This is the mail archive of the fortran@gcc.gnu.org mailing list for the GNU Fortran project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Record length 32bit vs. 64bit in gfortran


Hello,

for those not reading comp.lang.fortran, there is a longer discussion of
32 vs 64 bit record lengths in gfortran at:

http://groups.google.com/group/comp.lang.fortran/browse_thread/thread/062ce3447e5ef570

Richard E Main wrote:
"I don't like gfortran's choice of a default there. I know that it would
be a problem
for users of my codes with gfortran. And no, having a compiler switch
doesn't really solve it as long as the default is the "wrong" way."

And Steve Lionel of Intel wrote:
"While it is true that the DEC compilers (and much more so the Intel
compilers) have a variety of environment variables for adjusting
semantics for I/O, there is no such control for 32 vs. 64 bit record
lengths in unformatted files.  Instead, we use a record structure that
uses 32-bit lengths as long as the length is less than 2GB, and an
alternate format for longer records. This preserves compatibility with
most existing implementations while still allowing longer records
without having to "flag" the file first or deciding based on the
platform."

And further:
"I was somewhat disappointed a few years later to see that straight
64-bit lengths were used by the GNU compilers as this was incompatible
with standard practice for all files."

What do you think about the Intel way? I think it is a nice solution,
though one would have the incompatibility between older and newer
gfortran versions; i.e. one needed to have a method (compile time or run
time) to change between that and the current behaviour.


Tobias


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]