This is the mail archive of the gcc-help@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: gcc 4.2.0 build fails on mac os 10.4.9, intel, /usr/bin/libtool: fatal error in ld64


Hi All,

Just want to close the loop here in case anybody sees the original
message below and wonders about the resolution.

Short story is that this is apparently a multilib issue, which i
believe refers to the production or perhaps use of both 32 and 64 bit
libraries, and it can be avoided by disabling multilib support.  I
received help from Dave Seaman on comp.sys.mac.programmer.tools and
Martin Costabel on the darwin dev list. (And i guess i got some
indirect help from Alexander Wieder replying to Parvathi Kumar on this
list, with the always important advice to read the host-specific
instructions at http://gcc.gnu.org/install/specific.html.  Now in fact
for intel macs, there aren't any, but it was a crucial step i had
skipped.)

The longer story is that both Dave and Martin advised using fink,
which has a gcc42 package.  When i told Dave i didn't want to use fink
because of its global nature (i.e., it's not so clear how you can have
multiple simultaneous different versions of gcc using fink, and it
requires root access to the machine), he sent me the configuration
information from gcc-4 (fink's name) via "gcc-4 -v", and said that i
could study how fink did it.

And in fact if you do actually download fink and trace through what it
is doing, it puts a lot of effort (it seems to me) into disabling
multilib at specific points (it patches both the sources and the build
system).

So, if you take a much cruder approach and hack off multilib support
everywhere, you can configure with a command something like
    ...../gcc-4.2.0/configure --host=i686-apple-darwin8
--disable-multilib --prefix=........
(Just for reference, this build is a little shorter, about 26000
lines, than a fuller build, which i think will be about 40000 lines,
and at least 30000 or so even if there's no fortran.)

This builds a gcc which can at least compile hello-world, and doesn't
require any root access.

(This does not build a fortran compiler, which may or may not be easy
to do, depending on gmp and other factors i believe: a naive build of
gmp, doing make check, certainly raises complaints.  On the other
hand, fink does build gfortran, so presumably you could just study how
fink did it if you wanted to make a separate build.)

Finally, for somebody just interested in using a single version of gcc
and who doesn't mind using root access for getting it, fink (or i
guess darwinports) is probably the easiest approach.

I appreciate all the help i've been given on this.

dan

On 6/1/07, dan hitt <dan.hitt@gmail.com> wrote:
Hi,

I'm attempting to build gcc 4.2.0 on an intel mac, os
version 10.4.9.

I have a very plain setup, no fink or other software
installed (except for the developer tools).

I ran configure with no arguments (except for the
installation path).  Per standard, i configured in
an empty directory (not the source directory).

I ran make, and it chugged along producing about
30,000 lines of output before it conked out with the
error
   /usr/bin/libtool: fatal error in ld64

This was in stage 3, working on libgcj.la (5 layers
deep).  The exact command is very long, but
starts out
   ......bld/0/./gcc/xgcc -shared-libgcc........
and ends up
   .......-compatibility_version 9 -current_version 9.0

So
  (1) is this a issue?

[I suppose that it actually is an issue, because ideally
shouldn't configure detect a problem and say "Oh, your
version of libtool isn't up to snuff, you should use
a better one" or something like that?]

 (2) Any recommendations on the next step i should
take?

Thanks in advance for any info, corrections, suggestions,
pointers to faqs, etc.

dan



Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]