Re: libc5 policy decision
Philippe Troin <phil@fifi.org> writes:
> On Fri, 01 Nov 1996 22:32:00 PST Bruce Perens (bruce@pixar.com) wrote:
> > From: Guy Maor <maor@ece.utexas.edu>
> > > Since the number of packages that don't work with the old allocator is
> > > very small, 2 confirmed so far, this doesn't really make sense. What
> > > would make sense is to build 5.4.7 with malloc-dl, and make the old
> > > malloc available as a separate library.
> > > In other words, the default behavior should be the new malloc,
> > > malloc-dl.
> > >
> > > Would you reconsider?
> > I will reconsider and go along with your proposal,
> > but I reserve the right to revisit this question ten days before release.
> Now, the question is, should we rebuild all packages so that they
> link with 5.4.7, or will 5.2.18 be ok ?
> And what is the list of packages which break under 5.4.7 ?
netscape and octave
Who's going to make the .so file with the old malloc() in it? Also,
should we put those C++ new function in it.
FWIW, I make a gnumalloc.so. I did the following:
Untar the libc source.
cd libc
./configure
cd malloc-930716
ln -s /usr/include/linux .
ln -s /usr/include/asm .
make
cd ../elfshared/libc
gcc -shared -o gnumalloc.so *
The result was a 14k file, it would probably be slightly smaller if
stripped (strip doesn't remove dynamic linking info).
I then did
setenv LD_PRELOAD /lib/gnumalloc.so (after copying it to /lib)
And tried running the netscape binary against gnumalloc.so and a
pristine libc.so.5.4.7. Went to www.gamelan.com and no crash. So it
does work, and 14k is a lot smaller than including a complete copy of
libc-5.0.9.
I assume it's inappropriate to include the binary here, but I put a
copy in /home/Debian/dunham on master.debian.org.
Steve
dunham@gdl.msu.edu
--
TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to
debian-devel-REQUEST@lists.debian.org . Trouble? e-mail to Bruce@Pixar.com
Reply to: