Error reading a variable with cdms2

[jbeezley]

I get the following error when trying to slice a variable originating on an opendap server:

 $ python -c "import cdms2; cdms2.open('http://esgcet.llnl.gov/dap/ipcc4/20c3m/ncar_ccsm3_0/pcmdi.ipcc4.ncar_ccsm3_0.20c3m.run6.atm.mo.xml').variables['ua'][0, 0, :, :]"
ncvargets: ncid 65536: NetCDF: Variable has no data in DAP request
Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "<prefix>/cdms2/variable.py", line 82, in __getitem__
    return AbstractVariable.__getitem__(self, key)
  File "<prefix>/cdms2/avariable.py", line 1399, in __getitem__
    return self.getSlice(numericSqueeze=1, raw=0, isitem=1, *speclist)
  File "<prefix>/cdms2/avariable.py", line 656, in getSlice
    result = self.subSlice(*specs, **keys)
  File "<prefix>/cdms2/avariable.py", line 567, in subSlice
    axis = self.getAxis(i).subaxis(slice.start, slice.stop, slice.step, wrap=(numericSqueeze==0))
  File "<prefix>/cdms2/axis.py", line 1256, in subaxis
    fullBounds = self.getBounds()
  File "<prefix>/cdms2/axis.py", line 1931, in getBounds
    boundsArray = self.getExplicitBounds()
  File "<prefix>/cdms2/axis.py", line 1949, in getExplicitBounds
    boundsArray = numpy.ma.filled(boundsVar)
  File "/Library/Python/2.7/site-packages/numpy-override/numpy/ma/core.py", line 582, in filled
    return np.array(a)
  File "<prefix>/cdms2/avariable.py", line 138, in __array__
    return numpy.ma.filled(self.getValue(squeeze=0))
  File "<prefix>/cdms2/fvariable.py", line 161, in getValue
    return self.getSlice(Ellipsis, squeeze=squeeze)
  File "<prefix>/cdms2/avariable.py", line 656, in getSlice
    result = self.subSlice(*specs, **keys)
  File "<prefix>/cdms2/avariable.py", line 549, in subSlice
    d = self.expertSlice (slicelist)
  File "<prefix>/cdms2/fvariable.py", line 84, in expertSlice
    result = apply(self._obj_.getitem,slist)
IOError: Unknown error

However accessing the components one at a time works:

$ python -c "import cdms2; print cdms2.open('http://esgcet.llnl.gov/dap/ipcc4/20c3m/ncar_ccsm3_0/pcmdi.ipcc4.ncar_ccsm3_0.20c3m.run6.atm.mo.xml').variables['ua'][1, 10, 0, 0]"
-1.70405423641

netCDF4 python also reads it fine:

$ python -c "from netCDF4 import Dataset; print Dataset('http://esgcet.llnl.gov/dap/ipcc4/20c3m/ncar_ccsm3_0/pcmdi.ipcc4.ncar_ccsm3_0.20c3m.run6.atm.mo.xml').variables['ua'][1, 10, :, :]"
[[-1.70405424 -1.72662032 -1.74810171 ..., -1.62960672 -1.65557122  -1.68037927]
 [-1.28498518 -1.27757108 -1.26952016 ..., -1.30228746 -1.2974571   -1.29165149]
 [-0.92818403 -0.89008546 -0.85163492 ..., -1.038167   -1.00244784  -0.96571213]
 ..., 
 [ 5.33871937  5.37260056  5.40242577 ...,  5.21467352  5.259552     5.30097008]
 [ 4.2086339   4.22407246  4.23648071 ...,  4.14485407  4.16894293   4.19023132]
 [ 2.89284873  2.89466238  2.89463282 ...,  2.87655139  2.88376951   2.88921022]]

[doutriaux1]

@jbeezley that is odd indeed. Will look into this... I suspect a numpy 1.9 issue, we are overdue for a major overhaul of the numpy inheritance anyway.... 1.9.2 breaks evrything.

[durack1]

@jbeezley I think you already know this, but at least in netcdf-c 4.3.3.1 there does appear to be something problematic with the DAP integration as per Unidata/netcdf-c#115.

At least in the current UV-CDAT/master we're still using 4.3.2 https://github.com/UV-CDAT/uvcdat/blob/master/CMake/cdat_modules/netcdf_pkg.cmake

[jbeezley]

Yes, that will probably prevent us from updating the netCDF library, but the issue here seems to be unrelated.