[releng_4 tinderbox] failure on alpha/alpha

[Ed Hall]

On Sun, 17 Aug 2003 14:02:18 -0700, Andy Sparrow <spadger at best.com> writes:
> > > The same thing started in -PORTS quite some time ago, where I find 
> > > personally find that it generates more cr at p than the real traffic at 
> > > times.
> > 
> > You're entitled to your opinion,
> 
> Thanks, I will clarify it further for you.
> 
> > but since you've never had to deal
> > with the flood of support requests when INDEX builds were broken by
> > careless committers before I started the automated tinderbox,
> 
> Wouldn't the real issue be to control the careless committers then? Or 
> to target them specifically and directly with the Tinderbox failures?
> 
> When I automated overnight builds of mutiple branches of a commercial 
> product on mutiple OS platforms, sending those build results 
> company-wide was never considered as an option.
> 
> I just don't see why it isn't more appropriate to simply limit the 
> messages to people with a commit bit, a specific email alias, or even 
> people who checked stuff in since the last sucessful Tinderbox.
> 
> > I'd
> > suggest you try to consider it from point of view of those of us who
> > are actually involved in the support of the OS.
> 
> It's not that I don't appreciate the efforts that are being made so much 
> as I question the elegance of the solution employed.
> 
> Some people pay for (limited) bandwidth by time on-line, and cannot 
> filter except after receipt, thus have no choice but to *pay* to 
> retrieve those messages before filtering them, so it's not simply a 
> question of whether they "just hit delete" or filter them out or 
> whatever.
> 
> Those messages thus inevitably dilute the value of the list for them, I 
> suggest you try to consider it from *their* point of view.
> 
> There's also the issue that all the descriptive fields for -STABLE and 
> -PORTS say that these are "discussion" lists - which *used* to be true. 
> Multiple posts from Bots don't make for much of a "discussion", in my 
> book.
> 
> Whatever. Procmail works for me, but not everyone has that choice.

The tinderbox posts are quite valuable -- more valuable than most of
the posts here (including this one).  They can save me and others
here a lot of time since they let me know whether or not CVS is in a
buildable state.

They are sent to -stable for a couple of reasons that I can see:

1) Users are instructed to read -stable if they are running STABLE,
   and check the list before they report problems to see if a problem
   has already been reported.  A tinderbox failure is such a problem
   report, and no less valuable for being automatically generated.
   And even if a build on a different architecture than you're using
   is the one that is breaking, it is often a symptom of a problem
   that could affect everyone.

2) Sending build failures to a semi-public list provides peer pressure
   on committers to fix their stuff, quickly.  Despite your claim to
   the contrary, commercial organizations (even one I can think of in
   Redmond, WA) do this inside their engineering groups.  Subscribers
   to freebsd-stable form a similar group of peers and internal
   customers.

When a build fails, work cannot progress.  It's an urgent situation.
A lot of people are affected, not just the committer who broke the
build or even just committers in general.  Perhaps *you* aren't
affected, but honestly, now, what percentage of messages on this
list directly affect you?

		-Ed