[RFC] decimal float point patch based on libdecnumber: testcase

[Wu Zhou]

On Thu, 22 Jun 2006, Daniel Jacobowitz wrote:

> On Fri, Jun 23, 2006 at 07:27:29AM +0800, Wu Zhou wrote:
> > That is because they have different precision/exponent. Number 
> > 1234567890123456.dd have 16 digits, its coefficient is 1234567890123456, 
> > and exponent is 0, so it won't print back with an exponent.  While in the 
> > case of "1.2E10dd", the coefficient is 12, exponent is 9, so it will print 
> > back with an exponent. And in exponent display mode, the coefficient will 
> > be normalized, in this case, to 1.2; and exponent get to 10 respectively.
> > 
> > For "p 1200000000000000.dd", it will return 1200000000000000. It is 
> > equal to but different than 1.2E+15.  Their precison is not the same.
> 
> So they map down to different bit patterns.  How bizarre.

Yes, they do.  A little bizarre when I first saw this.  But I don't feel 
like this now.  :-)  It is desirable to be able to differentiate 
between floating numbers with same value but different precision anyway.

> 
> > Do you mean that we need to output something more close to the fact. to 
> > say, "Addition/Multiple of decimal floating point is not supported right 
> > now".  or something other like this.  If you want, I can do that.
> 
> No, I think what's there is fine - but add a test or two for that error
> message.

OK.  Understand.  We can use XFAIL for them, right?

Regards
- Wu Zhou