cvs commit: src/kerberos5/lib/libroken Makefile

[Andre Oppermann]

Ken Smith wrote:
> 
> On Tue, Feb 03, 2004 at 11:28:24AM -0500, Tom Rhodes wrote:
> > On Tue, 3 Feb 2004 01:32:11 -0800 (PST)
> > Ruslan Ermilov <ru at FreeBSD.org> wrote:
> >
> > > bland       2004/02/03 01:32:11 PST
> >   ^^^^^
> >
> > You're bland... :)
> >
> > >
> > >   FreeBSD src repository
> > >
> > >   Modified files:
> > >     kerberos5/lib/libroken Makefile
> > >   Log:
> > >   Take signal.c out of sources.
> > >
> > >   Reviewed by:    nectar
> >
> > Any idea what is causing this yet?  Perhaps I missed/overlooked/ignored
> > the discussion?
> >
> > --
> > Tom Rhodes
> 
> We're actively trying to track it down now.  No ETA on it being fixed
> at this point, it is probably buried deep inside some perl scripts
> but that's just a guess.  We have caught the last couple of examples
> to help try and track it down though (two or three have happened in
> the last two or three days).

All this magic happens in CVSROOT-src/cfg.pm where $PID for temporary
files and $COMMITTER are set.  $PID is probably not the problem but
$COMMITTER seems to be.  The text in this variable is put verbatim into
the commit message in CVSROOT-src/log_accum.pl function build_header().

I'm not a perl hacker so I can't really comment on the code snipped
below but it looks strange that the first $ENV{} has the environment
variable name in "" and the second one in '' quotes.  At least it seems
to be inconsistent.  I wonder if we ever can get into the case where
LOGNAME and USER are not set in the users environment.  It might be
that for some SSH setups the environment is cleared or altered on
freefall.  Nontheless I don't think any of the others should return
anything else than the current username.  Unless there is a bug in
the kernel wrt credential handling or some other bug in perl.

 # The username of the committer.
 $COMMITTER = $ENV{"LOGNAME"} || $ENV{'USER'} || getlogin
 		|| (getpwuid($<))[0] || sprintf("uid#%d",$<);


-- 
Andre