[#XJU-559475]: Re: Fwd: netcdf4 vs. netcdf3

Arlindo,

 nc_supress_endianess_conversion ( ncid, varid );

is certainly a simple interface.  I assume you would also want this to suppress 
conversion on input as well as on output, because users have reported 
byte-swapping on reads is also a bottleneck on little-endian machines.

We have had a few requests for a similar feature in netCDF-3, and as you know, 
something equivalent will be available in netCDF-4.  I'll consider this 
enhancement for netCDF-3, but there are several other features ahead of it on 
the TODO list, so I can't commit to adding it soon.  Performance has never been 
a primary goal for netCDF, since it often conflicts with other desirable 
software characteristics, but you've made a good case for how important this 
can be in some cases.

It may not seem like much work to add just one little function to netCDF, but 
we would also have to add test code, Fortran bindings, C++ bindings, netCDF-4 
equivalents, reference documentation, and User Guide documentation.  And it has 
to be supported in the OPeNDAP client libraries, so a user can just relink for 
OPeNDAP support.  This creates quite an obstacle to adding new functions.

Maybe this would be a better addition for pnetcdf, whose primary goal is 
performance on HPC platforms.  I'll have to think about it a bit ...

--Russ


Russ Rew                                         UCAR Unidata Program
address@hidden                     http://www.unidata.ucar.edu



Ticket Details
===================
Ticket ID: XJU-559475
Department: Support netCDF
Priority: Normal
Status: Closed