Subject: Re: IEEE_754 logic
From: Kurt Hornik (Kurt.Hornik@ci.tuwien.ac.at)
Date: Thu 21 Oct 1999 - 19:20:46 EST
>>>>> Thomas Hoffmann writes:
>> >>>>> Peter Dalgaard BSA writes:
>> > > Thomas Hoffmann <email@example.com> writes:
>> >> 3. The C9x draft introduces
> the is*() functionality under the name of "classification MACROS". If an
>> >>implementation implements this standard, autoconf misses isnan() and isfinite().
> The problem for me is, that if isnan and isfinite are the way C9x says
> (i.e. are MACROS), then configure concludes "not there" (AC_FUNCS
> misses macros (intentionally)).
> From this it reasons "not IEEE 754".
> Then I get back to a native finite() in Arith.h, but IEEE 754 mode was
> switched off.
> So I see two topics:
> 1. Maybe not to (only) check if isnan/isfinite are library functions,
> but also to check for working macros (I have not found a predefined
> macro in Autoconf for this).
> This seems of some importance to me, because it is not only an
> idiosyncracy of a singular system, but may become a more widespread
> problem if systems become C9x compliant someday.
This should be doable. In fact, this is what the code in Arith.h does
for one case, sort of. I don't think that there is a macro in autoconf
that checks whether FOO exists as a library function or is defined as a
preprocessor symbol for another library function.
(We used to have a similar problem with sigsetjmp, which e.g. under
GNU/Linux is a macro ... so code which later tests for HAVE_SIGSETJMP
would not do the right thing. I ended up writing a test which uses
Question: does IEEE 754 require the `#include <math.h>'? (And do all
implementations conform?) Then we could use AC_EGREP_HEADER tests for
finite and isfinite ...
> 2. Introduce a more bulletproof check for IEEE 754.
> Is there possibly a computation we can carry out and conclude from its
> result to IEEE 754? I just have no idea.
(Don't think so. Does anyone know?)
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !) To: firstname.lastname@example.org
This archive was generated by hypermail 2b25 : Tue 04 Jan 2000 - 14:16:09 EST