delorie.com/archives/browse.cgi   search  
Mail Archives: djgpp-workers/1999/06/24/10:29:29

Date: Thu, 24 Jun 1999 09:26:33 -0500
From: Eric Rudd <rudd AT cyberoptics DOT com>
Subject: Re: libm sources from cyberoptics
To: Eli Zaretskii <eliz AT is DOT elta DOT co DOT il>
Cc: djgpp-workers AT delorie DOT com, dj AT delorie DOT com
Message-id: <37724019.45E19654@cyberoptics.com>
Organization: CyberOptics
MIME-version: 1.0
X-Mailer: Mozilla 4.05 [en] (Win95; U)
References: <Pine DOT SUN DOT 3 DOT 91 DOT 990624120124 DOT 25071G-100000 AT is>
Reply-To: djgpp-workers AT delorie DOT com

Eli Zaretskii wrote:

> I would suggest to submit a bug report anyway, and let the maintainer
> decide what would be the best practical way of solving that.

I'll do that, after my work on the math functions tapers off a bit.

> The pow2 and pow10 cases are taken care of by libc/stubs.h, so you
> don't have to do anything.  As for exp2 and exp10, since they aren't
> ANSI, I suggest to remove _exp2 and _exp10 labels from pow2.S and
> pow10.S, and add two short .S files that jmp to __pow2 and __pow10,
> respectively.

Done, but bear in mind that the names exp2 and exp10 will be standard once C9x
issues, so we will eventually want to change this scheme back to the way it was.

> > I will zip up the corrected docs along with the new source.

I have just uploaded the changes to DJ's server in the file

   /incoming/math0624.zip

I also made modifications to about half of your docs.  Most of the changes were
to include additional information about behavior for exceptional arguments.  I
also changed the wording here and there to clarify the descriptions.  If you
have objections to any of the changes, edit them as you see fit.

You had also mentioned a possible problem with ldexp not setting ERANGE.  Last
night, while testing it further, I typed in some args wrong, and realized what
the problem might have been.  To test for overflow, I typed in 1, 1E400 for the
arguments, and didn't get ERANGE.  I then realized that the second argument in
an integer, and the %d format just read in the 1E400 as 1 for the exponent.
This, of course, shouldn't overflow.  I wonder if you did the same in testing
ldexp?

-Eric Rudd

- Raw text -


  webmaster     delorie software   privacy  
  Copyright © 2019   by DJ Delorie     Updated Jul 2019