Hi Steve, Steve McIntyre schrieb am Sat 05. Apr, 13:00 (+0100): > On Sat, Apr 05, 2008 at 01:39:00PM +0200, Jörg Sommer wrote: > >Steve McIntyre schrieb am Sun 23. Mar, 19:51 (+0000): > >> Package: slrn > >> Version: 0.9.9~pre97-1 > >> Severity: serious > >> > >> On the mips buildd "ball", slrn builds correctly, but the debian/rules > >> file does a check on the output binary objects to see if the hostname > >> is encoded anywhere. > > > >We are talking about the *FQDN*. The check uses hostname -f. > > It's not uncommon that various machine configurations (e.g. chroots, > vservers) may leave the domain bits of the FQDN empty. Also, as I > mentioned to people on IRC late last night, valid FQDNs may match too: > > 00:51 < Sledge> zobel / bzed: it's entirely possible to have a valid > FQDN that will break the slrn build > 00:51 < Sledge> imagine "mat.is", which I've just looked up and it > exists
Right, but this can be prevented by calling grep with -F. I'll improve
the grep call by this.
> >Because slrn has still the option to hard-code the domain name, I treat
> >this check as useful. I don't see why it's okay if the hostname of the
> >build machine is hard-coded in the binary.
>
> Oh, I agree on that completely. I don't agree that that justifies
> using a test that's too fragile and likely to break. What happens if
> (by chance) random optimisations in a newer version of gcc happen to
> cause strings in the binary to match the FQDN?
I don't expect this.
> Should that cause the build to fail too?
Yes, I've no problem with it. This test is there since version 0.9.6.3-5,
i.e. since more than 140 package releases and more than 7 years. It's the
first time this test catched a false postive, as far as I can see in the
changelog. I really would wait until gcc introduces random optimisations
that concanates strings without the delimiter (\0) of the first string
and think then about a solution and don't make assumptions about this.
> >If you still think it's a bug in slrn, explain why it's okay to find the
> >*FQDN* (not only the hostname ball) in the binary. Otherwise I mark this
> >bug as wontfix.
>
> I don't care about this particular case (ball), that's not the
> point. I'm pointing at a more general problem that could cause future
> random build failures, both on buildds and for users. Do you
> understand where I'm coming from?
You think of these hosts:
% strings /usr/bin/slrn G -io '[a-z.]*[a-z]\.[a-z][a-z][a-z]\?' | \
while read i; do host $i &> /dev/null && echo $i; done
space.mit.edu
gmx.de
ftp.gtk.org
slrn.sourceforge.net
jedsoft.org
I would like to talk to John Davis about this problem. I found out that
the values USE_DOMAIN_NAME and MY_DOMAIN_NAME aren't used anywhere in the
code. So, my assumption the test is still needed, might be wrong.
Bye, Jörg.
--
“Politics is for the moment, equations are forever”
(Albert Einstein)
signature.asc
Description: Digital signature http://en.wikipedia.org/wiki/OpenPGP
