cvs commit: src/sys/sys cdefs.h src/include nl_types.h stdio.h

[Warner Losh]

From: Steve Kargl <sgk at troutmask.apl.washington.edu>
Subject: Re: cvs commit: src/sys/sys cdefs.h src/include nl_types.h stdio.h
Date: Wed, 1 Feb 2006 09:58:04 -0800

> On Wed, Feb 01, 2006 at 12:40:34PM -0500, Kris Kennaway wrote:
> > On Wed, Feb 01, 2006 at 09:37:58AM -0800, Steve Kargl wrote:
> > > On Wed, Feb 01, 2006 at 12:35:27PM -0500, Kris Kennaway wrote:
> > > > 
> > > > Didn't we go through this the other week?  'make buildworld' is the
> > > > only safe way to build your world.  If you try some shortcut method,
> > > > and it fails, then don't complain until you've done 'buildworld' and
> > > > it's also failed in the same way.
> > > 
> > > The error message I quoted was from a buildworld!
> > 
> > OK, this was unclear to me.
> > 
> 
> It's a catch-22.  The panics on amd64 appear to be from a
> 2006-01-26 src tree.  Stefan happened to change stdio.h on
> that day.  If one panicks his system 10 to 15 times trying
> to isolate the problem, one runs the risk of damaging a
> filesystem.  In this case, the panic appears to be related
> to devfs or file descriptor code.  Well, "make buildworld"
> in parallel with a "gmake check-gfortran" in a GCC tree
> writing to /usr can trigger the panic in 5 to 10 minutes,
> but this scrambles /usr.  So, yes, my /usr/include and src/
> are probably out-of-sync, but it is what it is. 

This is definitely a cross-threading situation.  However, I'll see
what I can do to mitigate the effects.  I have a more general fix for
the building current on 5.4, but that won't apply here.

If you really need to get a good set of includes, then you'll have to
cvsup from 2006-01-25 and rebuild from those sources.

I'm actually surprised that make includes doesn't install the sys
directory as well, which would pick up the new sys/cdefs.h.  Does
'make -f Makefile.inc _includes' also exhibit the same problem?

Warner