Bug#848634: python-cartopy: FTBFS on *i386: test_lambert_azimuthal_equal_area fails

[Bas Couwenberg]

On 2017-01-31 11:57, Ghislain Vaillant wrote:
> Reassigning this bug to src:proj, following Aaron's investigation.
> 
> On Thu, 26 Jan 2017 11:34:40 -0500 ucko at debian.org (Aaron M. Ucko) 
> wrote:
>> Ghislain Vaillant <ghisvail at gmail.com> writes:
>> 
>> > So, you think it would help to request src:proj to be build with 
>> > -ffloat-store then?
>> 
>> Yes; testing in a personal chroot confirms that adjusting its build
>> settings is both necessary and sufficient.  (I don't recommend turning
>> this flag on blindly.)

I'm not sure if this is desirable in proj, I think PROJ.4 is among "most 
programs" for which "the excess precision does only good".

It's probably better to make test_lambert_azimuthal_equal_area less 
strict.

The impact of this change in proj affects many reverse dependencies and 
users due to its central role in the geospatial ecosystem.

I'll need to discuss this with PROJ.4 upstream before I can consider 
incorporating the suggested change in the proj package.

If you need to fix the test_lambert_azimuthal_equal_area failure for 
python-cartopy to be included in stretch, you'll need to handle this in 
the that package since the discussion with upstream won't reach a 
conclusion before the hard freeze next Sunday (when many of us will be 
at FOSDEM).

Kind Regards,

Bas