CompoundCurve with empties parses differently from WKB and WKT

[mloskot]

Related to ​WKT/WKB/GeoJSON output always valid input thread, this might seem like a fabricated example, so I'm not sure if could be considered as a bug. Here it is:

PostGIS successfully parses this WKB

010900000002000000010200000000000000010800000000000000

which is equivalent to this WKT

COMPOUNDCURVE(EMPTY,CIRCULARSTRING EMPTY)
COMPOUNDCURVE(LINESTRING EMPTY,CIRCULARSTRING EMPTY)

But fails, what is expected, to parse that WKT throwing:

ERROR:  incontinuous compound curve

In fact, all combinations of multiple empty components are rejected:

COMPOUNDCURVE (EMPTY, CIRCULARSTRING EMPTY)
COMPOUNDCURVE (LINESTRING EMPTY, CIRCULARSTRING EMPTY)
COMPOUNDCURVE (CIRCULARSTRING EMPTY, CIRCULARSTRING EMPTY)

Interestingly, mixing empty with non-empty:

COMPOUNDCURVE(EMPTY, CIRCULARSTRING(1 5,6 2,7 3))
COMPOUNDCURVE(LINESTRING EMPTY, CIRCULARSTRING(1 5,6 2,7 3))
COMPOUNDCURVE(CIRCULARSTRING EMPTY, CIRCULARSTRING(1 5,6 2,7 3))

results in different kind of error:

ERROR:  getPoint4d_p: point offset out of range

Shouldn't the last error be:

ERROR:  incontinuous compound curve

[pramsey]

We've always had different rules for WKB and WKT inputs (try, for example, an unclosed WKB polygon and WKT polygon, one gets int the other does not). So this is not really a huge deal, IMO (except insofar as our general policy is a bit of a huge deal, but we've lived with it for a long long time).

I'd like to examine the last case, if only to try and catch it more cleanly that it looks like it's being caught.

[mloskot]

Paul,

Thanks for the feedback, yet another brick to my understanding of PostGIS. Certainly, I agree with you, not a big deal and as I mentioned, it is a case from very far corner, I wasn't sure though, hence the ticket.

I admit that recently I've been stress-testing PostGIS vs my interpretation of OGC+SQL/MM.

[robe]

"not a big deal" is all I need to push this.

[pramsey]

Changed the code to catch the last error a little more validly and safely than the hail-mary "you almost walked off the end of the point array" catch. r13452

[mloskot]

The fix looks good to me, thanks Paul.