pgsql2shp makes dangerous call to liblwgeom on POINT EMPTY

[strk]

It looks like dumping empty points is doing bad things from dumper:

$ pgsql2shp pg25 "select 'POINT EMPTY'::geometry"
Initializing... 
Done (postgis major version: 3).
Output shape: Point
Dumping: Xlwgeom_api.c [135] called with n=0 and npoints=0

I tested this in current trunk (r17559) but I suspect it happens in older versions too. BTW, that message was once an ERROR or fatal assertion, and only "recently" turned into a notice instead.

[pramsey]

This is interesting… there is no explicit handling of empty in the dumper.

If you ask for a LINESTRING EMPTY it will create for you a line string with zero vertices in the shape file. Similarly for a POLYGON EMPTY.

For a POINT EMPTY there is no structural analogue to create in a shape file. So we are left with two options:

• Treat POINT EMPTY as SHP_NULL type (which is also used for actual SQL NULLs)
• Treat _all_ empties as SHP_NULL type
• Treat POINT EMPTY as a special case as in WKB, filling in the coordinates with NAN (hmm).

The second case changes existing behaviour the least, so I'm inclined to it. The third has a certain symmetry with WKB, but then we also need the inverse behaviour to convert POINT(NAN NAN) in a shape file into an empty…

[pramsey]

In 17650:

Handle POINT EMPTY in shape loader/dumper
References #4437

[pramsey]

In 17651:

Handle POINT EMPTY in shape loader/dumper
References #4437

[pramsey]

In 17652:

Handle POINT EMPTY in shape loader/dumper
References #4437

[pramsey]

In 17653:

Handle POINT EMPTY in shape loader/dumper
Closes #4437