On 1/4/2011 11:49 AM, Christopher Barker wrote:
> On 1/4/11 10:15 AM, John Caron wrote:
>>> So exactly how much does NetCDF-Java do for you? Can I ask it for a
>>> land-sea mask and it will look for the multiple ways that might be
>>> stored and give it back to me? If so, pretty cool!
>>
>> It cant do that right now, but that functionality could be provided in a
>> standard API, and there is a framework for implementing the API by
>> different dataset conventions. This idea of providing a common API is
>> pretty much the idea of the "Common Data Model" or CDM.
>
> well, there is a data model, and then there are conventions. Having
> the library understand the conventions is a great idea. For this
> simple example, it would solve the problem, and doing it in one place,
> rather than separately in all client code. I've thought of working on
> a CF python package to build on Jeff Whittaker's NetCDF4 package --
> putting all the CF convention stuff in one place.
that would be great
>
> > IFAIU, Python cant interact directly with Java (ie link to it).
>
> Correct. It can potentially work with it through the JNI.
>
> Here is a package to make that easier:
>
> JCC 2.7: a C++ code generator for calling Java from C++/Python
>
> http://pypi.python.org/pypi/JCC/2.7
>
> Another option is GNU CNI -- have you tried to compile any of your
> code with gcj? That would have the advantage of not having to run a
> JVM at all.
thanks, i never hear much about this option. I wonder if there are
problems with this approach?
>
>> We have
>> an experimental project here to write a C library that communicates
>> across a socket to a "CDM stack" in Java, presumably much easier than
>> duplicating the CDM in C.
>
> yup -- though then you have to running a server of some wort in Java,
> yes? And does this work well across platforms (i.e. Windows?). It
> could be pretty cool, though.
yes and yes (though we havent done it yet)
>
>> One of the pieces of that project is to put a
>> Python layer on top of that, as a way of making it immediately useful
>> and to get feedback on the API.
>
> Sounds good -- Python is better for test code, too.
>
>> We are thinking of making Python an
>> important component of Unidata data access technologies (resources
>> permitting, etc, etc).
>
> Also sounds good -- Python use is really growing in the scientific
> community.
>
> Another option is to duplicate the CDM in Python -- easier than C for
> sure, probably easier than JAVA (though you've already got JAVA), but
> not so useful to C/C++/Fortran developers.
>
> There is also Jython, which can work directly with JAVA classes --
> that would be a fine way to do some scripting and testing, but not
> much use to most Scientific Python users, who rely on the "cpython"
> interpreter, and other C packages, like numpy.
yup
>
> -Chris
>
>
>
Received on Wed Feb 02 2011 - 18:17:53 GMT
