You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The test targets Pass 5a sweeps, mirroring Pass 1a's surface: OCCTCurveTests, OCCTGeom2dTests, OCCTSurfaceTests, OCCTMathTests.
The question is different for a test phase
Passes 5a to 5d sweep tests, not source, so this audit asks the test-side form of the same
question, against the refman, for the geometry primitives lane:
Under-coverage. An OCCT behaviour the refman documents, which we wrap and document, and which no test exercises. Distinct from a missing test in general: the finding is a capability we
advertise on the strength of the refman's contract without ever having checked it holds.
Over-coverage. A test that asserts behaviour the refman does not promise. These are the
dangerous ones. They pass today, they encode an implementation detail of one kernel version as if
it were the contract, and they turn a legitimate upstream change into a false regression. CI never validates a kernel-patch PR against the patched kernel it claims to fix #585 is
the shape: seven suites were red against a kernel that was simply older than the tests, and
nothing distinguished that from a real break.
Method
Read the refman through the context MCP (occt-refman, pinned to Package.swift's OCCT
version), never from recall. Land one committed executable artifact under Scripts/repro/<this issue>/ printing the lane's table, per the census rule in docs/v2.0.0-plan.md.
Relationship to the test-validity phase
Adjacent to #766 and deliberately not the same question. #766 asks whether a test can fail;
this asks whether the thing it asserts is something OCCT actually promises. A test can pass both
checks, fail both, or pass either one alone, and #766's own evidence is that passing one has been
treated as evidence for the other before.
Done when
The artifact is committed and re-runnable.
Every refman-documented behaviour in the lane is marked tested / untested / asserts-more-than-promised.
Each untested is filed or covered.
Each asserts-more-than-promised is either re-anchored on the documented contract, or annotated
in place with why the stricter assertion is wanted.
Part of #807. Post-v2.0.0. Runs after #389.
Lane
The test targets Pass 5a sweeps, mirroring Pass 1a's surface:
OCCTCurveTests,OCCTGeom2dTests,OCCTSurfaceTests,OCCTMathTests.The question is different for a test phase
Passes 5a to 5d sweep tests, not source, so this audit asks the test-side form of the same
question, against the refman, for the geometry primitives lane:
no test exercises. Distinct from a missing test in general: the finding is a capability we
advertise on the strength of the refman's contract without ever having checked it holds.
dangerous ones. They pass today, they encode an implementation detail of one kernel version as if
it were the contract, and they turn a legitimate upstream change into a false regression. CI never validates a kernel-patch PR against the patched kernel it claims to fix #585 is
the shape: seven suites were red against a kernel that was simply older than the tests, and
nothing distinguished that from a real break.
Method
Read the refman through the
contextMCP (occt-refman, pinned toPackage.swift's OCCTversion), never from recall. Land one committed executable artifact under
Scripts/repro/<this issue>/printing the lane's table, per the census rule indocs/v2.0.0-plan.md.Relationship to the test-validity phase
Adjacent to #766 and deliberately not the same question. #766 asks whether a test can fail;
this asks whether the thing it asserts is something OCCT actually promises. A test can pass both
checks, fail both, or pass either one alone, and #766's own evidence is that passing one has been
treated as evidence for the other before.
Done when
tested/untested/asserts-more-than-promised.untestedis filed or covered.asserts-more-than-promisedis either re-anchored on the documented contract, or annotatedin place with why the stricter assertion is wanted.