gdal_polygonize may generate invalid polygons according to ST_IsValid()

[rouault]

Cf "POLYGON ((1 4,1 3,0 3,0 1,1 1,1 0,3 0,3 1,4 1,4 3,3 3,3 4,1 4),(1 3,3 3,3 1,1 1,1 3))" from https://github.com/OSGeo/gdal/pull/7344/commits/82dfa7ad434f7548362f7d3b8629ae05cdf64b6b. This may be specific to 8-connectedness. The issue is not specific to PR #7344

[rouault]

Actually this is likely related only to 8-connectedness, and this is a bit by design. It is impossible to have a single Polygon that uses 8-connectnedness that is valid according to ST_IsValid(). Either we should just document that, or possibly add an option to generate valid MultiPolygon

[jratike80]

Maybe we should not output invalid polygons, or at least give users an easy option for making them valid. Possible solutions:

• Add an option --make-valid to the polygonize program https://gdal.org/en/stable/programs/gdal_raster_polygonize.html
• Document how to make the result valid with a pipeline (needs developing/testing so that the pipeline really works) gdal raster pipeline --progress ! read input.tif ! polygonize -c ! make-valid ! 8conn_valid.json

gdal raster polygonize --of GeoJSON -c -i P3412A.tif -o 8conn.json --progress    
0...10...20...30...40...50...60...70...80...90...100 - done in 00:01:05.

ST_IsValid finds 27390 invalid features from the output file. Errors can be corrected with "make-valid"

gdal vector make-valid --of GeoJSON -i 8conn.json -o 8conn-valid.json --progress
0...10...20...30...40...50...60...70...80...90...100 - done in 00:01:36.

For some reason this pipeline command begins but then just burns the CPU at 0..

gdal pipeline --progress ! read P3412A.tif ! polygonize -c ! write 8conn.json

EDIT: It is slow because the pipeline writes a temporary GeoPackage. Perhaps polygons are inserted one-by-one each as a new transaction?

CC @dbaston