Re: [global-libreoffice-ci] Tinderbox on Master for Windows-Debug - Build # 25612 - Still Failing!

On 21/02/2020 14:50, Luboš Luňák wrote:
> On Friday 21 of February 2020, Stephan Bergmann wrote:
> > On 21/02/2020 11:15, Luboš Luňák wrote:
> > >    The build is apparently incremental, so the PCH is still built
> > > with -DBOOST_SPIRIT_USE_OLD_NAMESPACE from a previous build.
> >
> > I think it is unfortunate that the Windows tinderbox builds can break in
> > such a way (currently affected are Jenkins' "Tinderbox on Master for
> > Windows", "Tinderbox on Master for Windows-Debug", and "Tinderbox on
> > Master for Win64-Debug").
> >
> > If anybody who feels responsible for those tinderboxes or for Windows
> > --enable-pch in general could get that fixed, that would be appreciated.
>
>   I've tried to add a gbuild rule to rebuild a PCH if its build flags change,
> but that's actually rather tricky - the PCH target would need to depend on a
> rebuild target, but the rebuild target would need the flags that are set only
> for the PCH target. It could possibly be handled somehow by making the rules
> more complicated, but I think in this case it's simpler to just do 'make
> clean' on the affected incremental tinderboxes. This kind of PCH breakage
> doesn't happen very often, or does it?

So we're hoping for some Meson magic to make that work reliably out of 
the box.  :)

Seems like all three affected Windows tinderboxes self-healed meanwhile 
thanks to 
<https://gerrit.libreoffice.org/plugins/gitiles/core/+/9ad252b2e79576119c2d733a1a45fdd9e9f83140%5E!/> 
"Drop o3tl::optional wrapper" causing substantial rebuild.

Anyway, if --enable-pch is known to be broken, it might make sense for 
those unattended tinderboxes to use --disable-pch instead?  CC'ing Cloph 
as their presumed owner.