Layer algebra methods should be allowed to fail if GEOS methods fail.

[Ari Jolma]

GEOS methods may fail due to topology problems at least. Currently the methods may crash due in those cases or continue (the error is probably emitted). Should SKIP_FAILURES be extended to this case or should we unconditionally fail?

[Even Rouault]

If SKIP_FAILURES is defined, it makes sense to skip failures due to GEOS errors. And in any case, crashes should be avoided.

[Ari Jolma]

Hm. Additional complication: GEOS methods, which return booleans (e.g. intersects()). Do they throw topology exceptions and if yes can we catch those?

[Even Rouault]

intersects() doesn't seem to always reports topology exceptions as show below, but anyway you can check with CPLGetLastErrorType() afterwards (and by clearing with CPLErrorReset() before)

>>> from osgeo import gdal, ogr
>>> g = ogr.CreateGeometryFromWkt('POLYGON((0 0,1 1,0 1,1 0,0 0))')
>>> g.Intersects(g)
True
>>> g = ogr.CreateGeometryFromWkt('POLYGON((0 0))')
>>> g.Intersects(g)
ERROR 1: IllegalArgumentException: point array must contain 0 or >1 elements

ERROR 1: IllegalArgumentException: point array must contain 0 or >1 elements

False
>>> gdal.GetLastErrorType()
3

[Ari Jolma]

Fixed in r33751. More extensive tests could be useful.