[MITgcm-devel] Re: [MITgcm-support] Re: MITgcm-test

[chris hill]

Hi Ed,

 What you need to do add is a test compile and run in the same way as
you do for system(), clock(), fdate() etc... and then set the config
appropriately.

 "which netCDF" can be used to find a candidate -lnetcdf but it may
still give problems due to linking conventions so compiling and testing
a code fragment is needed too.

 I realise this slows down genmake2, but it will then work and we can
think about a split form later.

Chris
On Mon, 2004-09-27 at 09:31, Alistair Adcroft wrote:
> Ed,
> 
> Martin's right. netcdf should not be enabled if it isn't available. It's 
> not good enough to say everyone should install it since it isn't always 
> possible. Can you modify genmake2 to intelligently configure mnc 
> appropriately?
> 
> A.
> 
> Martin Losch wrote:
> > Hi,
> > 
> > new week, new game, new unqualified suggestions (o:
> > Would it be possible to check, whether netcdf is avaible at the genmake2 
> > or even testreport step and then decide on the type of output in the 
> > verification experiments? I guess, that's what my suggestion last Friday 
> > was aiming at anyway.
> > For example, a "which ncdump" would show the ${whereever}/bin directory, 
> > where the netcdf utilities that come with the distribution are; then, if 
> > looking for netcdf.inc and libnetcdf.a in ${whereever}/include and 
> > ${whereever}/lib fails, one can assume that netcdf is not installed 
> > properly, issue a comment/warning and just not use the mnc-package for 
> > testing with testreport on this particular machine until the use has 
> > created a customized build option file.
> > 
> > In the end, it's probably a political issue, isn't it? If you want users 
> > to completely switch to netcdf in the long run, anything that would let 
> > anyone avoid installing and using netcdf, is a bad idea.
> > 
> > Martin
> > 
> > 
> > On Sep 24, 2004, at 11:06 PM, Patrick Heimbach wrote:
> > 
> >> Hi Ed,
> >>
> >> Quoting Ed Hill <eh3 at MIT.EDU>:
> >>
> >>> On Fri, 2004-09-24 at 16:24, Patrick Heimbach wrote:
> >>>
> >>>> Hi Ed,
> >>>>
> >>>> that was me (with -addr).
> >>>> I was going to show her how she could get
> >>>> to compile & run the model on her platform in 10 min,
> >>>> and then got surprised by the netcdf stuff.
> >>>>
> >>>> The issue is that it's quite unfortunate
> >>>> that one of our basic (since closest to real application)
> >>>> setups is now failing on all platforms that don't
> >>>> have netcf installed.
> >>>> I would opt to use a less high-profile verif. for this (e.g. exp2).
> >>>
> >>>
> >>>
> >>> Hi Patrick,
> >>>
> >>> Since this is a new user, why not show her how to install NetCDF and
> >>> then she'll have output thats easier to understand?  I mean, its *so*
> >>> easy to do!
> >>>
> >>> Or point her towards the MITgcm-support list and we'll gladly walk her
> >>> through the install process.
> >>>
> >>> Ed
> >>
> >>
> >> I already did that, and she already downloaded the rpm's for her RH9.
> >> But I still think that this misses the point of using
> >> something (currently) non-standard for a "high-profile"
> >> verification. Just my opinion...
> >>
> >> -Patrick
> >>
> >>
> >> _______________________________________________
> >> MITgcm-devel mailing list
> >> MITgcm-devel at mitgcm.org
> >> http://dev.mitgcm.org/mailman/listinfo/mitgcm-devel
> > 
> > 
> > _______________________________________________
> > MITgcm-devel mailing list
> > MITgcm-devel at mitgcm.org
> > http://dev.mitgcm.org/mailman/listinfo/mitgcm-devel
>