Re: solaris 10: address@hidden

[David Lee]

On Mon, 4 Jul 2005, Howard Chu wrote:

> Bob Friesenhahn wrote:
> > On Mon, 4 Jul 2005, David Lee wrote:
> > > I'm using GNU/ld; but I'd be happy,as part of the experiments, to try to 
> > > redirect our "gcc" to use the Solaris/ld.  (How to redirect it?)
> >
> > I recommend that anyone who installs GCC under Solaris use the Solaris 
> > linker rather than the GNU linker.  This is based on past experience 
> > learned the hard way.
>
> Agreed. The GNU linker used to be pretty good up to the SunOS 4 days, but we 
> always use the Solaris linker on our builds now. Again, past experience...

Thanks for your replies.  Much appreciated!

> From the perspective of the individual end-user, starting afresh with 
their own installation, what you say is probably appropriate.

But from the perspective of multiple suppliers of autotools-based software 
packages, each to many hundreds (or more) of end-user sites e.g. as being 
listed at:
  http://forum.sun.com/thread.jspa?threadID=24107&tstart=0

it would be useful to keep flexibility to a maximum, so far as is 
reasonably practical.  (Yes, the consensus view may be to discourage 
GNU/ld on Solaris, but it would be useful to be able to use it, or, at the 
very least, to be able to offer a statement to the effect "Sorry, but 
GNU/ld on S10 doesn't work because..." and to offer a decent explanatory 
text at the "..." as to why S10 has made things so much worse.)

For instance, do we _know_ that this S10/address@hidden problem is 
_definitely_ 100% attributable to GNU/ld?  Is "libtool" known/proven to be 
not contributing in any way to this problem?  As was mentioned in my email 
of 29th June, the "make check" phase of libtool-1.5.18 _itself_ suffers 
from this problem.  Or if we can demonstrate an S10 bug, that, too, would 
be useful.

Because this problem shows in libtool itself, it brings us one step closer 
to the underlying problem than do other suffering autotools applications. 
So investigating this failure of libtools own "make check", which is what 
I'm offering to try to assist in doing, could be fruitful in pinning down 
the problem to one of S10, GNU/ld, libtools (or to a better understanding 
of a possible double-fault problem).

So guidance would be appreciated in:
 1. how to drive libtool and/or gcc to use a different "ld" (for instance 
to Solaris/ld and a separately built latest GNU/ld);
 2. whether my earlier view of possible overloaded "build_libtool_need_lc"
is moderately sensible... in which case, what next?.

Thanks again.

--

:  David Lee                                I.T. Service          :
:  Senior Systems Programmer                Computer Centre       :
:                                           Durham University     :
:  http://www.dur.ac.uk/t.d.lee/            South Road            :
:                                           Durham DH1 3LE        :
:  Phone: +44 191 334 2752                  U.K.                  :