[raster] ST_Union only considering first band if uniontype specified

[robe]

I'm not sure if this was just missed or too hard to do. If difficult, I guess we just need to amend the docs. Anyrate if I explicitly say 'LAST' vs. leaving it out I should get the same answer.

It seems ST_Union is still only returning first band if you specify a union type. Observe:

— this returns bright beautiful picture —

(WITH mset AS 
( SELECT ROW_NUMBER() OVER() As rnum, rast
  FROM (SELECT  ST_Tile(rast, (ST_Width(rast))::integer/2, ST_Height(rast)/2) As rast 
FROM (SELECT ST_Resize(rast,0.10,0.10) As rast
FROM pics 
WHERE filename='Mona_Lisa.jpg' LIMIT 1) As  foo
) As foofoo )
SELECT ST_Union(ST_SetUpperLeft(rast,ST_UpperLeftX(rast)*-1,ST_UpperLeftY(rast)*-1))
FROM mset where rnum IN(1,2))

— this returns grey scale — verified it ST_NumBands only returns 1 —-

SELECT ST_Union(ST_SetUpperLeft(rast,ST_UpperLeftX(rast)*-1,ST_UpperLeftY(rast)*-1), 'LAST')
FROM mset where rnum IN(1,2))

See attached images.

Using same image I described here from wikipedia

​http://www.bostongis.com/blog/index.php?/archives/207-Waiting-for-PostGIS-2.1-ST_Tile.html

[Bborie Park]

Gah! I'm on it.

[Bborie Park]

Looking at the docs, I evidently stated that "ST_Union(rast) now unions all bands of all input rasters". I believe I intended ST_Union(rast) unioning all bands to only apply to variant #1.

[robe]

Ah I see you wrote ST_Union(rast). I guess I wasn't looking closely. Maybe we can make it clearer. I'll just change this to a documentation enhancement and fix later, since its technically not wrong just could be clearer.

[robe]

Okay I reworded at r11081 just so I don't get confused

​http://postgis.net/docs/manual-dev/RT_ST_Union.html

I'll put in as an enhancement if not ticketed already

[robe]

I reworded yet again at r11082 as I realized when writing this long diatribe and trying to explain this oddball that it's only

ST_Union(rast,uniontype) that is problematic.

Also documented that the unionarg variant is new in 2.1.0.