Floating point exceptions in IOM (NetCDF)

[ufla]

While hunting other bugs, we have seen floating point exceptions occur in the iom_g[123]d subroutines, which are interface'd to iom_get. This meant that the code would crash when compiling with ifort plus the fpe0 flag. The actual FPE's occurred further down the call tree, in iom_nf90_get or the NetCDF lib, which seems to have a problem when the data array (pvar) arrives uninitialised.

Even though this might be dependent on the NetCDF implementation, it could be good style (and doesn't harm much) to initialise the pvar arrays in iom_g[123]d since they are intent(out). See the attached patch for suggestion.

[ufla]

Patch for iom.F90 adding initialisation for pvar in iom_g[123]d

[smasson]

Uwe,

Our "philosophy" is to avoid any initialization that is not needed. We had bad experiences with "harmless" initializations that were in fact hiding problems. pvar should be completely initialized by iom_get_123d. If it is not the case, we have a bug and we must find it.

I don't really understand your problem. As far as I understand, it looks that nf90_get_var from your NetCDF library is not accepting non-initialized arrays? But this argument is declared as an intent(out) in nf90_get_var, so its initial values should not have any impact? If it is the case, I would say the problem is more coming from your netcvdf library than from NEMO, no?

Which version are you using? What is the revision number of your iom.F90 and lib_mpp.F90?