[Proj] Proj 4.9.1RC4 Released
Greg Troxel
gdt at ir.bbn.com
Fri Feb 27 11:24:24 EST 2015
Howard Butler <howard at hobu.co> writes:
>> On Feb 24, 2015, at 9:00 PM, Greg Troxel <gdt at ir.bbn.com> wrote:
>>
>>
>> netbsd-6 i386:
>>
>> diff tv_out with tv_out.dist
>> 313c313
>> < 13 -10 2073986.9490880956873 -1680858.2722242793534 0.0000000000000
>> ---
>>> 13 -10 2073986.9490880956873 -1680858.2722242795862 0.0000000000000
>>
>> I am getting the same answer as the debian i386 boxes, I think.
>>
>> This seems to be two parts in 10^16, and it seems unreasonable to expect
>> that from doubles after actual computation.
>
> I have updated the test precision, but I'm getting a weird either/or.
>
> x86_64 is returning 0.00000
> i386 is returning -0.00000
>
> accordingly, this makes it hard to set the test case.
>
> Any ideas?
>
> https://trac.osgeo.org/proj/ticket/255#comment:8
No good ideas, other than on amd64 with more registers more of the
computation is likely to be in 80-bit mode. So +/- zero can end up
differently, and that's not necessarily wrong.
My current leaning is that for floating point tests, the only reasonable
thing to do is to have a range compare on the output.
It's not at all clear this should hold up a release, though.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 180 bytes
Desc: not available
Url : http://lists.maptools.org/pipermail/proj/attachments/20150227/6d054939/attachment.pgp
More information about the Proj
mailing list