The full Windows leg on the VM stopped at the coverage-map contract and was
reported green. Two defects, one hiding the other:
- The VM verdict counted the completion marker anywhere in the log. The
isolated-venue contract's stubbed entry prints that marker, and a capacity
case prints "expected=0 passed=True", so a leg that stopped at a later
contract, with no suite run, read as complete. The marker now counts only as
the log's last step header, which is where scripts/test.sh prints it; the
verdict contract pins the observed shape (14 cases).
- Under the native python3 the Windows legs install, the test-impact
contracts failed: the coverage-map builder cannot start the leg's bash
fakes, the scripts wrote CRLF, and cmp (diffutils) is not installed there.
selection.py, coverage-merge.py and coverage-map.py now write LF on every
host, so their outputs are byte-identical across platforms; the two cmp
checks compare bytes in the shell; the builder's functional leg is skipped
under a native Windows python with the reason and what was tried (the map
is built only by the ubuntu publish job), while both pure-python parser
tests still run there.
Production was not affected by the CRLF: run-tests-parallel.sh strips CR from
the selection and the runner trims its tokens.
Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>
Preserve the current selector, frozen-input inventory, result codec,
coverage collection, runner filtering, and isolated Windows venue work.
Component-level regression proofs are recorded in the PR handover.
The live classified provider, public execution integration, artifact
admission, fresh oracle/shadow evidence, and full platform gates remain
incomplete. This checkpoint does not enable or claim verified narrowing.
Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>
Martin Vogel