Yes, your interpretation is correct. ST_ReducePrecision is only guaranteed to work on VALID geometries. It will also work on *some* invalid geometries (such as ones which are invalid due to containing adjacent - but not overlapping - polygons). So the safest thing is to use ST_MakeValid if there is doubt about whether the inputs are valid.
On Tue, Jun 11, 2024 at 1:00 PM Marco Boeringa <ma...@boeringa.demon.nl> wrote: > Hi, > > Currently, the help page of PostGIS's ST_ReducePrecision command more or > less suggests that ST_ReducePrecision can be used as kind of "drop-in" > replacement of ST_MakeValid. The actual text reads: > > > > *"Returns a valid geometry with all points rounded to the provided grid > tolerance, and features below the tolerance removed. Unlike ST_SnapToGrid > the returned geometry will be valid, with no ring self-intersections or > collapsed components.* > > *Precision reduction can be used to: * > > - > > * match coordinate precision to the data accuracy * > - > > * reduce the number of coordinates needed to represent a geometry * > - > > * ensure valid geometry output to formats which use lower precision > (e.g. text formats such as WKT, GeoJSON or KML when the number of output > decimal places is limited). * > - > > * export valid geometry to systems which use lower or limited > precision (e.g. SDE, Oracle tolerance value)"* > > However, a first test replacing: > > ST_CollectionExtract(ST_MakeValid(<SOME OTHER GENERALIZATION CODE>),3) > > with: > > ST_ReducePrecision(<SOME OTHER GENERALIZATION CODE>),<GRID SIZE>) > > failed on some geometries with a processing error. > > Replacing the first try with: > > ST_ReducePrecision(ST_CollectionExtract(ST_MakeValid(<SOME OTHER > GENERALIZATION CODE>),3),<GRID SIZE>) > > did succeed. > > This suggest to me that the help page should actually read something like: > > > > *"Returns a geometry with all points rounded to the provided grid > tolerance, and features below the tolerance removed. Unlike ST_SnapToGrid > the returned geometry is guaranteed to be valid, with no ring > self-intersections or collapsed components, if the input geometry itself is > valid. If the input geometry is invalid, there is no guarantee the output > will be valid. Use ST_MakeValid to ensure the input is valid before running > this command if the input data may contain invalid geometries."* > > Is my interpretation of how ST_ReducePrecision handles validity as > reflected by this suggested updated help text correct? > > Marco >
