Re: Confusing change number info in smoke status

Op een druilerige winterdag (Wednesday 07 January 2004 12:47), schreef 
Nicholas Clark:

> On Wed, Jan 07, 2004 at 10:58:16AM +0000, Alan Burlison wrote:
> > larry is up [SunOS larry 5.6 Generic_105181-35 sun4u sparc SUNW,Ultra-1]
> >   Perl 5.8.x smoke is running
> >     change number 22081 started on 07/01/2004 03:00:54 GMT
> >         Change number 21996 started on Wed Jan  7 03:04:51 2004.
> >
> > The first change number is obtained by my script which uses 'p4 changes',
> > the second one comes from Smoke, which uses either the contents of .patch
> > or parses patchlevel.h.  Unfortunately the 3 things aren't the same.  For
> > example, at the moment the 'p4 changes' number for maint-5.8 is 22081,
> > the .patch number is 21996 and the local_patches value from patchlevel.h
> > is 21995.
>
> I suspect that your p4 changes command is getting the latest change
> anywhere in the repository. Compare:
>
> $ p4 changes -m1
> Change 22084 on 2004/01/07 by rgs@sixop 'Hardcoded numerical flag masks.'
>
> $ p4 changes -m1 //depot/maint-5.8/perl/...
> Change 22081 on 2004/01/06 by nicholas@entropy 'Move pod/perldelta.pod to
> pod/p'
>
> You'll want the latter.

and right after your p4 update, you should write that number to F<.patch> in 
your "master-tree", that should solve this inconsistency.

Perhaps we should start work on Test::Smoke::Syncer::P4?

> As to why the patch number and the local_patches level differ like that,
> I'm not sure.
> The line MAINT21995 only gets updated when the Changes file is updated
> (which is done by a script that Jarkko gave me, but I have to fire that by
> hand). So it can be quite out of date. (Which reminds me that this is
> not obvious, and when Randal uses that value to report things it's not
> actually that useful)
>
> However, I'm confused as MAINT21995 was added as change 21998 (not 21996)
> If you'd said 21998 then I'd expect that the explanation was that you'd
> happened to sync just after the change. Either that or you're faking the
> .patch number, because that file is added by the system that makes the
> rsync master, not p4.

Yup, *I* fake F<.patch> in Test::Smoke::Syncer for the case none was found, 
and found (at least in Jarkko's snapshots) that the change that commits the 
MAINTxxxxx is xxxxx+1, so I bump it one (Test/Smoke/Syncer.pm:311).
I use this for the case where smokers use the snapshots as the base for a 
clean sourcetree and apply all individual patches after that changenumber 
(from the right branch). Some people have problems rsync()ing from behind a 
firewall and most of them do not have access to p4.

Any suggestion for better guessing of the changenumber is always welcome.

(one often wonders "how random is rand() when it's picking these sigs")
Good luck,

Abe
-- 
I admit that there was too much waving the chicken and too little
looking at the chicken's genome in that change.
                                       -- Alan Burlison on p5p @ 2003-08-11