[gdal-dev] RFC 39

[Even Rouault]

Selon Ari Jolma <ari.jolma at gmail.com>:

> Even,
>
> I've uploaded a version, which fixes these. I added a note that GEOS is
> required, as these probably segfault without it (the methods return
> NULLs and not geometries).

Hi,

Hum, I didn't think at crashes. I thought that the algorithm would just nicely
fail. But yes indeed you are right, reviewing the code, there would be NULL
derefs, so I'd suggest adding a preliminary call to
OGRGeometryFactory::haveGEOS() to exit early if it returns FALSE. This will save
crashes to people running the test suite on a build without GEOS support.

I'm also wondering if there would not be cases where the GEOS routines could
return NULL geometries. Looking at the implementation of GEOSIntersection_r()
(and other methods) in http://svn.osgeo.org/geos/trunk/capi/geos_ts_c.cpp , in
case an exception is thrown in GEOS C++ code, it will be caught by the GEOS C
API and returned as a NULL pointer to the caller. So I suspect that if invalid
geometries are processed, we could get a NULL geometry, and it might be
appropriate to handle it gracefully.

Looking quickly at the patch, I think it would be for the occurrences of
OGRGeometry *x_geom_diff_new (2 places) in OGRLayer::Union(), same in
::Identity(), in ::Update() and ::Clip().

Best regards,

Even