I am currently working on a general-purpose IOSP for Matlab files, based upon
the JMatIO package (http://sourceforge.net/projects/jmatio/) which can handle
Matlab file version 5 files rather well. With some modifications JMatIO can
handle Matlab file version 4 as well.
Currently, the approach for mapping between Matlab and the CDM is:
Represent all 1xN char arrays as global attributes
Represent other 1x1 arrays as global attributes
Represent all other Matlab arrays as ucar.ma2.Array.
Most numerical types map directly, except:
UINT8 maps to SHORT
UINT16 maps to INT
UINT32 maps to LONG
UINT64 maps to LONG (may cause overflow).
Matlab Structures map to ucar.nc2.Structure
Matlab Cell arrays are handled in one of three ways:
- If all the data is the same type, it is represented as a regular
ucar.ma2.Array.
- If all the data is identical within each column of the array, it is
represented as a ucar.nc2.Structure with a member corresponding to each column
named "[name]_[col#]".
- Otherwise the cell array is represented as a ucar.nc2.Structure,
where each cell is named "[name]_[row#]_[col#]" for MxN cell arrays,
"[name]_[row#]" for Mx1 cell arrays, or "[name]_[col#]" for 1xN cell arrays.
Open questions:
Is anyone else interested in a general purpose Matlab IOSP?
Does my approach make sense?
How to represent imaginary components of arrays? Perhaps represent as a 2nd
matrix with a prefix like "_i_" or a suffix such as "_imag"?
Is there a better way to name the members of Structures that result from
handling Matlab Cell arrays?
This will need ongoing effort. Is there a "best" place to host IOSP
development projects (SourceForge, Google Code, etc?)
Thanks , Mike
_____________________________________________
Michael A. Godin
Software Engineer
Monterey Bay Aquarium Research Institute
Phone: 831-775-2063 http://www.mbari.org <http://www.mbari.org>
