test failure, glibc make check (chroot pass)

Bob Morgan wb5aoh at wb5aoh.dyndns.org
Mon Feb 23 13:00:15 PST 2004

Ken Moffat writes:
 > On Fri, 20 Feb 2004, Bob Morgan wrote:
 > > spitting out.  Those will be this weekend's projects.  The box in
 > > question is at home, and the email accessibility is at work, so it
 > > will probably be Monday afternoon here before I get a chance to
 > > get back on the list.
 > >
 >  Hope this response catches you, then
 > > I guess my question about @INC is:  I can see the string being spit out
 > > in the error message, but where is perl finding it in the first place?
 >  My belief is that it starts with any prefix (or failing that,
 > /usr/local) then generates the options from its own version number -
 > this would all, or mostly, be done in the configure script, and
 > exported.
 >  I think ldd is the first resort.  Maybe you logged off, and forgot to
 > set the PATH when you logged back in.  Don't know, this scenario hasn't
 > been common until now.
 > Ken
 > -- 


Well this turns out to be an embarrasing typo.  It's the old lowercase L
vs. numeric 1 character bug, and it bit me real bad.  I had created
perl directories as per15 instead of perl5 (look closely, I haven't been
able to SEE the difference on any of 3 largescreen monitors all week).
Simply renaming directory per15 to perl5 made perl -V behave, and then
I blew away and reran the glibc build, and everything worked, and I am
off and running again.  I think I got as far as building the patch binary,
and that is where I quit late last night.

I did run ldd on not only perl, but did: ldd /tools/usr/bin/* |more
(and again for sbin), and didn't turn up anything that didn't start
with the /tools directory.  But the fact that in the beginning you
suspected a perl problem certainly got me looking in the right places.
Without that tip I would still be looking for mtrace nowhere in particular.

Anyhow, the original error was a typo when I created the directories in the
first build of perl, typing a 1 instead of an l.  Simply renaming that one
directory, and starting over with glibc configure/make straightened everything out,
and I am off and running again very smoothly.  Between the first and second
pass I got GPM working, and that gets rid of the rekeying errors.
There's probably 1-3 pixels difference in the 1 and l, and the brain
just filters it out as noise, and the tab-completion always found
what was nearby and convenient depending on the context at the time.
I only SAW it by pasting the error output from perl -V back into
the command line for ls, to see why I could do manually what it couldn't,
and finally I got a manual command to give me a file not found error, and
at that point I finally saw it on the screen.  As for @INC, it obviously
knew how its directory should have been spelled, and was not knowledgeable
of how its directory name was actually spelled, in other words, a
precoded list of possibilities.

Thanks to you and the list for all of the help straightening out this
silly typo.


More information about the lfs-support mailing list