[geos-devel] test failures

[Greg Troxel]

Sandro Santilli <strk at keybit.net> writes:

> Theoretically -ffloat-store is already used when available,
> is that working for you guys ?

My build did use -ffloat-store.  So apparently the issue is not about
excess precision.  I have run paranoia on NetBSD/i386 - with
-ffloat-store it passes, and there are issues without it.  On amd64
paranoia passes.

I updated the '32-bit regression errors' ticket, but regression tests
pass on NetBSD 5 amd64, and had 3 errors on NetbSD 6 i386.

I wonder if there is some type that is 32 on ILP32 and 64 on LP64, and
it's having overflow, and this isn't about i386 fp precision.  I'll see
if I can manage to build on sparc (it may take all night), which is
ILP32 but does not have extended-precision double, just regular IEEE754.

> There's also a ticket to replace -ffloat-store with
> -fexcess-precision=standard
> http://trac.osgeo.org/geos/ticket/504

Hmm.   I don't know how widespread that is.

If the extra precision causes real issues (vs slightly different but not
wrong answers) then there is perhaps something else going on.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 194 bytes
Desc: not available
URL: <http://lists.osgeo.org/pipermail/geos-devel/attachments/20130820/f181c55e/attachment.pgp>