phobos commit, revision 1557


user: rsinfu

msg:
We should not expect exact precision of strtold().
The precition may vary between platforms or even different versions of libc.

Tested on FreeBSD and Gentoo Linux with dmd r501.

http://www.dsource.org/projects/phobos/changeset/1557

On 5/25/2010 10:59 PM, dsource.org wrote:
> phobos commit, revision 1557
> 
> 
> user: rsinfu
> 
> msg:
> We should not expect exact precision of strtold().
> The precition may vary between platforms or even different versions of libc.
> 
> Tested on FreeBSD and Gentoo Linux with dmd r501.
> 
> http://www.dsource.org/projects/phobos/changeset/1557

I disagree with this 'fix'.  D has taken a pretty strong stance on numerical accuracy so far.  That some platform libc's have bugs means we shouldn't use those parts of those libc's and instead provide an implementation that works correctly.

Brad Roberts <braddr at puremagic.com> wrote:
> On 5/25/2010 10:59 PM, dsource.org wrote:
> > phobos commit, revision 1557
> > 
> > 
> > user: rsinfu
> > 
> > msg:
> > We should not expect exact precision of strtold().
> > The precition may vary between platforms or even different versions of libc.
> > 
> > Tested on FreeBSD and Gentoo Linux with dmd r501.
> > 
> > http://www.dsource.org/projects/phobos/changeset/1557
> 
> I disagree with this 'fix'.  D has taken a pretty strong stance on numerical accuracy so far.  That some platform libc's have bugs means we shouldn't use those parts of those libc's and instead provide an implementation that works correctly.

That makes a lot of sense. I see; strict unittests are requirements for D, and those unittests should fail on these platforms.

I'll revert the change later. (dsource.org is down now)


Shin