hubert@entropy.UUCP (Steve Hubert) (02/08/85)
Libm says log(negative) = -HUGE and sqrt(negative) = 0.
Both set errno = EDOM which is reasonable. I guess the values
0 and -HUGE must come from the idea that the functions should
be continuous at zero or some voodoo like that.
If I am not mistaken, Fortran programmers are used to being yelled at
when they try to pass an out-of-domain argument to a function.
C programmers, as always, are a different story. Anyway, does anyone,
besides me, think it would be a good idea to explicitly warn the user
when s/he attempted to call log with a negative argument? If so,
have you come up with a smooth way of doing it?
Steve Hubert
Dept. of Stat., U. of Wash, Seattle
{allegra,decvax,ihnp4,ucbvax!lbl-csam}!uw-beaver!entropy!hubert
hubert%entropy@uw-beavergwyn@brl-tgr.ARPA (Doug Gwyn <gwyn>) (02/11/85)
> Anyway, does anyone, > besides me, think it would be a good idea to explicitly warn the user > when s/he attempted to call log with a negative argument? If so, > have you come up with a smooth way of doing it? Yes, on UNIX System V there is a matherr(3M) library function that is invoked when a math library routine detects an error. By default it prints an appropriate warning message on stderr. The programmer can provide his own matherr() and do whatever he thinks is appropriate upon occurrence of a math library error.
pedz@smu.UUCP (02/21/85)
I thought that on a Vax with 4.2 there was a whole set of signal codes (plus one signal) which is used for floating point errors. Why not pick the appropriate signal along with its code and send it out? I do not know how to get the code of the error in with the signal since I do not remember that with the kill call but in any case, I would make it an "interrupt" not a error or flag which must be tested everywhere. Perry convex!smu!pedz