c-cpp-coverage

name	c-cpp-coverage
description	Generate code coverage for C/C++ projects using gcov and lcov, and upload to Codecov. Use this skill when users need to add coverage to C or C++ projects, set up lcov pipelines, or troubleshoot gcov/lcov issues. Covers autotools, CMake, and Meson build systems.

C/C++ Coverage Generation Skill

This skill provides detailed guidance for generating code coverage from C/C++ projects using gcov/lcov and uploading it to Codecov. It encodes hard-won knowledge from real-world onboarding of C/C++ projects where the standard "just add --coverage" advice is insufficient.

When to Use This Skill

Use this skill when:

• The codecov-onboarding skill detects a C/C++ project
• A user asks about gcov, lcov, or C/C++ code coverage
• Coverage generation fails with gcov/lcov errors
• A user needs to add a coverage CI job to a C/C++ project
• A user encounters _FORTIFY_SOURCE, negative count, or format errors

Why C/C++ Coverage Needs Special Handling

Unlike Go (go test -coverprofile) or Python (pytest-cov), C/C++ coverage requires a multi-step pipeline:

1. Compile with --coverage flags (produces .gcno files)
2. Run tests (produces .gcda counter files)
3. Capture with lcov (reads .gcda → produces .info report)
4. Filter system headers and test code from the report
5. Upload the .info file to Codecov

Each step has gotchas. This skill documents them all.

Prerequisites

• GCC or Clang (for --coverage / gcov support)
• lcov (often not pre-installed in CI containers)
• A build system: autotools, CMake, or Meson
• An existing test suite (e.g., make check, ctest, meson test)

Instructions

Step 1: Detect the Build System

# Autotools (configure.ac / Makefile.am)
ls configure.ac Makefile.am 2>/dev/null

# CMake
ls CMakeLists.txt 2>/dev/null

# Meson
ls meson.build 2>/dev/null

# Plain Makefile
ls Makefile GNUmakefile 2>/dev/null

Also check how tests are run:

grep -E "^check|^test" Makefile 2>/dev/null
grep "add_test\|enable_testing" CMakeLists.txt 2>/dev/null
grep "test(" meson.build 2>/dev/null

Step 2: Configure Coverage Compiler Flags

The core flags are the same regardless of build system:

CFLAGS="-g -O0 --coverage"
CXXFLAGS="-g -O0 --coverage"
LDFLAGS="--coverage"

• -g — debug symbols (needed for source mapping)
• -O0 — no optimization (accurate line-level coverage)
• --coverage — equivalent to -fprofile-arcs -ftest-coverage

Workaround: _FORTIFY_SOURCE conflict

On Fedora, RHEL, and CentOS, the system's default CFLAGS include _FORTIFY_SOURCE=2, which requires at least -O1. Since coverage needs -O0, the compiler will error:

error: _FORTIFY_SOURCE requires compiling with optimization (-O1 or higher)

Fix: Explicitly undefine and redefine _FORTIFY_SOURCE to 0:

CFLAGS="-g -O0 --coverage -Wp,-U_FORTIFY_SOURCE,-D_FORTIFY_SOURCE=0"
CXXFLAGS="-g -O0 --coverage -Wp,-U_FORTIFY_SOURCE,-D_FORTIFY_SOURCE=0"

The -Wp,-U,-D syntax passes preprocessor directives through the compiler driver.

Workaround: -Werror with -O0

Many projects use --enable-gcc-warnings or -Werror in their default build. At -O0 some warnings that are suppressed at higher optimization levels become visible, causing the build to fail.

Fix: Pass --disable-gcc-warnings (autotools) or remove -Werror from the coverage build configuration.

Step 3: Apply Flags to the Build System

Autotools

./configure \
  CFLAGS="-g -O0 --coverage -Wp,-U_FORTIFY_SOURCE,-D_FORTIFY_SOURCE=0" \
  CXXFLAGS="-g -O0 --coverage -Wp,-U_FORTIFY_SOURCE,-D_FORTIFY_SOURCE=0" \
  LDFLAGS="--coverage" \
  --disable-gcc-warnings
make
make check

Important: If the project's ci/build.sh or configure wrapper assembles ./configure arguments from multiple sources, ensure that per-job overrides (like --disable-gcc-warnings) come last on the command line so they take precedence.

CMake

cmake -B build \
  -DCMAKE_C_FLAGS="-g -O0 --coverage -Wp,-U_FORTIFY_SOURCE,-D_FORTIFY_SOURCE=0" \
  -DCMAKE_CXX_FLAGS="-g -O0 --coverage -Wp,-U_FORTIFY_SOURCE,-D_FORTIFY_SOURCE=0" \
  -DCMAKE_EXE_LINKER_FLAGS="--coverage" \
  -DCMAKE_SHARED_LINKER_FLAGS="--coverage"
cmake --build build
cd build && ctest

Some CMake projects support a CMAKE_BUILD_TYPE=Coverage or have a -DENABLE_COVERAGE=ON option. Check the project's CMakeLists.txt for existing coverage support before adding flags manually.

Meson

meson setup builddir \
  -Db_coverage=true \
  -Dc_args="-Wp,-U_FORTIFY_SOURCE,-D_FORTIFY_SOURCE=0" \
  -Dcpp_args="-Wp,-U_FORTIFY_SOURCE,-D_FORTIFY_SOURCE=0" \
  -Dwerror=false
cd builddir
ninja
ninja test

Meson has built-in coverage support via -Db_coverage=true and can generate reports with ninja coverage.

Step 4: Generate the lcov Report

After tests run, .gcda files are scattered throughout the build tree. Use lcov to collect them into a single .info report:

lcov --capture \
  --directory . \
  --output-file coverage.info \
  --ignore-errors format,gcov,unsupported,negative

Then filter out system headers and test code:

lcov --remove coverage.info \
  '/usr/*' \
  '*/tests/*' \
  '*/test/*' \
  --output-file coverage.info \
  --ignore-errors format,negative

Verify the report:

lcov --list coverage.info --ignore-errors negative | tail -5

Why --ignore-errors is needed

• negative — When tests run in parallel (make -j N), gcov's internal counters can race, producing .gcda files with negative execution counts. lcov >= 2.0 treats these as hard errors by default. This is a known gcov limitation with parallel test execution.

• format — Stale .gcda files from a previous build or a different compiler version may have incompatible format.

• gcov — Some source files may fail gcov processing (e.g., generated code or assembly files).

• unsupported — Some gcov extensions used by newer GCC versions may not be recognized by the installed lcov version.

Step 5: Separate Unit and Integration Test Coverage (Optional)

If you want to upload unit tests and integration tests as separate Codecov flags, use lcov --zerocounters between test phases:

# Phase 1: Unit tests
lcov --zerocounters --directory .
make check TESTS="$UNIT_TESTS"
lcov --capture --directory . --output-file unit-tests.info \
  --ignore-errors format,gcov,unsupported,negative

# Phase 2: Integration tests
lcov --zerocounters --directory .
make check TESTS="$INTEGRATION_TESTS"
lcov --capture --directory . --output-file integration-tests.info \
  --ignore-errors format,gcov,unsupported,negative

For autotools projects, you can classify tests by file type:

• Compiled C binaries (test-* without .sh) → unit tests
• Shell scripts (test-*.sh) → integration tests

Step 6: Upload to Codecov

Read the Codecov instance configuration from codecov-config/CONFIG.md to determine the correct instance URL.

In GitHub Actions, prefer the codecov/codecov-action@v6 with use_oidc: true (see Step 9 template). OIDC eliminates the need for a token secret.

In GitLab CI, Tekton, or any CLI-based upload, a token is always required — the CLI does not support OIDC:

curl -Os https://cli.codecov.io/latest/linux/codecov
chmod +x codecov

# For app.codecov.io:
./codecov upload-process \
  --token "${CODECOV_TOKEN}" \
  --flag unit-tests \
  --file coverage.info

# For self-hosted instances, add --codecov-url:
./codecov upload-process \
  --codecov-url "${CODECOV_URL}" \
  --token "${CODECOV_TOKEN}" \
  --flag unit-tests \
  --file coverage.info

Step 7: Handle Heavy Test Suites

Some C/C++ projects have test suites that become impractical under coverage instrumentation. Coverage overhead amplifies I/O, CPU, and timing in ways that can cause tests to hang or time out.

Common problem categories

Category	Symptom	Example
VM-based tests	Extremely slow boot/I/O	Tests spawning QEMU/guestfish
Timing-sensitive	Assertions on wall-clock time fail	Timeout or rate-limit tests
Exponential backoff	Delays compound with overhead	Retry tests
FFI/plugin tests	Missing gcov symbols at load time	OCaml/Rust plugins linked without libgcov
Shell-heavy tests	Hang from repeated process spawning	Eval/shell plugin tests

Solution: Skip problematic tests

Create a script that replaces problematic test files with exit 77 stubs (automake's SKIP status) before running make check:

#!/bin/bash
# ci/skip-coverage-tests.sh
set -euo pipefail

TESTS_TO_SKIP=(
    tests/test-heavy-io.sh
    tests/test-timing-sensitive.sh
    # ... add more as needed
)

for test_file in "${TESTS_TO_SKIP[@]}"; do
    if [ -f "$test_file" ]; then
        echo "  SKIP: $test_file"
        cat > "$test_file" <<'SKIP'
#!/bin/bash
echo "Skipped: too slow under coverage instrumentation"
exit 77
SKIP
        chmod +x "$test_file"
    fi
done

Important: Use count=$((count + 1)) instead of ((count++)) in bash scripts with set -e, because ((0++)) returns exit code 1 when the value is 0, causing the script to abort.

Solution: Timeout for make check

Add a timeout to prevent any single test from blocking the CI job:

# Use -s KILL (not --signal=KILL) for BusyBox compatibility (Alpine)
timeout -s KILL 30m make check || true

The || true ensures the CI job continues to the coverage upload step even if some tests fail — partial coverage is better than none.

Step 8: GitLab CI Job Template

Here is a complete GitLab CI job for C/C++ coverage with all workarounds applied. Adapt to the specific project:

coverage:
  stage: builds
  timeout: 2h
  variables:
    CFLAGS: "-g -O0 --coverage -Wp,-U_FORTIFY_SOURCE,-D_FORTIFY_SOURCE=0"
    CXXFLAGS: "-g -O0 --coverage -Wp,-U_FORTIFY_SOURCE,-D_FORTIFY_SOURCE=0"
    LDFLAGS: "--coverage"
    CONFIGURE_OPTS: "--disable-gcc-warnings"
    BUILD_ONLY: "true"
  script:
    - ci/skip-coverage-tests.sh      # Skip problematic tests
    - ci/build.sh                     # Configure + compile (no tests)
    - ci/codecov.sh                   # Run tests + lcov + upload
  artifacts:
    paths:
      - "config.log"
      - "**/test-suite.log"
      - "unit-tests.info"
      - "integration-tests.info"
    when: always
    expire_in: 1 week
  rules:
    - if: '$CI_PIPELINE_SOURCE == "merge_request_event" && $CI_MERGE_REQUEST_TARGET_BRANCH_NAME == "master"'
    - if: '$CI_PIPELINE_SOURCE == "push" && $CI_COMMIT_BRANCH == "master"'
    - when: 

Key design decisions:

• BUILD_ONLY: "true" — The build script compiles without running tests. The codecov script handles test execution so it can separate unit and integration test phases with lcov --zerocounters between them.
• timeout: 2h — Coverage-instrumented builds and tests are significantly slower than normal.
• || true on test execution (inside codecov.sh) — Ensures coverage data is always uploaded, even if some tests fail.
• Artifacts include .info files — For debugging and manual inspection of coverage data.

Step 9: GitHub Actions Job Template

For C/C++ projects on GitHub:

coverage:
  runs-on: ubuntu-latest
  timeout-minutes: 120
  permissions:
    id-token: write  # Required for OIDC authentication with Codecov
  steps:
    - uses: actions/checkout@v4

    - name: Install lcov
      run: sudo apt-get install -y lcov

    - name: Build with coverage
      run: |
        ./configure \
          CFLAGS="-g -O0 --coverage" \
          CXXFLAGS="-g -O0 --coverage" \
          LDFLAGS="--coverage" \
          --disable-gcc-warnings
        make -j$(nproc)

    - name: Run tests
      run: |
        timeout 1800 make check || true

    - name: Generate coverage report
      run: |
        lcov --capture --directory . --output-file coverage.info \
          --ignore-errors format,gcov,unsupported,negative
        lcov --remove coverage.info '/usr/*' '*/tests/*' \
          --output-file coverage.info \
          --ignore-errors format,negative

    - name: Upload to Codecov
      uses: 
      
           
         
         
         
        
        
        

Step 10: ci/codecov.sh Template Script

A reusable script for generating lcov reports and uploading to Codecov. Includes all known workarounds:

#!/bin/bash
# ci/codecov.sh - Run tests, generate lcov reports, upload to Codecov.
#
# Expects the project to be already compiled with --coverage flags.
# Handles two test phases (unit + integration) with separate flags.

set -euo pipefail

MAKE="${MAKE:-make -j $(getconf _NPROCESSORS_ONLN)}"
LCOV_IGNORE="--ignore-errors format,gcov,unsupported,negative"
LCOV_IGNORE_FILTER="--ignore-errors format,negative"

# Install lcov if not present
if ! command -v lcov &>/dev/null; then
    echo "== Installing lcov =="
    if command -v dnf &>/dev/null; then
        dnf install -y lcov
    elif command -v apt-get &>/dev/null; then
        apt-get update && apt-get install -y lcov
    else
        echo "ERROR: Cannot install lcov — unknown package manager"
        exit 1
    fi
fi

capture_coverage() {
    local name="$1"
    lcov --capture --directory . --output-file "${name}.info" $LCOV_IGNORE
    lcov --remove "${name}.info" '/usr/*' '*/tests/*' \
         --output-file "${name}.info" $LCOV_IGNORE_FILTER
    echo "==  coverage summary =="
    lcov --list  --ignore-errors negative |  -3
}




lcov --zerocounters --directory .
 
 -s KILL 30m  check || 
capture_coverage 





curl -Os --connect-timeout 10 --max-time 120 \
  https://cli.codecov.io/latest/linux/codecov
 +x codecov
./codecov upload-process \
  --token  \
  --flag unit-tests \
  --file coverage.info

Note for internal GitLab: See codecov-config/CONFIG.md for the pip-based codecov-cli approach which is recommended for internal GitLab environments.

Troubleshooting

"error: _FORTIFY_SOURCE requires compiling with optimization"

Cause: Fedora/RHEL default CFLAGS set _FORTIFY_SOURCE=2 which needs -O1, but coverage requires -O0.

Fix: Add -Wp,-U_FORTIFY_SOURCE,-D_FORTIFY_SOURCE=0 to CFLAGS.

"geninfo: ERROR: Unexpected negative count '-NNN'"

Cause: Parallel test execution (make -j) races gcov counters.

Fix: Add --ignore-errors negative to all lcov commands.

"lcov: command not found"

Cause: lcov is not pre-installed in the CI container.

Fix: Install it: dnf install -y lcov (Fedora/RHEL) or apt-get install -y lcov (Debian/Ubuntu).

"timeout: unrecognized option: signal=KILL"

Cause: Alpine Linux uses BusyBox which has different timeout syntax. BusyBox does not support the GNU --signal=KILL long option.

Fix: Use -s KILL instead of --signal=KILL. The short option works on both GNU coreutils and BusyBox.

"undefined symbol: __gcov_merge_add" when loading plugins

Cause: A dynamically-loaded plugin (.so) was compiled without --coverage, or was compiled with a language toolchain (OCaml, Rust) that doesn't link against libgcov.

Fix: Skip tests that load such plugins by adding them to the skip list in ci/skip-coverage-tests.sh.

Tests hang indefinitely under coverage

Cause: Coverage instrumentation slows down I/O-intensive and timing-sensitive tests enough to cause hangs or infinite loops.

Fix:

1. Add a timeout -s KILL 30m to the make check command
2. Skip the heaviest tests via ci/skip-coverage-tests.sh
3. Use || true so hangs don't prevent coverage upload

Build fails with -Werror at -O0

Cause: Some warnings only appear at -O0 and are treated as errors when -Werror is active.

Fix: Pass --disable-gcc-warnings (autotools) or remove -Werror from the coverage build configuration.

if [ "$failed" ] is always true

Cause: In shell, [ "$failed" ] tests if the string is non-empty. The string "0" is non-empty, so the test is always true.

Fix: Use [ "$failed" != "0" ] for numeric comparison.

Reference

• gcov documentation: https://gcc.gnu.org/onlinedocs/gcc/Gcov.html
• lcov documentation: https://github.com/linux-test-project/lcov
• Codecov lcov support: https://docs.codecov.com/docs/supported-coverage-report-formats
• Codecov CLI: https://github.com/codecov/codecov-cli
• Codecov instance routing: see codecov-config/CONFIG.md
• General Codecov onboarding: see codecov-onboarding/SKILL.md

Summary

This skill covers the full C/C++ coverage pipeline:

1. Detecting the build system (autotools, CMake, Meson)
2. Configuring the right compiler/linker flags with workarounds
3. Running tests with timeout protection
4. Generating lcov reports with error tolerance
5. Optionally separating unit vs integration test coverage
6. Skipping problematic tests under coverage instrumentation
7. Uploading to the correct Codecov instance
8. Providing reusable CI job templates and scripts