postgis alpha3 extension fails to install -- documentation st_pixelaspolygon

[robe]

alpha3 extension doesn't install because the st_aspixelpolygon documentation is inconsistent with the function.

complains that no such function

st_pixelaspolygon(raster,integer,integer,integer) does not exist

Guys it would be really nice if someone could at least test the extension install before we release an alpha. Not that I'm not greatful, but I don't think this Wednesday thing is working out too well. At least not for me.

[pracine]

Should be fixed by Bborie in r8999.

[robe]

Pierre,

I seriously doubt it. alpha3 is r9003 (after Bborie's change). I discovered it when testing the package  http://www.postgis.org/download/windows/experimental.php

So I had to hand patch it by removing the ST_Aspixelaspolygon comment from the extension.

The extension picks up documentation / function inconsistencies because both get installed together as a single transaction whereas the regular installs the comments get installed as a separate file.

It still holds true that before we make a release we should test this. Right now extensions aren't tested by our regress system and its a little too late to discover them once we've declared an alpha. It was the same issue with alpha1.

I suppose we could have our regress install the comments as well to catch these consistencies.

[pramsey]

The point of regular alpha releases is to find this kind of thing, not to produce production software. This is important, it points to a process improvement needed, since we have new dependencies cropping up: extensions depend on comments and comments depend on the functions being there. So there needs to be a consistency test somewhere that all the comments are on functions that exist. Agreed?

[robe]

To me these simple checks should be done before we release an alpha for plain reason that I plan to build binaries to coincide with alpha, and it really screws up my tempo when extensions don't install since I can't really release the alpha.

Yes I believe I have a ticket for that already -- #1532

BTW: Can we have extensions build NOT require xsltproc? We distribute the postgis comments files with the tar balls, so there isn't a reason to require it. Besides if we do, I can't build extensions because I don't have xlstproc under mingw since it crashes a lot or really slows down my building.

[pramsey]

No cake and eating it too.

The original request was for extensions to auto build/install in the presence of 9.1, and for developers that means they'll be building extensions when working on svn, when they might not have the comments, which means they also have to get the comments built. I can remove the dependency and also remove the autobuild of extensions. Or I can leave the dependency in place.

[pramsey]

Or you could make extensions not depend on comments. Or only include comments if they happen to exist.

[dustymugs]

Fixed in r9117