Re: Bug in XSHd5.1 tgamma

[Michael Kerrisk <mtk.lists@xxxxxxxxx>]

Fred J. Tydeman wrote:
> On Thu, 24 Jul 2008 17:52:42 +0200, Michael Kerrisk wrote:
>
> > > Close, but wrong.  Look at the wording in exp() for underflow.
> > Point taken.  But given that C99 doesn't specify the exception,
> I believe that C99 covers both overflow (for +large, +small, -small,
> -large) and underflow (-large).
The existing text C99 doesn't cover the point, because underflow is not 
a matter of -large.  As we pass through a certain continuous range of 
negative reals, we cycle through: finite, representable result; 
underflow; finite, representable result; underflow; etc.  The underflow 
and overflow conditions here are not simply a matter of the linear 
magnitude of x.  Would it not be better to describe the underflow 
exception using text which speaks in terms of the magnitude of the 
result, not the magnitude of x?

> However, based upon the comments
> in this group, it could be made clearer (I am willing to get C99/C1x
> changed if someone provides the better wording).
I'll try to spend some time on this tomorrow.