Re: Debian package pspp-0.8.5-2 is on mentors - added a patch to fix the s390 regression failure

[Ben Pfaff]

On Mon, Jun 22, 2015 at 08:28:01AM -0700, Ben Pfaff wrote:
> On Mon, Jun 22, 2015 at 04:43:10PM +0200, Friedrich Beckmann wrote:
> > i have uploaded a new debian version 0.8.5-2 on mentors
> > 
> > http://mentors.debian.net/package/pspp
> 
> Thanks, I built and uploaded this in turn to Debian.
> 
> > This package contains the debian/patches/render.patch (see below) which
> > hopefully fixes the s390 regression failure.
> > 
> > I made the patch based on the assumption, that the s390 regression failure
> > and bug https://savannah.gnu.org/bugs/?45272 have the same root cause.
> > 
> > I find the problem analysis quite difficult but my change is based on the
> > behavior that the current xr_measure_cell_height function will result zero 
> > height, 
> > if the width is 0. The MacOS bug is provoced by a text field which includes
> > just carriage returns. Then the computed width is 0 and 
> > as a result the height is also 0. This results in the assertion if this 
> > happens at
> > the end of a page which is the case as the previous rendered item is a 
> > chart.
> > 
> > With the patch below the xr_layout_cell function is called in 
> > xr_measure_cell_height
> > which yields a non zero height and the problem disappears. 
> 
> Oh, that's a really good observation.
> 
> > Do you think this change may break other parts? I thought that the code
> > was just a shortcut to avoid the call to the xr_layout_cell.
> 
> I'm not sure; although the core rendering code is (almost?) all mine,
> it's complicated and I don't recall the details.  I'll take a detailed
> look if it fixes the s390x bug.

Unfortunately it failed in the same way:
        
https://buildd.debian.org/status/fetch.php?pkg=pspp&arch=s390x&ver=0.8.5-2&stamp=1434989331

I'll still try to take a detailed look tonight, especially since you
added a reproduction case to the bug report.