Coverage iterator not translated properly to rasql on WCPS trimmings

[ungarj]

Just found out that some of the WCPS queries in our ​WCPS Tutorial cause an error now though having worked in earlier versions.

It seems after some testing the last version the queries worked was ​rasdaman 8.3.1-3.

Not sure whether it's relevant but the only queries affected (just tested our example queries though) are the ones with 1D output.

catalina.out indicates the "Specified domain does not intersect with spatial domain of MDD."

Example Query:

for c in (FSC_Scandinavia)
  return
    encode(
      coverage temp
        over
        $pt t(150190:150200)
      values (count(c[t($pt)]=30)) / (count(c[t($pt)]!=0)) * 100,
      "csv"
    )

The two files attached contain the tomcat output of the same query running with rasdaman 8.3.1-3 and 8.4.0-rc2.2.

Maybe I'm missing a schema or a syntax change. Do you have any idea what could be wrong here?

Thanks, Joachim

[ungarj]

catalina.out with rasdaman 8.3.1-3

[ungarj]

catalina.out with rasdaman 8.4.0-rc2

[Dimitar Misev]

Yes indeed, the coverage expression is not properly translated to rasql. Alireza can you look at this?

Old

select csv(marray i_i in [150190:150200] values ((count_cells((((c) [*:*,*:*,i_i[0]])=(30))))/(count_cells((((c) [*:*,*:*,i_i[0]])!=(0)))))*(100)) from FSC_Scandinavia AS c

New (notice the iterator variable is not used, only a 0 is used instead:

select csv(marray i_i in [150190:150200] values ((count_cells((((c) [*:*,*:*,0])=(30))))/(count_cells((((c) [*:*,*:*,0])!=(0)))))*(100)) from FSC_Scandinavia AS c

[Dirk Daems]

I can confirm that we are experiencing the same problem: iterator variables are not replaced but a zero is used when transforming the WCPS query into a RASQL query.

[Dimitar Misev]

Fixed in changeset:63a177a

[Dimitar Misev]

New bug discovered by dirk:

I have a 3 dimensional (t,x,y) coverage NDVI_SPOT_VGT.
The temporal resolution on axis t is a dekad, which is about 10 days. Each month has 3 dekads so one year has 36 dekads.

When I execute query:

for c in (NDVI_SPOT_VGT)
return encode(
  coverage averagesOverTime
    over $T t(0:10)
      values ( avg(c[x(15:20),y(0:5),t(2+$T*36)] ) ),
  "csv")

it is translated into RASQL command:

select csv(marray i_i in [0:10] values avg_cells((c) [(2)+((i_i[0])*(36)),3696:4256,3696:4256])) from NDVI_SPOT_VGT AS c

This seems to be OK.
But when I specify a range for the time axis (t) as well:

for c in (NDVI_SPOT_VGT)
return encode(
  coverage averagesOverTime
    over $T t(0:10)
      values ( avg(c[x(15:20),y(0:5),t(2+$T*36:4+$T*36)] ) ),
  "csv")

it is translated into RASQL command:

select csv(marray i_i in [0:10] values avg_cells(c[0:0,3696:4256,3696:4256])) from NDVI_SPOT_VGT AS c

So the range specified for axis t (2+$T*36:4+$T*36) is mapped to 0:0 which looks like a bug to me.
Is this a known issue?

[Piero Campalani]

Some more info can be found as well in #360 (closed as duplicate).

[Dimitar Misev]

Dirk, can you test this query again?

[Dirk Daems]

Replying to dmisev:

Dirk, can you test this query again?

Dimitar,

Just tested with the code I fetched from the git repository yesterday, but the problem remains.

Dirk

[Alireza]

DimensionIntervalElement does not handle non-numeric variables, it takes a ScalarExpr which resolve to a NumericScalarExpression from which it takes the numeric value "dvalue".
Setting `transformedCoordinates' to false works in case of GRID_CRS slice but but not in this case. Moreover slicing with an arithmetic operation involving a variable reference doesn't work either. This patch deserves a more comprehensive solution on Dimension*Element.java to admit a non-numeric outcome in the subset

[Piero Campalani]

transformedCoordinates is meant to differentiate between values already converted for RasQL, or not.
So, setting transformedCoordinates to false as in changeset:63a177a works for this specific case but doesn't make much sense to me

[Dimitar Misev]

So what's the conclusion, Piero/Alireza? How do we proceed?

[Dimitar Misev]

I'll take over the ticket, but I need details on where did you stuck last time (let's sync next week).

[Dimitar Misev]

I submitted patch that fixes it. All WCPS/WCS tests pass so it's good. I had to change the lo/hi in DimensionIntervalElement from BigInteger to String in order to get this working. This also fixed the test for #371 actually, which was only partially correct in the rasql translation.