End-to-end testing#
The unit / integration suite runs against real PyQGIS in CI and exercises every algorithm's
initAlgorithm / parameter construction. What it does not cover is running an algorithm
end-to-end (processAlgorithm) in a real QGIS session on real data, nor the GUI / dialog behaviour.
That still has to be done by hand, and it is the gate before a non-experimental release.
This page is that by-hand validation guide. Install the plugin first — see Installation.
What to run — cover both wiring paths and all four groups#
The raster / conversion algorithms call the pyramids CLI in-process; the mesh / EO ones call the library directly. Run at least one of each path:
- [ ] Raster (CLI path): Reproject (warp) a GeoTIFF to another CRS; confirm the output layer loads and looks right.
- [ ] Raster: Clip (by mask and by extent), Merge, Raster calculator (e.g. NDVI over two inputs), Assign CRS / NoData.
- [ ] Conversion: Rasterize a vector, Polygonize a raster (GeoJSON output), Create COG, Convert format.
- [ ] Mesh (library path): Convert a
.2dm/ Selafin mesh to UGRID NetCDF, and Mesh → vector (needspyramids-mesh). - [ ] EO (library path): Contrast stretch, and Tasseled cap on a Landsat reflectance stack
(needs
pyramids-eo).
What to watch for — things CI cannot catch#
- [ ] PyQGIS parameter wiring — each algorithm's
parameterAs*calls, enum / attribute names, and the dialog rendering. (initAlgorithm/ parameter construction is covered automatically by the real-QGIS integration tests; the dialog rendering and actual execution are not.) - [ ] Forced-GeoJSON output — the provider pins vector outputs to
.geojson; confirm Polygonize / Mesh → vector produce loadable GeoJSON and the extension matches the content. - [ ] Dependency guards — with
pyramids-mesh/pyramids-eoabsent, the mesh / EO algorithms should raise a clear "install ..." message; the vendored-GDAL warning should appear only for a vendored build. - [ ] Error surfacing — a bad CRS / missing file should come back as a clean
QgsProcessingExceptionin the Processing log, not a raw traceback. - [ ] Qt5 / Qt6 — ideally smoke-test on a QGIS 3.x (Qt5) and a 4.x (Qt6) build, since the plugin
uses the
qgis.PyQtshim.
Done when#
- [ ] At least one algorithm per group runs successfully on real data and loads a correct output layer.
- [ ] The dependency-guard and error-surfacing behaviour is confirmed.
- [ ] Any bugs found are filed as their own issues (and fixed) before flipping
experimental=Falseinmetadata.txt.