Re: Failing tests on openSUSE

2017-05-22 11:38 GMT+02:00 Khaled Hosny <khaledhosny@eglug.org>:

> On Mon, May 22, 2017 at 11:28:20AM +0200, Tomáš Chvátal wrote:
> > 2017-05-22 10:53 GMT+02:00 Khaled Hosny <khaledhosny@eglug.org>:
> >
> > > On Mon, May 22, 2017 at 10:28:03AM +0200, Miklos Vajna wrote:
> > > > Hi Tom,
> > > >
> > > > On Sat, May 20, 2017 at 11:29:03AM +0200, Tomáš Chvátal <
> > > tomas.chvatal@gmail.com> wrote:
> > > > > I really have no clue what the test is doing or how to fix it. The
> > > commits
> > > > > introducing the tests are:
> > > > > https://cgit.freedesktop.org/libreoffice/core/commit/vcl/
> > > qa/cppunit/complextext.cxx?id=62592cf1ebcc8ef381fe53855684342d963e6138
> > > > > https://cgit.freedesktop.org/libreoffice/core/commit/vcl/
> > > qa/cppunit/complextext.cxx?id=9af45ef67824f3ec6e95c16e7deb005a73532d65
> > > > It can be a good idea to CC Khaled, then. :-)
> > >
> > > The test is pretty obvious, it lays out the text and the checks the
> > > returned character widths.
> > >
> > > If the test is failing then either 1) your version of HarfBuzz is
> > > returning different glyph advances which would almost certainly be a
> bug
> > > in HarfBuzz, or 2) your version of DejaVu Sans have different glyph
> > > advances but that is unlikely, or 3) VCL ended up selecting a different
> > > font other than the one we requested for whatever reason, or 4)
> > > something else fishy is going on in VCL.
> >
> > Oky Is there way to dump what dejavu sans it found?
>
> I think no, may be there are some SAL_LOG if you are building with
> logging enabled but I’m not sure.
>
> > We are using new harfbuzz:
> >
> > [   22s] [326/807] cumulate libharfbuzz0-1.4.6-1.1
> >
> > And with the dejavu sans I get the feeling they are not present in the
> > buildroot at the moment, only liberation sans and google carlito fonts. I
> > will try to add them and see if it changes anything (results to be seen
> in
> > couple of hours).
>
> The test (naïvely?) assumes DejaVu Sans is always present since it is
> one of the bundled fonts, if this is not true since the test needs to be
> conditionally enabled somehow.
>
> > Anyway I think the test should first font-config verify it got at least
> the
> > font it requiires.
>
> VCL does not have a public API for that AFAIK.
>
> Yep the problem was caused by missing Dejavu Sans font.

Adding it to buildroot fixed the issue.

Now I have another failing test for debugging :)

Cheers

Tom