Coverage

LLVM source-based coverage (-C instrument-coverage) over the same suites CI runs. llvm-profdata/llvm-cov must match the LLVM that built the binary — the devcontainer ships llvm-profdata-22/llvm-cov-22; locally use whatever suffix matches your LLVM (drop the suffix if you only have one). Also needs lcov (brew install lcov / apt install lcov).

Full coverage (Rust unit + Python + flow + TCK)

# clean stale coverage data first
find . -name "*.profraw" -delete
rm -f cov.profdata codecov.txt codecov.txt.all

# build + Rust unit tests, instrumented.
# The link-arg is required on Linux/CI so the embedded RediSearch static libs
# link without duplicate-symbol errors; drop it with macOS's linker. RUSTDOCFLAGS
# instruments doctests too, matching CI.
COV_RUSTFLAGS="-C instrument-coverage -C link-arg=-Wl,--allow-multiple-definition"
RUSTFLAGS="$COV_RUSTFLAGS" RUSTDOCFLAGS="$COV_RUSTFLAGS" cargo build
RUSTFLAGS="$COV_RUSTFLAGS" RUSTDOCFLAGS="$COV_RUSTFLAGS" cargo test -p graph

# instrumented Python + flow + TCK suites (module already built above)
source venv/bin/activate 2>/dev/null || source /data/venv/bin/activate
pytest tests/test_e2e.py tests/test_functions.py tests/test_mvcc.py tests/test_concurrency.py -vv
./flow.sh
TCK_DONE=tck_done.txt pytest tests/tck/test_tck.py -s

# pass each unit-test binary as an extra -object so llvm-cov also credits
# coverage recorded by test-only code paths (CI does this; without it those
# lines under-report)
TEST_OBJECTS=""
for bin in $(RUSTFLAGS="$COV_RUSTFLAGS" cargo test -p graph --no-run 2>&1 \
    | grep -w Executable | grep -oE 'target/debug/deps/[^)]+'); do
  TEST_OBJECTS="$TEST_OBJECTS -object $bin"
done

# merge the .profraw files and export lcov (libfalkordb.dylib on macOS)
llvm-profdata-22 merge --sparse $(find . -name "*.profraw") -o cov.profdata
llvm-cov-22 export --format=lcov --instr-profile cov.profdata target/debug/libfalkordb.so $TEST_OBJECTS > codecov.txt.all
lcov --ignore-errors unused -r codecov.txt.all -o codecov.txt
llvm-cov-22 report --instr-profile cov.profdata target/debug/libfalkordb.so $TEST_OBJECTS   # console summary

Narrow the run when you don't need whole-project numbers:

• Rust unit only — stop after cargo test -p graph and skip straight to the merge/export step (fast local check).
• Quick — run everything except ./flow.sh (flow tests dominate the wall-clock).

Running outside a devcontainer

The versioned llvm-*-22 tools and venv at /data/venv live in the devcontainer. To reproduce the exact CI environment from a plain host, run the same sequence inside the container image:

docker build -t falkordb-dev -f .devcontainer/Dockerfile .
docker run --rm -v $(pwd):/workspace -w /workspace falkordb-dev bash -c "
  find . -name '*.profraw' -delete
  CF='-C instrument-coverage -C link-arg=-Wl,--allow-multiple-definition'
  RUSTFLAGS=\"\$CF\" RUSTDOCFLAGS=\"\$CF\" cargo build
  RUSTFLAGS=\"\$CF\" RUSTDOCFLAGS=\"\$CF\" cargo test -p graph
  source /data/venv/bin/activate
  pytest tests/test_e2e.py tests/test_functions.py tests/test_mvcc.py tests/test_concurrency.py -vv
  ./flow.sh
  TCK_DONE=tck_done.txt pytest tests/tck/test_tck.py -s
  TEST_OBJECTS=''
  for bin in \$(RUSTFLAGS=\"\$CF\" cargo test -p graph --no-run 2>&1 | grep -w Executable | grep -oE 'target/debug/deps/[^)]+'); do TEST_OBJECTS=\"\$TEST_OBJECTS -object \$bin\"; done
  llvm-profdata-22 merge --sparse \$(find . -name '*.profraw') -o cov.profdata
  llvm-cov-22 export --format=lcov --instr-profile cov.profdata target/debug/libfalkordb.so \$TEST_OBJECTS > codecov.txt.all
  lcov --ignore-errors unused -r codecov.txt.all -o codecov.txt
  llvm-cov-22 report --instr-profile cov.profdata target/debug/libfalkordb.so \$TEST_OBJECTS
"

Output & notes

• codecov.txt is the filtered lcov report (upload this to Codecov); codecov.txt.all is the raw pre-filter export; cov.profdata is the merged profile; *.profraw are per-process files safe to delete afterward.
• Browse locally with genhtml codecov.txt -o cov_html.
• Flow tests require the instrumented debug build to exist first.
• Outside the devcontainer, the first run is slow because it builds the image.

If coverage fails, report the errors clearly and help diagnose.