Netcdf Fortran developers,
Using netcdf-4.2.1.1 and netcdf-4.2-fortran, currently Windows and Mac OSX.
I use current Intel compilers everywhere. I haven't gotten around to Linux,
but I will very soon.
I have a couple of comments I would like to share about the fortran
interface (both f77 and f90).
[1] Use of "integer" rather than "integer*4" and "integer*8": By using
integer for all arguments, the fortran interface is not consistent with the
C interface. On the fortran side, I decide either "i4" or "i8" as the
default size for integer. On the C side, all functions return "int", all
ids and many parameters are also "int", ranges are typically given by
"size_t" (which is almost never int on modern systems). I have no easy way
to write both integer*4 and integer*8 data. Yes I know about the int64
variants in the nc4 interface, but the index and count variables that allow
me to get a section of data will still be i4 if that happens to be the
default.
For better consistency, all C int parameters should be integer*4, all C
size_t paramters should be integer*8, and data choices should correspond to
typical fortran kind(1, 2, 4, 8). That may not match on all platforms, but
it will cover the majority. I know people may not be happy with having to
update their code, but at least it would be consistent moving forward.
[2] To support i8 on Windows, I added support for "NF_INT_IS_C_LONGLONG"
everywhere. I will be happy to share this code, but it is incomplete. I did
not patch the f90 code because it is in such flux as you attempt to move to
f03 support (in the 4.4 beta).
[3] I hope you do not abandon the f77 branch. We still depend on it to
build on older platforms such as IRIX (and older systems where we cannot
update the compilers). I completely understand if you decide to remove it.
I will continue to use it in older versions as necessary.
Thanks,
Roy Dennington
Semichem, Inc.
