Re: More on 32bit/64bit ABIs under Linux

[Geoff Keating <geoffk at ozemail dot com dot au>]

> Cc: libc-hacker@cygnus.com, egcs@egcs.cygnus.com
> From: Alexandre Oliva <oliva@dcc.unicamp.br>
> Date: 23 Apr 1999 15:41:25 -0300
> User-Agent: Gnus/5.070083 (Pterodactyl Gnus v0.83) XEmacs/20.4 (Emerald)
> 
> On Apr 22, 1999, hjl@lucon.org (H.J. Lu) wrote:
> 
> > The reason I suggested ${libc_version} is it tells me which C
> > library is compiled against. In many cases, this information
> > is very useful. It is just my opinion.
> 
> Doesn't ldd give you this information already?
> 
> IMO, introducing sonames of dependency libraries into dependent ones
> *is* a good idea, because it solves the problem of libc migration in a
> much cleaner way than the ld.so hack introduced in the migration from
> libc5 to 6.  The point is that it's not just libc that can change,
> affecting relinks of all libraries that depend on it: any other
> library that may have other libraries depending on it may cause the
> same kind of problem that the libc5->6 upgrade did.

I don't see why libc should be special, though.  The sonames should
include the names of every library linked to.  What happens when X12R1
comes out?  Or when libdb changes its interface (which happened last
year)?


Anyway, it doesn't seem likely that there will be a change to
libc.so.7 for Linux any time in the forseeable future.  It's simply
too much pain, and it can be avoided using symbol versioning.

-- 
Geoffrey Keating <geoffk@ozemail.com.au>