Turning the bootstrap checker into a facist...

[Phil Edwards <pedwards at disaster dot jaj dot com>]

...would be a Good Thing, IMHO.  (I'm replying to a public message of Mark's,
but shifting lists for general discussion.  (Hey, you posted it.  :-))


On Tue, Mar 13, 2001 at 08:30:45AM -0800, Mark Mitchell wrote:
> 
>   In fact, one of our SC members has suggested automatically removing
> patches that fail Geoff's regression checker,

Speaking as someone who just watched all 222 libstdc++ tests fail,
and is not looking forward to tracking down the cause, I like this idea.

I've been surprised before that the CVS committ scripts don't already fire
off some kind of checker (that's their intended purpose, as we all know),
but admittedly it would be a PITA to run a checkin and have to wait forty
minutes before learning that the checkin didn't take place.[*]

However -- and maybe the regression checker is in fact smarter than this;
Geoff please correct me if I'm wrong -- we'd have to make the checker look
closer than it currently does; a per-patch check rather than every-so-often.
Right now, I'll check in a documentation patch, and the next day get cc'd
on a message to five other people that /one/ of us broke the build, but
the checker doesn't know which one; my doc patch just happens to be within
+/- delta lines of the culprit in the ChangeLog.  Clearly that should not
cause my doc patch to be automatically rejected.


[*] We might see a return to the days of brutally and carefully hand-checking
the code because submitting a job to the compiler involved handing a stack of
paper cards to another human, and not getting the results back for 24 hours.
A misplaced comma would kill a day and a half if you weren't careful.
This wouldn't be a bad thing either; I see many patches that get committed
"while waiting for the bootstrap" only to generate three messages from
other people pointing out typos.


> and others have gone so
> far as to suggest requiring an automated cross-platform bootstrap
> *before* check-in.

That would be also be very good, but quite time-consuming unless the
Checker (note how it's taken on a life of its own now *grin*) is running
on a very fast machine.


Just my two timeslices,
Phil

-- 
pedwards at disaster dot jaj dot com  |  pme at sources dot redhat dot com
devphil at several other less interesting addresses in various dot domains
The gods do not protect fools.  Fools are protected by more capable fools.