svn commit: r259010 - in head/sys: conf powerpc/fpu

John Baldwin jhb at freebsd.org
Tue Dec 17 17:55:15 UTC 2013


On Monday, December 16, 2013 8:32:39 am Dag-Erling Smørgrav wrote:
> John Baldwin <jhb at freebsd.org> writes:
> > Dag-Erling Smørgrav <des at des.no> writes:
> > > John Baldwin <jhb at freebsd.org> writes:
> > > > LINT64 is yet another kernel config covered by 'make tinderbox',
> > > > but not by the periodic tinderbox.  It is probably worth adding to
> > > > the periodic tinderbox (someday it'd be nice if the two
> > > > tinderboxes built the same set of things).
> > > Some day it would be nice if people talked to me directly instead of
> > > sniping from the sidelines.
> > Ah, but when people have raised this exact issue before (that tcbuild
> > and 'make tinderbox' build different things), you have blown them off
> > repeatedly.
> 
> I have no idea what tcbuild is.

I meant it as the name of your scripts (probably have ports-mgmt/tinderbox on 
the brain too much).

> I blow people off when they complain that the tinderbox doesn't work
> exactly like 'make tinderbox' because it's not intended to.

I just want it to build the same things and report errors when they are 
broken.  In that sense they should both do the same thing.

> > > Oh, and 'make tinderbox' should die.
> > No, it is very developer friendly as it Just Works(tm) as a single
> > command from an existing source tree checkout.
> 
> We already had 'make universe'.
> 
> The story behind 'make tinderbox' is that Alfred threw a fit over a
> tinderbox breakage and insisted, despite being provided with ample proof
> to the contrary, that my tinderbox was a black box which nobody knew how
> worked and nobody could reproduce and that tinderbox reports were
> therefore worthless.  He then proceeded to implement 'make tinderbox'
> purely to piss me off.  It serves no other purpose and needs to die.

Ah, a bit touchy I see.  While that might have been Alfred's initial
motivation, the points still remain that:

1) 'make tinderbox' Just Works as a single command from an existing source
   tree checkout.  I don't have to track down some other thing from some
   other SVN tree to checkout and configure.  It's also easy to reproduce
   a single failed build step in the same tree by doing 'make buildkernel'
   etc. so I can fix an issue that pops up and do quick turnaround testing
   and do a final 'make tinderbox' to make sure all is still well.

2) 'make tinderbox' provides a summary of what failed.  'make universe'
   does not.  You have to go check all the log files by hand to see if
   anything failed.  That is less helpful.

It definitely serves a useful purpose for many developers, and I for one
don't sit and cackle maniacally while rubbing my hands each time I invoke
it because I think that somehow my doing so is pissing you off.

> > It also happens to build more of the tree than the periodic tinderbox
> > (by building more kernel configs, albeit doing quite a bit of
> > duplicate work for platforms like arm in the process).
> 
> That's simply not true.  The tinderbox builds exactly the same kernels
> as 'make tinderbox' if they're present.  The issue here is that a bug in
> the source tree prevents some of these kernel configurations from being
> generated, hence the tinderbox cannot build them.

Ah, I thought at one point it only built GENERIC and LINT type configs,
but presumably that has changed?  As long as they build the same things
I'm happy.  I just want to avoid running into unrelated breakages when I
do a 'make tinderbox' on a patched tree (the unrelated breakage thing is
what I recently ran into with LINT64).

-- 
John Baldwin


More information about the svn-src-head mailing list