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

[Support #DKO-246585]: make check fails f90 tests when built with 64

> End,
>
> I have certainly tasted madness.  I have spent several days trying to
> understand ncfortran.h mostly to no avail.
>
> You ask:
> You want to calculate with 64-bit integers, but write 32-bit integers?
>
>
> The answer is no. The application calling the library needs to use 64-bit
> integers. Therefore when using a Fortran interface to a library function
> any mention of an integer in the function call prototype will cause a
> 64-bit value to be passed at runtime. I assumed that if I built the netCDF
> libraries in a 64-bit data environment (Fortran -qintsize=8) then this
> should work.

OK, but netcdf is about storage, so what is it you want to store? 32-bit or 64
bit values?

I notice from rereading your emails that you said the f77 tests. You could
build with --disable-f90 and just use the f77 API, if it works. It's almost as
simple as the F90 API.

But I am having a hard time wrapping my head around this problem. Is what you
are doing even legal for netcdf?

You want to write a 64-bit integer, but you want to call it a NF_INT? That is
not correct.

The correct answer would be to change the code to explicitly use 8-byte
integers in the code, correct? Can't f90 do that with the "kind" keyword?

OK, but it is not a perfect world. Perhaps what you need to do is get netcdf-4,
which has native handling of 64-bit integer (with the NC_INT64 and NC_UINT64
types). But you would still have a problem that if you tried to hand netcdf an
integer you are giving it 64-bits and the library is expecting 32-bits.

Perhaps the answer is to build netcdf-4 as a 32-bit integer application and
then call it from your application, writhing NC_INT64s, and handing it your
data.

I need to discuss this with my colleague Russ about this...

Ed

Ticket Details
===================
Ticket ID: DKO-246585
Department: Support netCDF
Priority: Normal
Status: Open