woods@hao.UUCP (Greg Woods) (11/29/83)
The "new math" library (for f77(1)) that we got on our 4.1aBSD distribution
tape (for a VAX 11/750, if it matters) has serious bugs in it. I am talking
about the "-lnm" library, not the standard "-lm" math library. I had a program
that was somehow scribbling on the data and/or code, causing a wierd subscript-
out-of-range error (or memory fault, if I didn't use the -C flag when compiling)
even though the variable used in the subscript reference had a reasonable value.
I tore my hair out for three weeks over this, trying to track it down.
Eventually, I gave up and decided to go back to the old version of the code.
To my surprise, I got a floating exception. This was traced to a line in a
double precision subroutine that called "dexp". In this case, it occured to me
that the "more accurate" routines in -lnm might fail for certain input values
near the range limits, so I relinked with -lm. Surprise! Everything worked fine.
I'd been so frustrated by the new version of this program that just refused to
run, I was trying absolutely any wild theory, so I said, what the hell, why not
try relinking with the old math library. Another *big* surprise! It now works!
AAAAARRRGH! How can they release a math library that scribbles on the code?!?!?
Unless you are a masochist, DON'T USE -LNM!!!!!!!!
Greg "Disgusted" Woods
--
{ucbvax!hplabs | allegra!nbires | decvax!brl-bmd | harpo!seismo | ihnp4!kpno}
!hao!woods