makeValid generates unexpected output about geos HOT 8 CLOSED

So the result of MakeValid is the same for both the old/new overlay engines, which means the result isn't "wrong" per se, just that it's got a different way to turn an invalid input to a valid output.

This is the output, the system has decided that the way to reconcile the doubled-up bottom rings is to use the outer ring as the enclosing ring, and return the spare inner ring as a line in a collection. The MakeValid code is very ring-centric, so this kind of falls out in an expected way.

A "better" or at least different MakeValid might drop all the edges into a graph and then decide what is an "enclosed area" by doing a scan-line routine. This is probably what we're seeing when we look at the rendering of the geometry in QGIS. Is it the "right" way to interpret the invalid geometry? Maybe, at least insofar as it's the visualization that most/many user interface tools will provide. I think waiting to add a graph-based MakeValid might be better than trying to coerce the existing routine into being something it is not.

from geos.

With the GeometryFixer ported at 5156001, this will be resolved on next release. (You'll have to use GEOSMakeValidWithParams if you're on the CAPI, or change from MakeValid to GeometryFixer if you're on C++ API)

from geos.

Is it an area or a line? Make valid can only do so much for you, if you feed it something nonsensical, it might return nonsense.

from geos.

The OGC has a spec for simplicity, and ISO has a spec for both validity and simplicity, for LineStrings. Moreover, if you have a large number of geometries, you should expect that you should be able to call makeValid on them without large surprises.

from geos.

Is it an area or a line? Make valid can only do so much for you, if you feed it something nonsensical, it might return nonsense.

It is a multipolygon border, so the area enclosed is quite small.

from geos.

OK, so it's a narrow corridor... and perhaps the invalidity is at the "pole", aka the bottom, where it does this.

from geos.

This makes sense to me. It's not worth hacking the original routine with a special case. Is a graph-based MakeValid something that is on the backlog?

from geos.

Yes, a less hacked-together MakeValid is on the list of things, and will hopefully improve on the reliability of the current one a little. Unfortunately we might have these kind of behaviour changes, but I think as long as we return valid results, with some reliability that's all folks can reasonably ask.

from geos.