MapColonies / shigola

48%

LAST BUILD ON BRANCH master

branch: SELECT

CHANGE BRANCH

x

Sync Branches

• No branch selected
• chore/ci-security-updates
• feat/remove-viewer
• fix/land-ci-dependency-updates
• master
• refs/tags/backup/pre-rebrand-world-crs84quad-tiles-pr-fixes
• test/cite-mvt-postgis
• test/coverage-baseline
• test/postgis-athens-fixture

Build # 57a221c73

Build Type

push

github

Committed by web-flow

Commit Message

test(cite): re-base the OGC conformance suite onto db-side MVT (MAPCO-11481) (#8)

* test(cite): re-base the conformance suite onto db-side mvt

The OGC CITE run was configured entirely against a GeoPackage fixture, which
made the project's only external conformance evidence an invisible dependency of
the GeoPackage provider: removing the provider would have deleted the evidence
and said nothing. The fixture now serves the same three Athens layers out of the
PostGIS fixture through ST_AsMVT, so nothing in the conformance path is
file-backed.

The workflow provisions the fixture with the same `docker compose up -d` plus
`docker wait migration` gate on_pr_push.yml uses, and builds with CGO_ENABLED=0
and -tags noGpkgProvider so that "conformance passes with no GeoPackage present"
is a property of the binary rather than of the config. Path triggers swap
provider/gpkg/** for provider/postgis/**, testdata/postgis/** and
docker-compose.yml; the OGC and tiling-scheme triggers are untouched.

Two things are deliberate and are argued in the files:

- The layer SQL passes !BBOX! to ST_AsMVTGeom unwrapped. That is exact for
  WorldCRS84Quad, whose tile grid is linear in lon/lat, and out by about half a
  unit in 4096 for WebMercatorQuad at zoom 14. The transform-to-3857 pattern the
  provider README documents is exact the other way round and degenerate near the
  poles. One SQL string cannot be exact for both, because nothing in it says
  which grid the request came from. Hence the narrow 13-15 zoom window: the map
  does not advertise tiles it would render visibly wrong.
- The WorldCRS84Quad tile moves from row 4742 to 4740. A WorldCRS84Quad tile is
  half the width and a bit over half the height of a WebMercatorQuad one, so
  4742 holds no places_points rows -- and ST_AsMVT emits no layer at all for a
  layer with no rows, which the suite reports as a clean pass. 4740 is the
  nearest tile in the same column holding all three layers.

Verified: both schemes rep... (continued)

Coverage Stats

9841 of 20391 relevant lines covered (48.26%)

203.85 hits per line