[Met_help] possible error in METv1.1

John Halley Gotway johnhg at rap.ucar.edu
Wed Aug 27 07:11:39 MDT 2008


Glad to hear the BIGENDIAN flag is working.

As for the PrepBufr issues, I'm thinking that it's probably a 64-bit
versus 32-bit issue.  The only way to get the PB2NC tool to work on a
64-bit machine is to build MET and its dependent libraries as 32-bit. 
Please see a discussion of this here:

We're looking at things to address this difficulty in the future.


> John Halley Gotway wrote:
>> Matt,
>> Thanks for sending that.  I don't see anything obvious.  Have you had a
>> chance to try compiling with the -DBIGENDIAN architecture flag set?
>> John
>> J Matt Clark wrote:
>>> John Halley Gotway wrote:
>>>> Matt,
>>>> Did you ever receive my last email about a test program?  I got a
>>>> weird bounce message indicating that it might not have gone through.
>>>> I've copied it below.
>>>> Also, regarding endianess.  MET was built and initially tested on
>>>> little endian Linux machines.  However, we were able to port it to a
>>>> big endian IBM machine - NCAR's supercomputers.  If your PowerPC
>>>> is big endian, you'll need to set some architecture flags at compile
>>>> time.  Take a look in the top-level "Makefile_ibm" for an example, and
>>>> look for the "ARCH_FLAGS" line:
>>>> You could try setting your ARCH_FLAGS to -DBIGENDIAN and recompiling.
>>>> Here's my other message:
> John,
> Using -DBIGENDIAN appeared to have solved the problem w/ grid_stat - at
> least I don't get the segfault.
> There remains the problem of pb2nc, which seems to only read pbfiles
> created on the same architecture. Is there a means
> to convert pbfiles across architectures - i.e. to process little-endian
> files on a big-endian system?   I've not yet found a combination which
> works - because of differences in the record-marker length (4 vs 8) and
> endianess.
> Thanks,
> Matt

More information about the Met_help mailing list