[vtk-developers] Experimental buildbot builds are distracting

[Ben Boeckel]

On Tue, Oct 20, 2015 at 08:00:55 -0400, Ken Martin wrote:
> I can see the experimental buildbot group being useful for staging new
> builbots to see if they would be green, or moving master buildbots down if
> an OS upgrade breaks something and it has not been fixed yet (Trey would
> have been a candidate for this)

The problem with downgrading is that we generally have minimal cover
over the feature sets we're testing. For example, trey is the only
osx+python3 tester and the only osx+qt+opengl2 (kamino is just osx+qt)
tester. And since the experimentals are generally ignored…

> But having long running persistent experimental failing buildbots
> seems of questionable value.  I do not consider experimental buildbots
> in evaluating the quality of a topic.

Agreed. For VTK, agora should be fixed and made expected as it's the
only machine testing the Intel compiler and PGI. Kargad is flaky and
probably worth deciding whether to boot off or not. Amber10 has this
error:

..\..\..\..\source\ThirdParty\netcdf\vtknetcdf\libsrc4\nc4hdf.c(3666):
error C2177: constant too big
..\..\..\..\source\ThirdParty\netcdf\vtknetcdf\libsrc4\nc4hdf.c(3666):
error C2177: constant too big

which is weird since every other Win64 builder doesn't care. Is it
VS2010-specific? Unlikely since ParaView has 2010 builders which don't
have this error. And it looks like I need to get a Qt4 build on nemesis
which has QtWebKit enabled (or disable the webkit-using modules).

--Ben