Re: [Libreoffice] unit-test dependencies (Re: Tests on release build)

On 15/12/11 17:37, Michael Meeks wrote:
> On Thu, 2011-12-15 at 13:18 +0000, Caolán McNamara wrote:
> > On Thu, 2011-12-15 at 13:56 +0100, Michael Stahl wrote:
> > > we wouldn't have this problem if these were subsequenttests
> >
> > but then no one would run them :-), I still feel its far better to try
> > and track down and fix the dependency or parallel test problem, whatever
> > it is, then move things into a subsequenttest because if the immediacy
> > of "make" to "pass/failure" isn't there I just have no faith that people
> > as lazy as me will run them :-(
>
>       I tend to agree. If we have obscure crashers for our unit tests during
> make - then we should fix them :-)
>
>       Oddly hwpfilter (two of the three reports) is included into tail_build,
> and thus it's prj/build.lst dependencies are irrelevant, it should be in
> the new, pristine world of gnumake-ness. Which makes this distinctly
> odd. Those files should not be moving around after creation - surely ?
> and make is all a single process, so it'll wait until the file creation
> rule is complete before relying on it rather than statting and finding a
> 1/2 done file.

well that of course would be the case if the dependencies in the test
makefile were complete.  which they apparently are not.  which is in no
way obvious.  which is my point.

>       Seemingly tail_build depends on fpicker which depends on ucb - so the
> ucb1, ucbfile1 components should be in-place.

and i have already wasted half an hour looking at the libraries the darn
thing loads and couldn't find anything obvious that was missing...

>       Like Michael S, I too have seen this intermittently fail too - which is
> deadly annoying.
>
>       I guess we could add a '--disable-unit-tests' configure option for
> Gentoo-ers.

<cultural_stereotype>
hmm, presumably Gentooers don't care that much whether the stuff they
build actually works as long as it's fast :-P
</cultural_stereotype>