darkbio-wire 0.10.0

.PHONY: check coverage fuzz fuzz-loop fuzz-minimize fuzz-seeds generate vectors
.DEFAULT_GOAL := check

# The fuzz budget per target in seconds, the jobs to run it on and the
# sanitizer to build with. The sanitizer stays off by default, its runtime
# hangs before main on macOS.
FUZZ_TIME ?= 180
FUZZ_JOBS ?= $(shell nproc 2>/dev/null || sysctl -n hw.ncpu)
FUZZ_SANITIZER ?= none

# Environment of the fuzz feature builds. Every random draw goes through the
# seeded backend of the mocks, see src/transport/mock/random.rs, for reproducible
# vectors and deterministic fuzzing. A musl host also links dynamically, as
# libFuzzer's interceptors resolve the real libc functions with dlsym, which a
# static musl binary cannot do.
HOST_MUSL = $(findstring musl,$(shell rustc -vV | sed -n 's/^host: //p'))
FUZZ_ENV = RUSTFLAGS='--cfg getrandom_backend="custom"$(if $(HOST_MUSL), -C target-feature=-crt-static)'

# check runs the gates CI holds a push to, the formatting, clippy, the docs and
# the tests of every feature combination.
check:
	cargo fmt --all -- --check
	cargo clippy --all-features -- -D warnings
	cargo doc --all-features --no-deps
	cargo hack test --each-feature
	cargo test --all-features

# coverage measures the test coverage of the library code and opens the HTML
# report. It needs nightly to leave the test modules out of the numbers. The
# previous instrumented build is dropped first, as a test binary left behind
# by another toolchain would get merged into the report as uncovered code.
coverage:
	rm -rf target/llvm-cov-target
	cargo +nightly llvm-cov --html --open

# fuzz-seeds regenerates the fuzz seed corpora from the scenario tests, every
# script they run written out as the fuzzers read it. The fuzzers start from
# these alongside their own corpus. The seeder names the targets, so one left
# without seeds means a name drifted from the binary.
fuzz-seeds:
	rm -rf fuzz/seeds
	WIRE_SEEDS=$(CURDIR)/fuzz/seeds cargo test --features fuzz --quiet
	for target in $$(cargo +nightly fuzz list); do \
		test -d fuzz/seeds/$$target || { echo "no seeds for $$target"; exit 1; }; \
	done

# fuzz runs every fuzz target for the budget each on all cores, from its corpus
# and the seeds, stopping at the first finding.
fuzz:
	for target in $$(cargo +nightly fuzz list); do \
		mkdir -p fuzz/corpus/$$target; \
		$(FUZZ_ENV) cargo +nightly fuzz run -s $(FUZZ_SANITIZER) -j $(FUZZ_JOBS) $$target fuzz/corpus/$$target fuzz/seeds/$$target -- -max_total_time=$(FUZZ_TIME) || exit 1; \
	done

# fuzz-minimize uses set cover to retain coverage features with fewer inputs,
# keeping a corpus grown by fuzz runs small before it is committed.
fuzz-minimize:
	for target in $$(cargo +nightly fuzz list); do \
		$(FUZZ_ENV) cargo +nightly fuzz cmin -s $(FUZZ_SANITIZER) $$target -- -set_cover_merge=1 || exit 1; \
	done

# fuzz-loop runs the fuzz targets round robin until a finding stops it or the
# loop is interrupted, the corpus growing across rounds.
fuzz-loop:
	while true; do $(MAKE) --no-print-directory fuzz || exit 1; done

# vectors transcribes the client scenario tests for other client implementations
# to replay. src/transport/mock/vector.rs defines the recorded events. The build
# routes all randomness through getrandom's custom backend, which the recorder
# seeds per scenario, so the transcripts regenerate unchanged for a given
# version of the crypto. The flag rebuilds every crate, hence its own target
# directory.
vectors:
	rm -rf vectors/client
	$(FUZZ_ENV) CARGO_TARGET_DIR=target/vectors \
		WIRE_VECTORS=$(CURDIR)/vectors cargo test --quiet --features fuzz mock::server

# generate writes the protobuf bindings and the message conversions derived
# from the schema into src/protocol/generated, formatted like handwritten code.
# The generator vendors its own protoc, so consumers of the crate need neither
# protoc nor prost-build.
generate:
	cargo run --quiet --manifest-path generator/Cargo.toml
	rustfmt --edition 2024 src/protocol/generated/*.rs