Aggregation operations in petascope don't work on multiband covs

[Piero Campalani]

Petascope seems to provide wrong (and different at every request) results to operations applied on multiband coverages, whether the operation is applied on a single band or not.

Simple examples on the rgb test coverage:

1. AVG on 1 RGB cell (should return the cell itself?):

   for m in (rgb) return encode(
       trim(m,
           {x:"CRS:1"(200:200),y:"CRS:1"(200:200)}),
       "csv") 
   
   ==> {"242 27 103"}
   
   for m in (rgb) return encode(
       avg(
           trim(m,
               {x:"CRS:1"(200:200),y:"CRS:1"(200:200)})
       ), "csv") 
   
   ==> {"0 8.70018e-313 7.19645e-262"}
   

2. AVG on 1 RGB cell, only the RED band (should return the single value itself):

   for m in (rgb) return encode(
       trim(m,
           {x:"CRS:1"(200:200),y:"CRS:1"(200:200)}),
       "csv") 
   
   ==> {242}
   
   for m in (rgb) return encode(
       avg(
           trim(m,
               {x:"CRS:1"(200:200),y:"CRS:1"(200:200)})
       ), "csv") 
   
   ==> {0}
   

3. ADD over a multiple bands:

   for m in (rgb) return encode(
       trim(m,
           {x:"CRS:1"(200:201),y:"CRS:1"(200:201)}),
       "csv") 
   
   ==> {"242 27 103","242 27 103"}
   
   for m in (rgb) return encode(
       add(
           (float)trim(m,
               {x:"CRS:1"(200:200),y:"CRS:1"(200:201)})
       ), "csv") 
   
   ==> {"7.26573e-262 3.61544e-313 8.49162e-314"}
   ==> {"484 7563062 16"} without (float) cast
   ==> {"484 7563062 0"}  without (float) cast
   

4. ADD over a single band:

   for m in (rgb) return encode(
       trim(m.red,
           {x:"CRS:1"(200:201),y:"CRS:1"(200:201)}),
       "csv") 
   
   ==> {242,242}
   
   for m in (rgb) return encode(
       add(
           (float)trim(m.red,
               {x:"CRS:1"(200:200),y:"CRS:1"(200:201)})
       ), "csv") 
   
   ==> {8.10648e+16}
   

. Multiplication of multi-band results (but maybe this is not implemented at all):

for m in (rgb) return encode(
    (float)trim(m,{x:"CRS:1"(200:200),y:"CRS:1"(200:200)}) *
    (float)trim(m,{x:"CRS:1"(200:200),y:"CRS:1"(200:200)}), "csv") 

==> Error: Could not evaluate rasdaman query: 
'select csv((float)(((m[200:200,200:200])*((float)(m[200:200,200:200]))))) from rgb AS m'
Cause: Error evaluating rasdaman query: 'select csv((float)(((m[200:200,200:200])*((float)(m[200:200,200:200]))))) from rgb AS m

[Dimitar Misev]

1. I get "Error: Exception: Base type typeName: typeID: 7 typeSize: 16 struct { typeName: RAS_DOUBLE typeID: 15 typeSize: 8 red, typeName: RAS_DOUBLE typeID: 15 typeSize: 8 green } as ElementType is not supported yet." for

   for m in (rgb) return
       avg(m)
   

2. I get 242.0

   for m in (rgb) return
       avg(
           trim(m.red,
               {x:"CRS:1"(200:200),y:"CRS:1"(200:200)})
       )
   

3. Similarly gives "Error: Exception: Base type typeName: typeID: 7 typeSize: 16 struct { typeName: RAS_DOUBLE typeID: 15 typeSize: 8 red, typeName: RAS_DOUBLE typeID: 15 typeSize: 8 green } as ElementType is not supported yet."

   for m in (rgb) return
       add(
           (float)trim(m,
               {x:"CRS:1"(200:200),y:"CRS:1"(200:201)})
       )
   

4. I'm sure if you don't use the encode it will be fine as well like 2.
5. If you remove the casts it will work

   for m in (rgb) return encode(
       trim(m,{x:"CRS:1"(200:200),y:"CRS:1"(200:200)}) *
       trim(m,{x:"CRS:1"(200:200),y:"CRS:1"(200:200)}), "csv") 
   

[Dimitar Misev]

So in conclusion the real bug seems to be that aggregation operations over composite types do not work in petascope, but they work just fine in rasql.

[Dimitar Misev]

Wait, why is this blocker? You can certainly work around it.

[Piero Campalani]

Yes, the problem seems to reside indeed in the "csv" encoding, so no RasQL problem.

This works:

for m in (rgb) return
    avg(trim(m.1,
        {x:"CRS:1"(200:200),y:"CRS:1"(200:200)})
)

However condensing in parallel on a tuple still is not available:

Error: Exception: Base type typeName: typeID: 7 typeSize: 16 struct { typeName: RAS_DOUBLE typeID: 15 typeSize: 8 red, typeName: RAS_DOUBLE typeID: 15 typeSize: 8 green } as ElementType is not supported yet.

[Dimitar Misev]

Yes, csv should is meant to be used for array results, it doesn't make sense for scalars.

[Dimitar Misev]

Patch applied, changeset:0f7748d2366f67968c7f49308a8c1fcb0026e0f2

[Piero Campalani]

Re-opening since the associated WCPS test (n.85) fails again now:

select avg_cells(c[100:102,241:243]) from rgb AS c

I get this error:

SEVERE:   rasj[0] RasRNPImplementation.getResponse: type not supported: typeName: 
 typeID:  7
 typeSize: 16
struct 
{
  typeName: RAS_DOUBLE
 typeID:  15
 typeSize: 8
 red, 
typeName: RAS_DOUBLE
 typeID:  15
 typeSize: 8
 green
}

[Dimitar Misev]

I think it is unnecessary to reopen this, as the issue is in rasql and you have already opened #638

[Dimitar Misev]

Ah sorry, I didn't read your comment fully.

That should not happen, do you have the latest master?

[Piero Campalani]

Yes, maybe I did not include the very latest patches: I am recompiling now to make sure (do you get that too btw?)

[Dimitar Misev]

I'm not sure about this one, but #638 is confirmed.

[Piero Campalani]

Ok, re-compiling the query runs fine. WCPS test n.85 valid again.