[Cdash] Problem with cdash, subprojects and CTEST_USE_LAUNCHERS=TRUE ?

[David Cole]

On Fri, Aug 17, 2012 at 1:50 PM, Alexander Neundorf <neundorf at kde.org>wrote:

> **
>
> On Wednesday 08 August 2012, David Cole wrote:
>
> > Are the subprojects built via ExternalProject_Add calls?
>
> >
>
> > If so, and the default make command contains the "-i" flag, then
>
> > perhaps make will return no error even if there are build errors. And
>
> > if the make does not return an error, then ctest with launchers will
>
> > not report an error. (Whereas without launchers, it will simply scrape
>
> > the output looking for matches to the known set of error regular
>
> > expressions...)
>
> >
>
> > I'm pretty sure this is the case that we've seen before with a
>
> > subproject-based dashboard using a Super Build ExternalProject
>
> > approach...
>
> >
>
> > The solution is to either avoid using the launchers for such project
>
> > dashboards, or come up with a way to make "make" return errors when
>
> > there are build errors.
>
> >
>
> > If I am off base here, perhaps you could give steps to repro so I can
>
> > take a closer look?
>
> I got the same problem on a different project today.
>
> No cdash subprojects, 2 or 3 ExternalProjects and the big main project.
>
> When CTEST_USE_LAUNCHERS is TRUE, the build fails, but the dashboard is
> green, 0 build errors, 0 warnings.
>
> The build fails in the main project, not in the ExternalProjects.
>
> This is with cmake 2.8.9.
>
> Strange is that it works without problems for a similar project, which
> also has a few ExternalProjects, and where I didn't find differences in the
> setup or anything yet.
>
> Any tips how to debug what is going wrong ?
>

First try "--debug" and "-VV" on the ctest command line when you're calling
the -S script. Next, start adding printf debugging if necessary to the code
where the return values of launcher-launched programs are being calculated.
I bet one of those two things pinpoints what's going wrong. If not, you can
dig into the ctest code even deeper...



> I will look further into it next week.
>
> Alex
>
>
Good luck.

:-)
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/cdash/attachments/20120817/0e6dfdc8/attachment-0003.htm>