Add GEOSCoordSeq_set function that takes a list of coordinates

[olt]

Creating a GEOSCoordSequence requires calls to GEOSCoordSeq_setX and GEOSCoordSeq_setY for each point. There should be a GEOSCoordSeq_set function that takes a list of points.

Something like:

GEOSCoordSeq_set(const GEOSCoordSequence* s, double *val, unsigned int n)

[strk]

I don't understand that signature. Is it an array of pointers to doubles ? Does it save 1 call for point or is it meant to do 1 call per dimension for any sized CoordinateSequence ?

[hobu]

Attached is a patch that starts the attempt. I am unable to compile it because of the Lion ar/llvm bug ​https://svn.macports.org/ticket/30309

[olt]

updated the patch. compiles und works on Mac OS X Lion.

[olt]

Updated the patch. Added two tests and changed the signature to match GEOSCoordSeq_create. dim, n -> n, dim

[strk]

What happens if you specify a number of ordinate values which is bigger than the number of coordinates the sequence was created with ? What if it's smaller ? What's the use case to allow that ? Should this whole thing be a constructor instead (GEOSCoordSeq_createWithValues) ?

[hobu]

Sandro,

You're right, this should probably be changed to a constructor that takes in an array, dimension count, and point count.

We'll update the patch and provide it.

Howard

[robe]

punt

[strk]

@hobu, was this change found to not be worth the trouble ? Or did you just not have time ?

[hobu]

Seemed like sugar and not that interesting.

We were able to work around the function call overhead in a different way for Shapely, so our need for it went away.

[strk]

I confirm it's not that interesting, just tried quickly with a GEOSCoordSeq_createXYZ taking 3 arrays of doubles and using it from postgis. The result is slower than the current version !

So, closing

[strk]

For the record (if anyone wants to clone before it disappears): ​https://github.com/strk/libgeos/tree/svn-trunk-GEOSCoordSeq_createXYZ