This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


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

Re: scanf() cand read what printf() produced: NaNs and Inf


>I would have put this about #4 on my list of top 3 wants for g77.  AFAIK the
>standards for C and Fortran insist that the behavior of attempting to read
>back such data files is undefined; I don't think even the IEEE standards
>suggest a standard text form which these tokens should take.  I would find
>this extension useful if it were to work between gcc and g77.

Proper numerical I/O is crucial for g77.  If the libc/libm libraries
underlying libg2c don't take care of this for us on all systems, then
we should put ensuring that libg2c does, either by getting the
fixes into libf2c or putting them into libg2c ourselves.

(Anyone have a reference to an on-line version of a suitable document
explaining what "proper numerical I/O" *is*?  I'll check with my
Interval friends, but would like to get a balanced picture of peoples'
opinions.  I know there's at least one paper, of which I think I have a
physical copy.  I'd like to add references to such documents to the
g77 docs, perhaps as part of a new item under "Projects" about
reviewing and improving, as appropriate, libg2c's handling of
numeric<->decimal-ASCII I/O.)

        tq vm, (burley)


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