Re: Not-a-number (nan) issues

[Eric Norum <norume@aps.anl.gov>]

I wrote a little test program to check for proper handling of NAN/INF:

System       scanf/strtod    printf
Darwin          Y               Y
Linux           Y               Y
Solaris         Y               Y
vxWorks         N               Y
RTEMS           N               Y
Windows/Cygwin  N               Y
Windows         N               Y(1)

(1) Prints a NaN as -1.#IND and infinity as 1.#INF

scanf/strtod checked with "nan", "Nan", "NaN", "NAN", "INF", "+INF", "-INF"
printf checked with 0./0., 1./0. -1./0. and sqrt(-1.0)

So:
1) on 'unix-based' machines INF or NAN in a .db file works fine.
2) for all except windows I think that things would work if we provided an 
epicsStrtod routine (assuming that everything in EPICS is now using strdod).
3) The fact that windows prints NaNs and infinities differently poses a 
nastier problem.

--
Eric Norum                                 norume@aps.anl.gov
Advanced Photon Source                     Phone: (630) 252-4793
Argonne National Laboratory