xmap with added dimension runs into a StackOverflow

[felixcremer]

When I am trying to make an xmap call which adds a new dimension and reuses the old time dimension I get an StackOverflow Error.
Yesterday this example run into some strange vanishing dimensions not allowed error, but today this is giving a stackoverflow error and I am not sure, what I changed. The StackOverflow is than subsequently killing my Julia session.

using YAXArrays, Zarr, SignalDecomposition

bucket = "esdl-esdc-v3.0.2"
store = "esdc-16d-2.5deg-46x72x1440-3.0.2.zarr"
path = "https://s3.bgc-jena.mpg.de:9000/" * bucket * "/" * store
path = joinpath(homedir(), "Daten/") * "bids2025_" * store
ds = open_dataset(zopen(path,consolidated=true,fill_as_missing=false))
c = Cube(ds)
function decompose_TS(xout, xin)
           any(isnan,xin) && return xout .= missing
           xsmooth, xanom = SignalDecomposition.decompose(xin)
           xout[:,1] = xsmooth
           xout[:,2] = xanom
      end
 xmap(decompose_TS, c[Variable=At("gross_primary_productivity"),
               time=Date(2001)..Date(2018,12,31)] ⊘ :time, output = XOutput(:time, Dim{:Scale}(["Seasonal", "Anomalies"])))

[meggart]

Currently adding output dimensions using a Symbol name only is not yet supported. I modified you example slightly for now and added some other fixes in #534 . With my PR the following works well for me:

using YAXArrays, Zarr, SignalDecomposition

bucket = "esdl-esdc-v3.0.2"
store = "esdc-16d-2.5deg-46x72x1440-3.0.2.zarr"
path = "https://s3.bgc-jena.mpg.de:9000/" * bucket * "/" * store
#path = joinpath(homedir(), "Daten/") * "bids2025_" * store
ds = open_dataset(zopen(path,consolidated=true,fill_as_missing=false))
c = Cube(ds)
function decompose_TS(xout, xin)
           any(isnan,xin) && return xout .= missing
           xsmooth, xanom = SignalDecomposition.decompose(xin,MovingAverageSmoothing())
           xout[:,1] = xsmooth
           xout[:,2] = xanom
      end
 
c_subset = c[Variable=At("gross_primary_productivity"),
               time=Date(2001)..Date(2018,12,31)]

r = xmap(decompose_TS,c_subset  ⊘ :time, output = XOutput(c_subset.time, Dim{:Scale}(["Seasonal", "Anomalies"])))

mem = r[lon=Near(11),lat=Near(51)][:,:]

[meggart]

And if this is for a tutorial, can we remove the roundtrip Dataset -> Cube -> At("gross_primary_productivity") and instead do ds.gross_primary_productivity?