geometry regression tests not raster safe

[robe]

I'm still getting a lot of crashing when testing against VC 64-bit EDB builds for 9.0 and 9.1. I tried to run these outside of mingw to see if they crash by themselves, but I quickly ran into obstacles.

It seems a lot of our tests e.g. regress.sql

use constructs like ST_SRID('POINT(1 2)')

which if you have raster installed gives a st_srid(unknown) is not unique errors. I think we have quite a few of these that need to be changed to

ST_SRID('POINT(1 2)'::geometry)

Anyrate I have managed to crash my 9.0x64bit after some runs, outside of mingw environment so separate issue.

[pramsey]

Why doesn't this present on other platforms? I can run regression on raster/geometry without trouble.

[robe]

Because windows especially anything compiled with VC is special. proprietary compilers are evil .

Actually that's a separate issue. I think this ticket might be resolved since I think I got the last ST_SRID etc. offender.

It should be an issue on all platforms if we have code like the above in any of our geometry tests. Though it could be the way we run the tests now that we never have both raster and geometry installed while geometry tests are running which would be bad as we wouldn't be able to pick up bugs like the one we had with shp2pgsql/shp2pgsql-gui a while back that it was using lazy constructs that would never work on an install that has both raster and geometry installed. That's a separate issue though so I'm goner close this out.