXRuntime.cmake

@@ -265,6 +265,39 @@ elseif(UNIX)
 
     message(STATUS "ONNX Runtime Linux library: ${onnxruntime_SOURCE_DIR}/lib/libonnxruntime.so")
 
+elseif(WIN32)
+    # Windows: Download Windows binaries
+    if(NOT DEFINED ONNX_VERSION_WINDOWS OR "${ONNX_VERSION_WINDOWS}" STREQUAL "")
+        message(FATAL_ERROR "ONNX_VERSION_WINDOWS not defined in VERSIONS file")
+    endif()
+
+    if(CMAKE_SIZEOF_VOID_P EQUAL 8)
+        set(ONNX_URL "https://github.com/microsoft/onnxruntime/releases/download/v${ONNX_VERSION_WINDOWS}/onnxruntime-win-x64-${ONNX_VERSION_WINDOWS}.zip")
+    else()
+        set(ONNX_URL "https://github.com/microsoft/onnxruntime/releases/download/v${ONNX_VERSION_WINDOWS}/onnxruntime-win-x86-${ONNX_VERSION_WINDOWS}.zip")
+    endif()
+
+    FetchContent_Declare(
+        onnxruntime
+        URL ${ONNX_URL}
+        DOWNLOAD_EXTRACT_TIMESTAMP TRUE
+    )
+
+    FetchContent_MakeAvailable(onnxruntime)
+
+    add_library(onnxruntime SHARED IMPORTED GLOBAL)
+
+    set_target_properties(onnxruntime PROPERTIES
+        IMPORTED_IMPLIB "${onnxruntime_SOURCE_DIR}/lib/onnxruntime.lib"
+        IMPORTED_LOCATION "${onnxruntime_SOURCE_DIR}/lib/onnxruntime.dll"
+    )
+
+    target_include_directories(onnxruntime INTERFACE
+        "${onnxruntime_SOURCE_DIR}/include"
+    )
+
+    message(STATUS "ONNX Runtime Windows library: ${onnxruntime_SOURCE_DIR}/lib/onnxruntime.lib")
+
 else()
     message(FATAL_ERROR "Unsupported platform for ONNX Runtime")
 endif()

sdk/runanywhere-commons/cmake/LoadVersions.cmake

@@ -56,10 +56,13 @@ message(STATUS "    ONNX_VERSION_IOS: ${RAC_ONNX_VERSION_IOS}")
 message(STATUS "    ONNX_VERSION_ANDROID: ${RAC_ONNX_VERSION_ANDROID}")
 message(STATUS "    ONNX_VERSION_MACOS: ${RAC_ONNX_VERSION_MACOS}")
 message(STATUS "    ONNX_VERSION_LINUX: ${RAC_ONNX_VERSION_LINUX}")
+message(STATUS "    ONNX_VERSION_WINDOWS: ${RAC_ONNX_VERSION_WINDOWS}")
 message(STATUS "  Sherpa-ONNX:")
 message(STATUS "    SHERPA_ONNX_VERSION_IOS: ${RAC_SHERPA_ONNX_VERSION_IOS}")
 message(STATUS "    SHERPA_ONNX_VERSION_ANDROID: ${RAC_SHERPA_ONNX_VERSION_ANDROID}")
 message(STATUS "    SHERPA_ONNX_VERSION_MACOS: ${RAC_SHERPA_ONNX_VERSION_MACOS}")
+message(STATUS "    SHERPA_ONNX_VERSION_LINUX: ${RAC_SHERPA_ONNX_VERSION_LINUX}")
+message(STATUS "    SHERPA_ONNX_VERSION_WINDOWS: ${RAC_SHERPA_ONNX_VERSION_WINDOWS}")
 message(STATUS "  Other:")
 message(STATUS "    LLAMACPP_VERSION: ${RAC_LLAMACPP_VERSION}")
 message(STATUS "    NLOHMANN_JSON_VERSION: ${RAC_NLOHMANN_JSON_VERSION}")