Heap profiling + tcmalloc-style telemetry parity (Phases 2–11)

.bazelrc

@@ -8,4 +8,8 @@ test --test_output=streamed
 build:macos --macos_minimum_os=10.15
 build:macos --no@fuzztest//fuzztest:use_riegeli
 
+# Rust integration tests (rust_test) print to stderr; keep the output
+# from being suppressed so failures are diagnosable in CI.
+test --test_output=errors
+
 try-import %workspace%/fuzztest.bazelrc

.bazelversion

@@ -1 +1 @@
-8.2.1
+8.7.0

.github/workflows/bazel.yml

@@ -0,0 +1,57 @@
+name: Bazel build
+
+# Smoke-test that the Bazel target graph keeps working alongside the
+# Cargo build.  We exercise the rust_library variants and at least
+# one rust_test -- enough to catch the common regressions in the
+# dual-build layer.
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+
+jobs:
+  bazel:
+    name: bazel build + test
+    runs-on: ubuntu-latest
+
+    steps:
+      - uses: actions/checkout@v4
+
+      # The Bazel team officially recommends bazelisk on CI so a
+      # `.bazelversion` (or the MODULE.bazel) pins the toolchain
+      # rather than the system bazel.
+      - name: Install bazelisk
+        run: |
+          sudo curl -L -o /usr/local/bin/bazel \
+            https://github.com/bazelbuild/bazelisk/releases/latest/download/bazelisk-linux-amd64
+          sudo chmod +x /usr/local/bin/bazel
+          bazel --version
+
+      # Cache the Bazel disk cache so subsequent runs skip the
+      # rules_rust toolchain download (~150 MB) and the cmake
+      # action's output.  The cache key folds in MODULE.bazel.lock so
+      # any dependency bump invalidates the cache rather than
+      # silently reusing a stale repo set.
+      - name: Cache Bazel
+        uses: actions/cache@v4
+        with:
+          path: |
+            ~/.cache/bazel
+          key: bazel-${{ runner.os }}-${{ hashFiles('MODULE.bazel.lock', 'MODULE.bazel') }}-${{ github.sha }}
+          restore-keys: |
+            bazel-${{ runner.os }}-${{ hashFiles('MODULE.bazel.lock', 'MODULE.bazel') }}-
+            bazel-${{ runner.os }}-
+
+      - name: Bazel build :: snmalloc-rs Rust library (default)
+        run: bazel build //snmalloc-rs:snmalloc_rs
+
+      - name: Bazel build :: snmalloc-sys Rust library (default + profiling)
+        run: |
+          bazel build \
+            //snmalloc-rs/snmalloc-sys:snmalloc_sys \
+            //snmalloc-rs/snmalloc-sys:snmalloc_sys_profiling
+
+      - name: Bazel test :: snmalloc-rs integration tests
+        run: bazel test //snmalloc-rs:all

.github/workflows/main.yml

@@ -83,6 +83,18 @@ jobs:
             build-type: Release
             extra-cmake-flags: "-DSNMALLOC_TRACING=On"
             build-only: true
+          - os: "ubuntu-24.04"
+            variant: "Profile Build (gcc)"
+            build-type: Release
+            extra-cmake-flags: "-DSNMALLOC_PROFILE=ON"
+            build-only: true
+          - os: "ubuntu-24.04"
+            variant: "Profile Build (clang)"
+            build-type: Release
+            extra-cmake-flags: >-
+              -DCMAKE_CXX_COMPILER=clang++
+              -DSNMALLOC_PROFILE=ON
+            build-only: true
           - os: "ubuntu-22.04"
             variant: "clang libstdc++ (Build only)"
             build-type: Release
@@ -125,6 +137,33 @@ jobs:
             dependencies: "sudo apt install -y ninja-build libc++-dev"
             test-exclude-pattern: "memcpy|external_pointer"
             test-extra-args: "--repeat-until-fail 2"
+          # Profile + TSan: exercise the heap-profiling code paths
+          # (perf-profile_stress + func-profile_*) under ThreadSanitizer.
+          # Uses libc++ because TSan requires a TSan-instrumented C++
+          # runtime; libstdc++ is not instrumented on Ubuntu.  The
+          # `-R profile_` ctest filter restricts the run to profile
+          # tests so the sanitizer overhead stays within the CI budget.
+          - os: "ubuntu-24.04"
+            variant: "Profile + TSan (clang)"
+            build-type: "Debug"
+            extra-cmake-flags: >-
+              -DSNMALLOC_PROFILE=ON
+              -DSNMALLOC_SANITIZER=thread
+              -DCMAKE_CXX_COMPILER=clang++
+              -DCMAKE_CXX_FLAGS=-stdlib="libc++ -g"
+            dependencies: "sudo apt install -y ninja-build libc++-dev"
+            test-extra-args: "-R profile_"
+          # Profile + ASan: exercise the heap-profiling code paths
+          # under AddressSanitizer.  ASan is compatible with libstdc++,
+          # so no extra runtime dependency is needed beyond ninja.
+          - os: "ubuntu-24.04"
+            variant: "Profile + ASan (clang)"
+            build-type: "Debug"
+            extra-cmake-flags: >-
+              -DSNMALLOC_PROFILE=ON
+              -DSNMALLOC_SANITIZER=address
+              -DCMAKE_CXX_COMPILER=clang++
+            test-extra-args: "-R profile_"
     uses: ./.github/workflows/reusable-cmake-build.yml
     with:
       os: ${{matrix.os}}
@@ -190,6 +229,11 @@ jobs:
             build-type: Release
             extra-cmake-flags: "-DSNMALLOC_ENABLE_PAC=ON"
             variant: "PAC"
+          # Profile build with heap profiling support enabled
+          - os: "macos-15"
+            build-type: Release
+            extra-cmake-flags: "-DSNMALLOC_PROFILE=ON"
+            variant: "Profile Build (clang)"
     uses: ./.github/workflows/reusable-cmake-build.yml
     with:
       os: ${{matrix.os}}
@@ -472,6 +516,68 @@ jobs:
         cd ${{github.workspace}}/build
         ctest --parallel --output-on-failure
 
+  # ============================================================================
+  # Profile + PGO (clang) — two-stage profile-guided optimization build
+  #
+  # Runs scripts/run-pgo-build.sh end-to-end: stage 1 builds an
+  # instrumented snmalloc + func-profile_overhead-fast, executes it to
+  # populate .profraw data, merges via llvm-profdata, and stage 2
+  # rebuilds with -fprofile-use=<merged.profdata>. The use-stage
+  # libsnmallocshim-rust.a is uploaded as a release artifact so
+  # downstream consumers (snmalloc-rs and friends) can pick up the
+  # PGO-optimized static archive on every push to main.
+  #
+  # macOS is intentionally skipped — the matrix has limited macOS
+  # minutes and the AppleClang/Xcode profraw format is pinned per OS
+  # image, which would force re-merge across runner upgrades. Run
+  # scripts/run-pgo-build.sh locally on macOS.
+  #
+  # LLVM 19 matches the COMPILER_RT_LLVM_VERSION env at the top of
+  # this file and the coverage.yml job, so llvm-profdata's raw-profile
+  # format is consistent across CI legs.
+  # ============================================================================
+  pgo:
+    name: Profile + PGO (clang)
+    runs-on: ubuntu-24.04
+    timeout-minutes: 30
+    steps:
+    - uses: actions/checkout@v4
+    - name: Install clang-19 + llvm-19 + ninja
+      run: |
+        sudo apt-get update
+        sudo apt-get install -y ninja-build clang-19 llvm-19
+    - name: Run two-stage PGO build
+      env:
+        # Route stage artifacts to absolute paths under the runner
+        # workspace so the upload-artifact step below can find them
+        # regardless of where the script's repo_root resolves to.
+        CC: clang-19
+        CXX: clang++-19
+        PGO_STAGE1_DIR: ${{ github.workspace }}/build-pgo-gen
+        PGO_STAGE2_DIR: ${{ github.workspace }}/build-pgo-use
+        PGO_PROFILE_DATA_DIR: ${{ github.workspace }}/build-pgo-gen/pgo-data
+        PGO_PROFILE_FILE: ${{ github.workspace }}/build-pgo-gen/pgo.profdata
+        # SNMALLOC_RUST_SUPPORT=ON materializes libsnmallocshim-rust.a
+        # under the use-stage build directory; that file is the
+        # uploaded artifact below. Use CMake-provided clang names so
+        # the configure step does not fall back to system gcc.
+        PGO_EXTRA_CMAKE_FLAGS: >-
+          -G Ninja
+          -DSNMALLOC_RUST_SUPPORT=ON
+          -DCMAKE_C_COMPILER=clang-19
+          -DCMAKE_CXX_COMPILER=clang++-19
+      run: scripts/run-pgo-build.sh
+    - name: Verify PGO artifact
+      run: |
+        ls -l "${{ github.workspace }}/build-pgo-use/libsnmallocshim-rust.a"
+    - name: Upload PGO artifact (libsnmallocshim-rust.a)
+      uses: actions/upload-artifact@v4
+      with:
+        name: pgo-libsnmallocshim-rust-linux-x64
+        path: ${{ github.workspace }}/build-pgo-use/libsnmallocshim-rust.a
+        if-no-files-found: error
+        retention-days: 14
+
   # ============================================================================
   # vcpkg integration
   # ============================================================================
@@ -557,6 +663,7 @@ jobs:
       qemu-crossbuild,
       windows,
       format,
+      pgo,
       vcpkg-integration
     ]
     runs-on: ubuntu-24.04

.github/workflows/rust.yml

@@ -70,6 +70,50 @@ jobs:
     - name: Run tests
       run: cargo test ${{ matrix.release.flag }} --all ${{ matrix.features.args }}
 
+  # ============================================================================
+  # Heap-profiling feature build (Phase 7.5)
+  #
+  # Exercises the `profiling` cargo feature (which propagates
+  # SNMALLOC_PROFILE=ON to the C++ build via snmalloc-sys) on every push.
+  # Restricted to Linux + macOS because the profile code paths are validated
+  # there in the C++ matrix; Windows profile coverage can be added later if
+  # needed.
+  # ============================================================================
+  profiling:
+    runs-on: ${{ matrix.os }}
+    name: "profiling-${{ matrix.os }}-${{ matrix.release.name }}"
+    defaults:
+      run:
+        shell: bash
+        working-directory:
+          ./snmalloc-rs
+    strategy:
+      matrix:
+        os: [ubuntu-latest, macos-14, macos-15]
+        rust: [stable]
+        release:
+          - name: release
+            flag: "--release"
+          - name: debug
+            flag: ""
+      fail-fast: false
+    steps:
+    - uses: actions-rs/toolchain@v1
+      with:
+        toolchain: ${{ matrix.rust }}
+    - name: Checkout
+      uses: actions/checkout@v4
+    - name: update dependency
+      run: |
+        if bash -c 'uname -s | grep 'Linux' >/dev/null'; then
+          sudo apt-get update -y && sudo apt-get --reinstall install -y libc6-dev
+        fi
+      shell: bash
+    - name: Build (profiling)
+      run: cargo build ${{ matrix.release.flag }} --verbose --features profiling
+    - name: Run tests (profiling)
+      run: cargo test ${{ matrix.release.flag }} --all --features profiling
+
   publish-scan:
     runs-on: ubuntu-latest
     name: publish-scan

.gitignore

@@ -27,3 +27,8 @@
 
 # rust target
 /target
+
+# bazel convenience symlinks (created in the workspace root by `bazel
+# build` / `bazel test`).  The actual outputs live under the user's
+# bazel cache so the symlinks are pure noise on commit.
+/bazel-*

BUILD.bazel

@@ -8,6 +8,7 @@ filegroup(
             "src/test/*.h",
             "src/test/*.cc",
             "CMakeLists.txt",
+	    "cmake/**/*.cmake",
         ],
     ),
     visibility = ["//visibility:private"],
@@ -39,7 +40,7 @@ CMAKE_FLAGS = {
     "SNMALLOC_OPTIMISE_FOR_CURRENT_MACHINE": "ON",
     "SNMALLOC_USE_SELF_VENDORED_STL": "OFF",
     "SNMALLOC_IPO": "ON",
-    "USE_SNMALLOC_STATS": "ON",
+    "SNMALLOC_STATS": "ON",
     "SNMALLOC_BUILD_TESTING": "OFF",
 } | select({
     ":release_with_debug": {"CMAKE_BUILD_TYPE": "RelWithDebInfo"},
@@ -87,6 +88,36 @@ cmake(
     out_static_libs = [
         "libsnmallocshim-static.a",
         "libsnmalloc-new-override.a",
+        "libsnmallocshim-rust.a",
+    ],
+    postfix_script = "ninja",
+    visibility = ["//visibility:public"],
+)
+
+# Profile-enabled variant of the Rust shim archive.  Same source set as
+# `:snmalloc-rs` but with SNMALLOC_PROFILE=ON so the `sn_rust_profile_*`
+# exports in `rust.cc` switch from the no-op stubs to real bodies.  Used
+# by the `snmalloc_sys_profiling` Rust target.
+cmake(
+    name = "snmalloc-rs-profile",
+    cache_entries = CMAKE_FLAGS | {
+        "SNMALLOC_RUST_SUPPORT": "ON",
+        "SNMALLOC_PROFILE": "ON",
+    },
+    generate_args = ["-G Ninja"],
+    lib_source = ":srcs",
+    out_shared_libs = select({
+        "@bazel_tools//src/conditions:darwin": [
+            "libsnmallocshim-checks-memcpy-only.dylib",
+            "libsnmallocshim-checks.dylib",
+            "libsnmallocshim.dylib",
+        ],
+        "//conditions:default": [],
+    }),
+    out_static_libs = [
+        "libsnmallocshim-static.a",
+        "libsnmalloc-new-override.a",
+        "libsnmallocshim-rust.a",
     ],
     postfix_script = "ninja",
     visibility = ["//visibility:public"],