Hey Tom,
We got around to coming back to this issue and ran smack into a brick wall.
Worse it's toward the end of another big project and this function is a key
piece, so now I'm on the hook. Looking for information more than actual code
digging. A new piece of information that surfaced was that when we generated
netCDF files from the same BUFR point files on the 32 and 64-bit machines and
ran ncdump on them, a comparison of the ncdump output showed only some rounding
differences with the exception of the header info. Ncdump had no problems
displaying all of the data in the files. What I would like from you is
validation (or not) of my perception that this information shows we are
creating the netCDF files correctly on the 64-bit system, but that ncdfks is
having issues. We found that the 'hacked' version of ncdfks was not required
to read the files on the 32-bit systems; we are using the McIDAS core ncdfks.
If this theory is correct, and SSEC tells me they have almost no one with any
netCDF experience any more at least with the older McIDAS-X platforms so it's
possible, then I can leave the transition software alone and focus on the
server solution or eliminating the netCDF part completely and dumping it to
some other structure.
Am I missing something here, do you think? BTW we did make sure of all the
linking issues before this.
Thanks,
Brice
-----Original Message-----
From: Biggerstaff, Brice A9
Sent: Wednesday, May 15, 2013 8:26 AM
To: 'support-mcidas@xxxxxxxxxxxxxxxx'
Subject: RE: EXTERNAL: [McIDAS #ABG-678484]: netCDF and McIDAS
Thanks, Tom.
We'll will look into the linking issue and get back to you.
Brice
-----Original Message-----
From: Unidata McIDAS Support [mailto:support-mcidas@xxxxxxxxxxxxxxxx]
Sent: Tuesday, May 14, 2013 11:50 AM
To: Biggerstaff, Brice A9
Cc: support-mcidas@xxxxxxxxxxxxxxxx<mailto:support-mcidas@xxxxxxxxxxxxxxxx>
Subject: EXTERNAL: [McIDAS #ABG-678484]: netCDF and McIDAS
Hi Brice,
re:
> You're probably not the exact person to ask this question, but maybe
> you can direct it to someone that might be able to give me a hint of
> trail to follow.
Ready...
re:
> We're up-converting to RHEL 6 and McIDAS 2012 and we've run into a
> puzzling issue concerning our in-house decode to NetCDF. Since 2003
> we have been pulling BUFR point forecast files from NCEP and
> converting them to netCDF here and then serving them via McIDAS ADDE
> for in-house and the Cape Canaveral Air Force weather group. One of
> our team hacked the netCDF code, version 3.6.2, a little to add a
> character string length and made a minor mod to the SSEC ncdfks server
> and we've been running fine.
I assume that you are aware that SSEC and Unidata McIDAS come bundled with a
newer version of netCDF than 3.6.2. Given this, it may be necessary for your
team member to re-hack the netCDF code that McIDAS routines link against.
re:
> However, after we migrated the code to RHEL 6 64-bit we ran into an
> unexpected glitch. The ncdfks server fails with a 'Dimension
> mismatch' error.
This sounds like either a 32 bit/64 bit problem (improperly declared variable,
etc.) or linking against a different releases of netCDF than you think you are
linking against.
re:
> The strangeness is that if I take one of the converted netCDF files
> and put it on a RHEL 4 32-bit system, the ncdfks there, (ours or the
> unmodified SSEC version) does not fail but gives good data.
This really does sound like a 32 vs 64 bit issue somewhere.
re:
> I've looked into the ncdfks code some (although it's in C and that's
> not my best language) and I've sent samples and corresponded with the
> MUG, but they confess to being puzzled as well and not possessing the
> highest of expertise in netCDF. They suggested that I contact you
> guys. So I'm still going slowly through the code, but if something
> rings a bell about the 32-bit vs 64-bit oddity to you or someone up
> there, I would appreciate any ideas.
I am willing to take a look to see if anything jumps out at me. What I would
need to do this is:
- list of code that your team member modified in the first place
And the source for the modified routines.
- one or two example datasets (BUFR files and converted netCDF files)
I don't know exactly when I could get to the snooping, but I am willing to try
when I get a chance.
Cheers,
Tom
--
****************************************************************************
Unidata User Support UCAR Unidata Program
(303) 497-8642 P.O. Box 3000
support@xxxxxxxxxxxxxxxx<mailto:support@xxxxxxxxxxxxxxxx>
Boulder, CO 80307
----------------------------------------------------------------------------
Unidata HomePage http://www.unidata.ucar.edu
****************************************************************************
Ticket Details
===================
Ticket ID: ABG-678484
Department: Support McIDAS
Priority: Normal
Status: Closed
