ScFiltersTest::testOrcusODSStyleInterface fails with liborcus 0.13.3

[Kohei Yoshida]

Hi there,

> On February 22, 2018 at 8:01 AM Olivier Tilloy <olivier.tilloy at canonical.com> wrote:
> 
> 
> Hello,
> 
> I initially filed
> https://bugs.documentfoundation.org/show_bug.cgi?id=115931 but it was
> suggested to me to raise the issue on the mailing list instead.
> 
> So here it goes: when building libreoffice 6.0 with an external
> liborcus (version 0.13.3), unit tests fail:
> 
> subsequent_filters-test.cxx:2398:Assertion
> Test name: ScFiltersTest::testOrcusODSStyleInterface
> equality assertion failed
> - Expected: Color: R:254 G:255 B: 204
> - Actual : Color: R:255 G:255 B: 255
> 
> This is most likely caused by that upstream change:
> https://gitlab.com/orcus/orcus/commit/f821995022df8dd1e580dd22cf131584b2b1ac4f
> 
> I'm not familiar with that code, but I came up with a tentative patch
> (which I will actually put to the test soon):
> 
> --- a/sc/source/filter/orcus/interface.cxx
> +++ b/sc/source/filter/orcus/interface.cxx
> @@ -839,7 +839,7 @@ void ScOrcusStyles::fill::applyToItemSet
>          return;
>      }
> 
> -    rSet.Put(SvxBrushItem(maBgColor, ATTR_BACKGROUND));
> +    rSet.Put(SvxBrushItem(maFgColor, ATTR_BACKGROUND));
>  }
> 
> Thoughts and feedback welcome!

Yes, so, you have raised this in a timely fashion.  While I was backporting a patch against orcus for the 6-0 branch last night, I did notice the possibility of this very issue, and was thinking of how to address it for that branch.  Since you've already discovered this and emailed it to the list, let me provide my thoughts on this.

First of all, your patch should work fine.  However, with this change, we need to ensure that liborcus version 0.13.3 or later be used.  If you used an earlier version this change does break things.

What I was thinking of proposing is backporting the following commits:

20945a9a4de6684010fd5b3603595e6da543807d
a1c36eff089c3cd695bd78090575ca1c7677121e

raise the orcus baseline to 0.13.3, and commit all of these as a single backport commit to ensure that these all happen atomically.  This way if someone wants to build from the 6-0 branch using the system liborcus then the configure script should be able to catch it and fail.

How does this sound?

Kohei