Date: Mon, 29 Jul 1996 21:09:48 -0400
Message-Id: <199607300109.VAA25467@delorie.com>
From: DJ Delorie <dj>
To: leisner AT sdsp DOT mc DOT xerox DOT com
CC: eliz AT is DOT elta DOT co DOT il, djgpp-workers AT delorie DOT com
In-reply-to: <9607291934.AA17075@gnu.mc.xerox.com> (leisner@sdsp.mc.xerox.com)
Subject: Re: Long double support


> How useful is long double to most applications?
> 
> double provides such high precision, 15k of bloat to support something
> which is:
> 	not ANSI
> 	not used (by my estimate 99.999% of the time)
> I would say if it is needed have a special libldbl.a which has
> to be used to handle long doubles.

To summarize previous postings, I think the best approach is to

1) Add low-overhead support into libc's printf.  These would cast
everything to double and work with the existing double formatting
code.

2) Write high-precision printf/scanf routines and put them in libm.
Like the ansi math functions, you get a high precision and full IEEE
compliance if you link in libm.

This is only acceptable if there are no basic *functional* differences
between the two implementations; only precision and code bloat.
(i.e. they must obey the same format parameters, but libm's might
handle NaNs and large precision specifiers better).