[postgis-devel] PostGIS-on-GSERIALIZED

[Paragon Corporation]

Paul,
I enabled GSERIALIZED and submitted two more bug-tickets that you should fix
before I can even think about making an alpha test build.

Also on my list as a blocker is the ST_AddPoint.  Last I checked this still
did not work, but didn't check with the very last couple of builds.

So do not pass GO until you fix these:

http://trac.osgeo.org/postgis/query?status=assigned&status=new&status=reopen
ed&order=priority&priority=blocker&col=id&col=summary&col=status&col=owner&c
ol=type&col=priority&col=milestone&milestone=PostGIS+2.0.0

I'll probably have some more before end of day, but the |>> killed my tests
very early on.


Thanks,
Regina


--- ORIGINAL MESSAGE --
Well, I have now completed an alpha integration of PostGIS with
GSERIALIZED storage. To try it, edit ./postgis/gserialized.h and
change #undef to #define. Then make clean and make, and install into a
fresh database.

It currently fails 4 regression tests, but I think those are
"legitimate" failures. Several are memsize() tests, and the memory
size of these objects happen to be different from the old ones.
Another few are differences in 'box2d' representations, because I
changed the backing type for 'box2d' from BOX2DFLOAT4 to GBOX, so the
storage is now double and not float. So they are just more precise.
The final one is an obscure oddity in ST_Dump, which expects that the
dump of a collection of empty things will have zero entries. And
instead it has one entry for every non-collection element in the
collection.

There is a default index implementation, but it's currently n-d, not 2-d.

Anyhow, it builds! It regresses! When do we pull off the band-aid and
all start working on making this production ready?

Paul