This is the mail archive of the gcc-bugs@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: bootstrap/5187: gcc 3.0.3 fails to bootstrap on Sun Ultra 60withSolaris 8


"Joseph S. Myers" wrote:
 
> If you check out or update from CVS, or using patches between snapshots,
> you must ensure that timestamps are in the right order by running
> contrib/gcc_update --touch.  Snapshot and release tarballs should have the
> timestamps correct to ensure that autoconf is not run during build.

I've only used the release tarball - 3.0.3, so I assume autoconf would not be
needed. 

I renamed autoconf (which was version 2.50) to something else, so that there was
no autoconf in the search path. I then did a configure followed by a "make
bootstrap" on the gcc-3.0.3. This built okay, with no error messages in the
build process, suggesting that Joseph Myers is correct in assuming that autoconf
would not be needed. However, the 'make install' did not produce a usable gcc.
Hence while autoconf 2.50 may be broken, as suggested by Paolo Carlini and at
the web page:

http://gcc.gnu.org/ml/gcc-patches/2001-12/msg01727.html

a broken autoconf on my computer is *not* the reason I'm unable to get a decent
install of gcc-3.0.3. Clearly the same problem occurs with or without autoconf
in the search path on my machine. 

That is *not* to say that the 'configure' script in the gcc-3.0.3 distribution
is not broken, due to a broken autoconf being used to generate it. I've no idea
what version of autoconf was used to generate the configure script from the
configure.in file. 

As it is now late, and I've been to the pub this evening, I will not continue
with gcc-3.0.3 any more tonight. I will however try a rebuild of gcc-3.0.3
tomorrow morning while setting 'CONFIG_SHELL to ksh' as suggested by Craig
Rodrigues. This could just be the reason I'm unable to get a usable installation
on my Solaris machine.

However, given that I've used the same methods to build versions of gcc from
2.8.x, to 3.0.2, including 3.0 and 3.0.1, using the same sh and tcsh, I suspect
the problem is not with the Solaris tcsh or sh, but is due to a bug in
gcc-3.0.3. Most of my experience has been with a Sun SPARCstation 20 with a
32-bit kernel - the new machine I'm now using is an Ultra 60 with a 64-bit
kernel (I'm not however attempting to build a 64-bit gcc). 

I'll rebuild tomorrow with:
a) No autoconf in the search path (which I have already done)
b) CONFIG_SHELL set to ksh (which I have not done).

and let you all know the results for certain. My strong suspicion is that there
is a bug in 3.0.3 though. 

If anyone has any other explanations/suggestions, I'd be interested. 

-- 
Dr. David Kirkby Ph.D,
email: drkirkby@ntlworld.com 
former email address: davek@medphys.ucl.ac.uk
web page: http://www.david-kirkby.co.uk/
Amateur radio callsign: G8WRB


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