Adding support for disabling exceptions

Adding a mode of error handling that does not rely on C++ exceptions,
with Abseil-like Status and StatusOr return types (but not adding a
dependency on Abseil). The old exceptions-based API is still available,
and the new functions are suffixed with "NoExceptions".

Refactoring the CMake build system. Organizing examples and tests and
reducing repetition. Adding the HNSWLIB_ENABLE_EXCEPTIONS option (ON by
default). Exceptions are also enabled when building Python bindings, and
in a few examples/tests where the code currently relies on exception
handling, even if HNSWLIB_ENABLE_EXCEPTIONS is set to OFF.

Adding ctest support. The only tests not supported by ctest are
test_updates invoked with and without the "update" argument, because
that requires installing numpy and running a data generation script and
doing that from a CMake test might be controversial. Those tests are
still run as part of GitHub Actions CI.

Enabling sanitizer support using ENABLE_ASAN / ENABLE_UBSAN /
ENABLE_TSAN / ENABLE_MSAN CMake options. Adding a combined ASAN/UBSAN
build type to the C++ test matrix in GitHub Actions in non-Windows
builds. Also adding Clang compiler on Ubuntu to the build matrix.

Making sure that assertions are enabled during tests. Using the
RelWithDebInfo build type and customizing it to remove the NDEBUG flag.
Also removing the default /EHsc flag when building with MSVC and instead
enabling exceptions when needed.

Fixing race conditions when accessing random number generators and
distributions from multiple threads in multiThreadLoad_test. These race
conditions were caught using ASAN/UBSAN.

Author: michaelbautin

.github/workflows/build.yml

@@ -3,18 +3,18 @@ name: HNSW CI
 on: [push, pull_request]
 
 jobs:
-  test_python:
-    runs-on: ${{ matrix.os }}
+  python:
+    runs-on: ${{ matrix.os }}-latest
     strategy:
       matrix:
-        os: [ubuntu-latest, windows-latest, macos-latest]
-        python-version: ["3.9", "3.10", "3.11", "3.12", "3.13"]
+        os: [ubuntu, windows, macos]
+        python-version: ["3.9", "3.10", "3.11", "3.12", "3.13.2"]
     steps:
       - uses: actions/checkout@v4
       - uses: actions/setup-python@v5
         with:
           python-version: ${{ matrix.python-version }}
-          
+
       - name: Build and install
         run: python -m pip install .
 
@@ -36,11 +36,34 @@ jobs:
           python -m unittest discover -v --start-directory tests/python --pattern "bindings_test*.py"
         shell: bash
 
-  test_cpp:
-    runs-on: ${{ matrix.os }}
+  cpp:
+    runs-on: ${{ matrix.os }}-latest
     strategy:
       matrix:
-        os: [ubuntu-latest, windows-latest, macos-latest]
+        os: [ubuntu, windows, macos]
+        sanitizer: [no_sanitizers, asan_ubsan]
+        exceptions: [no_exceptions, with_exceptions]
+        compiler: [clang, gcc, msvc]
+        exclude:
+          # No sanitizers on Windows
+          - os: windows
+            sanitizer: asan_ubsan
+          # No sanitizers on macOS -- might not be well supported on arm64
+          - os: macos
+            sanitizer: asan_ubsan
+          # No clang or gcc on Windows
+          - os: windows
+            compiler: clang
+          - os: windows
+            compiler: gcc
+          # No MSVC on Unix
+          - os: ubuntu
+            compiler: msvc
+          - os: macos
+            compiler: msvc
+          # No GCC on macOS
+          - os: macos
+            compiler: gcc
     steps:
       - uses: actions/checkout@v4
       - uses: actions/setup-python@v5
@@ -51,11 +74,48 @@ jobs:
         run: |
           mkdir build
           cd build
-          cmake ..
-          if [ "$RUNNER_OS" == "Windows" ]; then
-            cmake --build ./ --config Release
+          cmake_cmd=( cmake -S .. -B . )
+          if [[ "${RUNNER_OS}" != "Windows" ]]; then
+            c_compiler="${{matrix.compiler}}"
+            if [[ "${c_compiler}" == "gcc" ]]; then
+              cxx_compiler=g++
+            elif [[ "${c_compiler}" == "clang" ]]; then
+              cxx_compiler=clang++
+            else
+              echo "Invalid compiler ${c_compiler} for OS ${RUNNER_OS}" >&2
+              exit 1
+            fi
+            cmake_cmd+=( 
+              -DCMAKE_BUILD_TYPE=RelWithDebInfo
+              -DCMAKE_C_COMPILER=${c_compiler}
+              -DCMAKE_CXX_COMPILER=${cxx_compiler}
+            )
+          fi
+          if [[ "${{ matrix.sanitizer }}" == "asan_ubsan" ]]; then
+            cmake_cmd+=( -DENABLE_ASAN=ON -DENABLE_UBSAN=ON )
+          fi
+          if [[ "${{ matrix.exceptions }}" == "with_exceptions" ]]; then
+            cmake_cmd+=( -DHNSWLIB_ENABLE_EXCEPTIONS=ON )
+          else
+            cmake_cmd+=( -DHNSWLIB_ENABLE_EXCEPTIONS=OFF )
+          fi
+          if [[ "${RUNNER_OS}" != "Windows" ]]; then
+            cmake_cmd+=( -G Ninja )
+          fi
+          "${cmake_cmd[@]}"
+
+          # This is essential for debugging the the build, e.g. tracking down
+          # if exceptions are enabled or NDEBUG is specified when it should not
+          # be.
+          echo
+          echo "Contents of CMakeCache.txt:"
+          echo
+          cat CMakeCache.txt
+          echo
+          if [[ "${RUNNER_OS}" == "Windows" ]]; then
+            cmake --build ./ --config RelWithDebInfo --verbose
           else
-            make
+            ninja -v
           fi
         shell: bash
 
@@ -67,26 +127,16 @@ jobs:
         shell: bash
 
       - name: Test
-        timeout-minutes: 15
+        # Without sanitizers, 15 minutes might be sufficient.
+        timeout-minutes: 30
         run: |
           cd build
-          if [ "$RUNNER_OS" == "Windows" ]; then
-            cp ./Release/* ./
+          if [[ "${RUNNER_OS}" == "Windows" ]]; then
+            cp "./RelWithDebInfo/"* ./
           fi
-          ./example_search
-          ./example_filter
-          ./example_replace_deleted
-          ./example_mt_search
-          ./example_mt_filter
-          ./example_mt_replace_deleted
-          ./example_multivector_search
-          ./example_epsilon_search
-          ./searchKnnCloserFirst_test
-          ./searchKnnWithFilter_test
-          ./multiThreadLoad_test
-          ./multiThread_replace_test
+          ctest --build-config "RelWithDebInfo"
+          # These tests could be ctest-enabled in CMakeLists.txt, but that
+          # requires auto-generating test data and installing numpy for that.
           ./test_updates
           ./test_updates update
-          ./multivector_search_test
-          ./epsilon_search_test
         shell: bash

.gitignore

@@ -11,3 +11,5 @@ var/
 .vs/
 **.DS_Store
 *.pyc
+venv/
+tests/cpp/data/