Re: Jenkins verification of gerrit patches

On 01/14/2015 03:36 PM, Norbert Thiebaud wrote:
> On Wed, Jan 14, 2015 at 8:25 AM, Stephan Bergmann <sbergman@redhat.com> wrote:
> > On 01/12/2015 06:28 AM, Norbert Thiebaud wrote:
> > > FYI: the build are 'default build'. autogen.config contains pretty
> > > much nothing beyond
> > > --with-external-tar, --wtih-ant-home and sometimes
> > > --disable-dependency-tracking
> >
> >
> > at least --enable-werror should hopefully be uncontroversial to add?
>
> if is is, then why is it not the default ?

Purportedly for the benefit of both casual and distro builders, who 
might run into errors with exotic compilers and/or building against 
system alternatives of certain of our external modules where we apply 
patches to mitigate warnings.

The debate what the default should be is there since our work on Werror 
got started, and the winning-so-far argument always was that a 
responsible developer should know how to enable it, so the default could 
safely be to have it disabled (for the purported benefit of: see above).

> My problem is
> 1/ I'd like to avoid getting in the autogen.input management game

Understandable, but I see no better solution for this problem short of 
re-fuelling that tiring discussion.

> 2/ I'm concerned about false positive.. that is your gerrit patch
> verification build failed because the core you built on failed.

With the current compiler landscape, I would argue that the amount of 
warnings that go into master and cause one of the "mainstream" compilers 
to fail is sufficiently small.  And would be even smaller if the bots 
used to triage gerrit commits would be configured --enable-werror.