Subject: Re: static vs. dynamic runtime linking, again (was: PAM and su -K)
To: Jason Thorpe <thorpej@shagadelic.org>
From: James Chacon <jmc@NetBSD.org>
List: tech-userlevel
Date: 01/25/2005 11:40:10
On Tue, Jan 25, 2005 at 09:30:43AM -0800, Jason Thorpe wrote:
>
> On Jan 25, 2005, at 7:27 AM, James Chacon wrote:
>
> >This is why on solaris I patch gcc to at least push -R lines to the
> >linker
> >for this as well. Now, why the GCC maintainers insist the broken
> >behavior
> >is "correct" is beyond me...
>
> I've been meaning to cook up a patch that allows the -R for
> GCC-implicit libraries to be selected on a per-target-configuration
> basis. This would solve the problem for NetBSD, and allow the GCC
> people to retain the current behavior for their favorite platforms
> (they have reasons for wanting it; not that I agree with those reasons,
> but they do have a rationale, nonetheless).
I beleive the way I patched it was to add a new spec entry definition
and then have the portion of the code which happily auto-add's -L also add
-R in those cases as well via the spec expansion.
So, what is the rationale for shipping a compiler that will link and not
run programs "out of the box" on Solaris (a fairly major platform for gcc)?
It's amazingly simple to make the -R work correctly there and get the right
path...
James