[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: caballero disk space

Greetings!  I've uncovered further information, but must confess to
remaining stumped as to exactly what is going on.  In brief, a rebuild
of the same gcl source and installation of the resulting deb will most
likely reenable the maxima build on ia64.  Here are the 'facts':

1) gcl source successfully builds on March 3, then successfully builds
   maxima 5.9.0-4 later the same day.

2) gcl fails to build maxima 5.9.0-5 three days later, with no
   significant changes to the maxima source.

3) gcl installed on merulo dchroot (before merulo went down, as it
   apparently is now) immediately dies on startup

4) gcl built on merulo dchroot with 'dpkg-buildpackage -rfakeroot
   produces a viable deb, which, when unpacked in my home directory on
   merulo, produces a correct gcl which builds maxima.

5) The binaries have exactly the same size, but differing md5sums.  In
   particular, the .text start address is different for each.  The
   final gcl image is produced by an elf 'unexec' routine taken
   directly from emacs.  While I'm unsure of the details, it may be
   possible that changes in dynamic library dependencies might confuse
   the dynamic loader ldso at certain .text addresses, though this
   would certainly appear to be a ldso bug if true, and has never
   appeared anywhere else in my experience.  In any case, it does not
   appear that the dynamic libraries on which gcl  depends have
   changed in the three day interval in question.

6) It is known, as before stated, that fakeroot interferes on ia64
   with the mprotect/segv catching mechanism used by gcl's garbage
   collector.  However, both the autobuilder build on 3/3 as well as
   my own under merulo's dchroot ran the critical gcl build portion
   outside of fakeroot, apparently eliminating this explanation.

7) the gcl build produces two images, a traditional and an ansi one.
   both fail in the versions James installed under merulos' dchroot
   environment.  When using strace, the first appears to fail due to a
   lisp core corruption (i.e. bogus pointer values), while the latter
   appears to die on attempting to trap SIGSEGV for garbage

Maybe some of you might have an idea? In any case, will someone please
do a give-back on gcl and maxima to the autobuilder?  This is the only
course of action that makes sense at this point.

Take care,

James Troup <james@nocrew.org> writes:

> Camm Maguire <camm@enhanced.com> writes:
> > Has anything changed in the build daemon (between say 3/3, and 3/7?)
> I don't know about that but the ia64 buildd has always (AFAIK, and
> certainly at least while I was running it) used fakeroot for it's
> builds.
> > I'm having trouble checking out the actual gcl build in unstable as
> > neither merulo nor caballero appear to have chroots.
> Err, merulo does too - I just installed the build-deps for gcl and
> maxima in it's unstable chroot for you.
> -- 
> James

Camm Maguire			     			camm@enhanced.com
"The earth is but one country, and mankind its citizens."  --  Baha'u'llah

Reply to: