Provide support for nested arrays with rasdl datatypes

[barboni@…]

I tried to define a datatype composed by nested array

Rasdl works and data is created successfully but when i try to generate the collection rasql return "Unknown collection type"

---

#How to reproduce the issue
#Load new datatype - meaTypes.dl is attached
su -l rasdaman -s /bin/bash -c "/usr/bin/rasdl -r meaTypes.dl -i —connect RASBASE"

#check datatype with rasdl —print
su -l rasdaman -s /bin/bash -c "/usr/bin/rasdl —print —connect RASBASE" | grep profile22Pixel
su -l rasdaman -s /bin/bash -c "/usr/bin/rasdl —print —connect RASBASE" | grep SMHI

#create collection
rasql -q 'create collection SMHI SMHIImageSet' —user rasadmin —passwd rasadmin

#the error:
rasql: rasdaman query tool v1.0, rasdaman v8.3.1-g6877ac4 — generated on 21.01.2013 18:43:50.
opening database RASBASE at localhost:7001...ok
Executing update query…rasdaman error 956: Update error 956 in line 1, column 1, near token create: Unknown collection type.
aborting transaction…ok
rasql done.

---

rpm -qa | grep rasdaman
rasdaman-8.3.1-5.el6.x86_64
rasdaman-petascope-8.3.1-5.el6.noarch

[Dimitar Misev]

We discussed this with Alan a bit and came to the conclusion that this will lead to difficulties on several levels, including WCPS/WCS.

The suggestion is to implement selecting and importing a single subdataset from a netcdf file of several datasets, rather than importing several subdatasets at once and having to deal with nested composite types. What do you think? From the usage perspective it won't make much difference, except that the subdatasets will be directly accessed by collection name in rasdaman, instead of as fields of one collection.

[abeccati]

Yes, I'm not even sure that kind of nesting would be supported by GML in rangeType. Howhever a couple of options to overcome that come to my mind:

They can still belong to the same collection, just losing the nesting. The target datatype should be then something like 112 float bands, just labeled accordingly: lat; lon; nh4no3_1; nh4no3_2; …. nh4no3:22; sulfate_1 and so on…

I'm curious about the dimensions of your coverage however since you are putting geographic coordinates in the range set. What axis would be in the 2D and 3D case?
In case lat lon would be suitable as dimesnions, also band ranges should be, in that case you can define a domain such as:
[lat, lon, product] → 22 * float rngeset. In the metadata you can then map product labels to CRS1 values for nh4no3; sulfate; nh42so4; nitrate; pm10;
Giving you a smooth 22 float values range type.

[abeccati]

Changed to feature as it is not meant to be supported yet. Lowered priority since workaround exist and data can still be handled.