codecov.yml

# Codecov configuration — info-only, never a CI gate.
#
# Background: PANIC_AUDIT_AND_LINT_HARDENING_PLAN.md Stage 2.5
# (audit-site coverage check) added cargo-llvm-cov tooling but the
# project's existing `TEST_COVERAGE_PLAN.md` / `TEST_COVERAGE_PLAN_2.md`
# both explicitly reject line / branch coverage percentage targets
# ("reward testing trivial getters at the expense of real
# invariants"). Codecov's default behavior is to fail PR status
# checks when coverage drops below a target; that contradicts the
# project stance. The settings below make every status check
# `informational: true` so the dashboard and PR comments still
# surface the data but the check never blocks a merge.

coverage:
  # Round to whole percent in the dashboard. Sub-percent precision
  # is noise — what matters is the per-PR delta visible in the
  # commit comment.
  round: nearest
  precision: 0

  status:
    # Project-wide coverage delta on the PR. `target: auto` compares
    # against the base commit; `informational: true` posts the
    # number as a comment / status, never as a required check.
    project:
      default:
        target: auto
        threshold: 5%
        informational: true
        only_pulls: false

    # Patch coverage = coverage of just the lines this PR touched.
    # Useful signal ("did the change land with tests?") but also
    # informational so it doesn't gate.
    patch:
      default:
        target: auto
        threshold: 5%
        informational: true
        only_pulls: true

# Show one PR comment with the project + patch summary. The
# `behavior: default` setting edits the existing comment instead
# of posting a new one per push.
comment:
  layout: "reach, diff, flags, files"
  behavior: default
  require_changes: false
  require_base: false
  require_head: false

# Skip generated / vendored / external-binding code. Coverage of
# FFI shim crates doesn't reflect substrate quality (the language
# bindings get their own dedicated CI tests; double-counting them
# here just inflates the headline number).
#
# `net/crates/net/src/ffi/**` (the SUBSTRATE-side FFI module, not
# the bindings) is also ignored: every public function there is
# `pub unsafe extern "C" fn` and the real exercise comes from the
# Go / Node / Python binding test suites calling into the compiled
# cdylib. The substrate's own Rust unit tests only cover the parts
# that don't need a live FFI runtime on the other side, so on
# paper the module sits ~43% covered — that's a measurement
# artifact, not a real gap. The actual FFI surface is exercised
# by `go test`, vitest, and pytest in the binding CI jobs.
ignore:
  - "net/crates/net/bindings/**"
  - "net/crates/net/src/ffi/**"
  - "net/crates/net/fuzz/**"
  - "net/crates/net/benches/**"
  - "net/crates/net/examples/**"
  - "net/crates/net/target/**"
  - "**/tests/**"
  - "go/**"
  - "web/**"
  - "crew/**"