[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[netCDF #LBV-326209]: bug with creating variable with an unlimited dimension.

Hi Ben,

> I managed to reproduce the error in my tester which I will attach.  In 
> my tester the issue is when you try to define the time variable when 
> time was an unlimited dimension. The crash is on issue is on line 273 of 
> nc4var.c the suggested_size = (pow . . . . )
> 
> In that line you have 1/(double)(var->ndims - num_set), and when it 
> crashes ndims = 1 and num_set = 1. If turned out the reason our model 
> was crashing but that my tester was not was that in our model we compile 
> with ifort and enable the -fpe0 option that enables divide by zero 
> floating point exception.
> Initially I was not building my tester with this option but when I 
> compile with -fpe0 option my little tester program will crash in the 
> same way.
> 
> We are building with intel 13.1.2.183, presumably with gfortran it would 
> be something like --ffpe-trap=0 do the equivalent of -fpe0 in ifort.

The bug is fixed in the GitHub sources.  The only file involved in the
fix was libsrc4/nc4var.c, but if you want a patch, I can send that.  The
fix will be in netCDF-C 4.3.2, and additional release candidates before
4.3.2, should we decide any are needed.

Thanks again for reporting the bug and for your work in helping isolate
it.

--Russ



Ticket Details
===================
Ticket ID: LBV-326209
Department: Support netCDF
Priority: Normal
Status: Open