Re: libglade build failure at 2004-12-03 17:24:21.337

[Andy Balaam <andybalaam artificialworlds net>]

Luis Villa wrote:

> On Fri, 03 Dec 2004 23:43:10 +0000, Andy Balaam
> <andybalaam artificialworlds net> wrote:
>  
>
> > Even better, get it to retry a few times, since this normally works
> > after a few goes.
> >    
>
> In our situation, that's not actually true, given that we're drawing
> from the main cvs servers, and the build server is in a pretty high
> availability colo. In the two months I've been running the tinderbox,
> the only cvs failures I've seen occurred when the cvs server was down
> for the count. (In the case today, because the gnome cvs box got a new
> kernel.)
>
> Still not a bad idea, of course, assuming you limit the retries- can't
> hurt even in the server failure case, and if there is a legit network
> burp, does help.
>  
But seems unnecessary given what you've just said.

> > (And then suppress the message, of course.)
> >    
> Alternately it has been suggested that I keep the previous successful
> build around, and if a build fails for whatever reason, just copy the
> old tree in for that build cycle. I want to keep the last successful
> build around for other reasons, so I'll probably just look into that.
>  
Good idea, so long as it doesn't get out of hand.  An ancient version 
hanging around could really screw things up in other modules.

Do we need some way of saying "I've terriorised!" any given error.  E.g. 
should we reply to gnome-build-status saying where we reported a bug or 
which list we informed of the error, to prevent repetition of this kind 
of stuff?  Of is the number of lists pretty finite, so this sohuld be 
discoverable?

Andy