Re: Hash randomization test failures
On Sun, Jan 05, 2014 at 07:10:46PM +0200, Niko Tyni wrote:
> On Wed, Jan 01, 2014 at 04:49:19PM +0000, Dominic Hargreaves wrote:
> > The following bugs are almost certainly caused by tests broken by
> > hash randomisation: they don't represent functional failures in the
> > software, and doesn't always cause FTBFS. In this sense, my view is that
> > like other sporadic build failures, they can be downgraded to important.
> > Does this sound reasonable to others?
>
> As long as it's only the tests that are broken, that does sound reasonable
> to me. I guess the probability of the failure should influence the
> decision too, but I'm not sure where to draw the line. Clearly a package
> that fails to build (for instance) 90% of the time should be considered
> RC-buggy.
>
> Also, at the risk of stating the obvious: I don't think we can tell
> whether a build failure is caused by functional breakage or just test
> suite fragility without looking at the code, right? The code could
> actually be relying on the old hash behaviour, and that could cause very
> similar sporadic test suite failures.
Yes, absolutely. From the packages I've looked at it's generally fairly
obvious if the breakage is caused by assumptions in the test suite, though,
as opposed to in the functional code.
> > #711428 libcgi-application-plugin-authentication-perl
> > #711436 libdbix-abstract-perl
> > #711444 libgraph-readwrite-perl
> > #711446 libgraph-writer-graphviz-perl
> > #711604 libmime-lite-html-perl
> > #711614 libscriptalicious-perl
> > #711615 libsearch-estraier-perl
> > #720018 libhtml-wikiconverter-moinmoin-perl
> > #720271 libnet-jabber-perl
Cheers,
Dominic.
Reply to: