From the cold read of #679 (and the six stale maps headers of #668, which is how this was noticed). internal/goldens compares build output against the committed goldens for examples, pointers, block, blockhome, messages, stream, blobs and scalars only; testdata/golden/tables/maps, arms, lists and wide are pinned but nothing compares them, which is how six maps headers went stale on main across two merges. Extend the goldens test to every directory under testdata/golden/tables, derived from the directory list rather than a hand-kept set, so a stale pin is red on the PR that made it stale. Related: #676 (the map-walk gate), #674 (the list-walk set derived from the IR).
From the cold read of #679 (and the six stale maps headers of #668, which is how this was noticed). internal/goldens compares build output against the committed goldens for examples, pointers, block, blockhome, messages, stream, blobs and scalars only; testdata/golden/tables/maps, arms, lists and wide are pinned but nothing compares them, which is how six maps headers went stale on main across two merges. Extend the goldens test to every directory under testdata/golden/tables, derived from the directory list rather than a hand-kept set, so a stale pin is red on the PR that made it stale. Related: #676 (the map-walk gate), #674 (the list-walk set derived from the IR).