st_union(geom[]) does not union single-geometrycollection arrays

[Regina Obe]

This seems non-intuitive to me too.

The ST_Union(ARRAY[..])   should be giving you the same results as ST_Union(set geom)
But as you stated in your second example, I confirm it is not.

Should be giving the same answer as

SELECT  ST_AsText(ST_Union(f.geom))
FROM (SELECT
        st_geomfromtext(
          'GEOMETRYCOLLECTION(
            POLYGON((0 0,10 0,25 25,0 10,0 0)),
            POLYGON((20 20,30 20,30 30,20 30,20 20))
          )') ) AS f(geom);

Which returns:   POLYGON((22 20,10 0,0 0,0 10,20 22,20 30,30 30,30 20,22 20))

Anyone have thoughts on this, if we should flag as a bug.  I would say it's a bug.

Thanks,
Regina

> -----Original Message-----
> From: Dian Fay <di at nmfay.com>
> Sent: Saturday, February 24, 2024 6:16 PM
> To: postgis-devel at lists.osgeo.org
> Subject: st_union(geom[]) does not union single-geometrycollection arrays
> 
> Postgres 16.1
> postgis_version: 3.4 USE_GEOS=1 USE_PROJ=1 USE_STATS=1
> 
> I've produced some geometrycollection arrays from st_clusterintersecting.
> Now I want to combine each cluster, to end up with as many simple
> geometries as there were originally clusters. This turns out to work well as
> long as I have multiple clusters. When the array contains only one
> geometrycollection -- in other words, _all_ the original geometries overlapped
> and should be combined -- it doesn't.
> Here's a distilled example showing a successful cluster simplification with two
> clusters:
> 
> ```
> select
>   st_asewkt(
>     st_union(
>       array[
>         st_geomfromtext(
>           'GEOMETRYCOLLECTION(
>             POLYGON((0 0,10 0,25 25,0 10,0 0)),
>             POLYGON((20 20,30 20,30 30,20 30,20 20))
>           )'
>         ),
>         st_geomfromtext(
>           'GEOMETRYCOLLECTION(
>             POLYGON((100 100, 110 100, 110 110, 100 100))
>           )'
>         )
>       ]
>     )
>   );
> ```
> 
> The polygons in the first collection overlap and are merged:
> 
> ```
> MULTIPOLYGON((
>   (10 0,0 0,0 10,20 22,20 30,30 30,30 20,22 20,10 0)),
>   ((110 100,100 100,110 110,110 100))
> )
> ```
> 
> However, remove the second geometrycollection:
> 
> ```
> select
>   st_asewkt(
>     st_union(
>       array[
>         st_geomfromtext(
>           'GEOMETRYCOLLECTION(
>             POLYGON((0 0,10 0,25 25,0 10,0 0)),
>             POLYGON((20 20,30 20,30 30,20 30,20 20))
>           )'
>         )
>       ]
>     )
>   );
> ```
> 
> And st_union doesn't touch it:
> 
> ```
> GEOMETRYCOLLECTION(
>   POLYGON((0 0,10 0,25 25,0 10,0 0)),
>   POLYGON((20 20,30 20,30 30,20 30,20 20))
> )
> ```
> 
> Since this does pull it out of the array, I can get the result I'm after by nesting
> `st_union(st_union(array[....]))`. But it's counterintuitive, at least to me, that
> `pgis_union_geometry_array` skips these. Should the "one geom, good geom"
> case really apply to geometrycollections or is it meant more for simple
> geometries?