Newsgroups: php.internals
Path: news.php.net
Xref: news.php.net php.internals:14743
Mailing-List: contact internals-help@lists.php.net; run by ezmlm
Message-ID: <20050209091640.71237.qmail@lists.php.net>
To: internals@lists.php.net
References: <20050208191532.36752.qmail@lists.php.net> <719431098.20050209002446@marcus-boerger.de>
Date: Wed, 9 Feb 2005 10:16:37 +0100
Lines: 78
Subject: Re: [PHP-DEV] 'double' / 'float' revised.. think about it.
From: r.korving@xit.nl ("Ron Korving")

That's my whole point Marcus, but the nature of the native floating point
types is something you can get around by implementing your own floating
point type, based on a base-10 system, in the same way that for example
bcmath does it. So my suggestion is; stop using float and double, and make
your own type. Small performance hit, and it will probably cost you about 1
byte extra per floating point variable, but then it will be accurate.

Ron

"Marcus Boerger" <helly@php.net> wrote in message
news:719431098.20050209002446@marcus-boerger.de...
> Hello Ron,
>
>   it doesn't matter how accurate your floating or whatever you call it is.
> Financial applications require native support which can be realized using
> bcmath. Whatever you think besides that - the problem is that any errors
> accumulate - that's the nature of float.
>
> If you would like to know whats IEEE 754 read this
> http://docs.sun.com/source/806-3568/ncg_goldberg.html
>
> best regards
> marcus
>
> Tuesday, February 8, 2005, 8:15:31 PM, you wrote:
>
> > Personal recent frustrating float issues made me think of the
following...
>
> > At the moment apparently the C-type double is used for PHP's float (or
> > double if you please) storage. I believe it is the philosophy of PHP to
make
> > things easy for its users, and in this perspective I was thinking..
hasn't
> > the time come to reimplement the PHP float type in a decimal-safe way?
Of
> > course it would be a small performance hit, but not too noticable I
think,
> > especially when utilizing simd instructions which a compiler may or may
not
> > already do. I think, especially with todays powerful systems,
performance
> > (although still a big concern) is no longer the only primary concern.
> > Shouldn't PHP abandon the speed of the double in C on this one, and
> > implement a custom implementation of the floating point type which is
much
> > more accurate, even at the cost of a few extra bytes and cycles?
> > Userfriendlyness over performance. I think it would be one hell of a
feature
> > for PHP to be mathematically correct, a world in which 1.5555 is really
> > 1.5555 and not something that approaches it, and where (for example)
> > rounding it would result in expected behaviour. I think PHP would
> > automagically be more suitable for various (financial for example)
> > applications which could potentially result in more popularity for PHP
as a
> > whole.
>
> > I know there are alternatives for accurate mathematics, but they
(naturally)
> > depend on strings which make them a lot slower than a native solution
would
> > be, and besides that, they are less nice to use compared to +,-,/,*,%
> > operators.
>
> > It's just an idea of course, and I'm curious if there are others who
feel
> > the same way about this.
>
> > Ron
>
>
>
>
> -- 
> Best regards,
>  Marcus                            mailto:helly@php.net