[raster] Upgrade scripts fail with addbandarg does not exist and other things

[robe]

The new additions to raster make it impossible to upgrade from an earlier 2.1.0SVN or 2.0… to present.

Get addbandarg[] does not exist if you are coming from lower 2.1.0SVN.

These new ones need to be added to rtpostgis_upgrade_cleanup.sql.in.c

I think we can safely get rid of 8.4 logic since 9.0 is our minimal now. So I'm going to wipe that out too.

[dustymugs]

Added addbandarg and agg_samealignment TYPEs to rtpostgis_upgrade_cleanup.sql.in.c in r10157

[robe]

That seems to fix my issues, though I don't like the CASCADE since if someone happens to use these types in their own functions, there functions will be destroyed as well.

[dustymugs]

Neither do I. I'm writing some code to minimize the impact…

[dustymugs]

I've added some logic in r10161 to only DROP and/or CREATE if need be.

[dustymugs]

Ignore my last. r10162 should work better as it tries not to be as destructive.

[robe]

Is their a reason you chose to rename addbandarg instead of just dropping it?

The problem is if I'm coming from 2.0.1, I don't have an addbandarg to rename so it fails.

[dustymugs]

I tried to find a solution that removes the CASCADE. Renaming the datatype allows us to continue the upgrade while not touching user code that uses the datatype. Granted, I think the statement a few lines later attempting to drop renamed datatype will also fail if there is user code. Alright, I'll keep things stupidly simple.

[dustymugs]

Fixed in r10169.

[robe]

Still doesn't work I'm afraid. I think we'll have to put these in a loop to create in the DO like I did with the box3d raster cast. or go back to your CASCADE

The issue is you can't even check if those functions exist since they reference non-existent types. So it breaks in the DROP IF EXISTS.

I'm going to get rid of the 8.4 section since we won't be supporting 8.4 in 2.1. eventually it would be nice to just have this DO section where you create the type, so we don't have to state the CREATE twice but I think the way strk's perl script is written, it would wipe it out. So I'll deal with that some other day.

[robe]

fixed in r10170. Feel free to reopen if you are not satisfied with my change.

I tested on my 9.1 instance upgrading from 2.0.1 to current 2.1.0 both the old fashioned way and using extensions. It should work fine on 9.0 since DO is supported in 9.0. I also got rid of the 8.4 legacy code.