Re: [Quickfix-developers] Possible Rounding Errors

[Jody H. <jod...@at...>]

On Tue, 27 Apr 2004 14:28:40 -0500
Oren Miller <or...@qu...> wrote:

> Looking at this more closely, we have several tests that have trailing
> 
> zeros, and this works fine.  It looks like there are just too many 
> significant digits.  Are they running on a 64 bit architecture?  I 
> think the only thing you can do right now is to change the type to a 
> STRING in the data dictionary.  The alternative is to comment out the 
> validation code for doubles.

I doubt a 64 bit architecture is to blame.  More than likely, a
printf("%.15f"), or setting a large precision in an ostream, or
something similar.  The code to process doubles could be a *little* more
understanding, by using the standard limits to truncate the value to the
number of digits that can be represented on that hardware.  In the case
of a double, I'd think that would be better than rejecting the value. 
By definition, you can not throw around floating point values and expect
to get the exact same value on different machines.

That said, is there anything in the FIX spec that dictates precision of
price values to something like 6 or 8 decimal places?