> Curt Sampson wrote:
> > In other words, in my limited scope of vision at this time I don't
> > see why we can't just add i386-type shared libraries to the alpha
> > ECOFF format files and solve all of our problems in one fell swoop.
> > Not to mention that we'll avoid the `is this a NetBSD ELF file or
> > a Digital UNIX ELF file?' question when we load a binary. (ELF
> > doesn't appear to have any OS type; just CPU type.)
> 
> I don't know the answer to this.

It more or less boils down to something like:

if you're going to add support for shared libraries for a given
executable format, and that format already defines the shared library
format, then you should use it.


I don't particularly _like_ all of the aspects of ELF or ELF shared
libraries, but it's better (in the long term) than going with ECOFF,
and i don't particularly feel inclined to create my own executable
format...  8-)


A.out has historically had no 'well-defined' shared library scheme.
Paul Kranenburg took the closest thing to that, the Sun method, and
implemented something pretty close to that...


> I just accepted that Chris would have a good reason for doing all
> that work on gnu tools (knowing how he *loves* gnu's copyleft 8-).

I've actually done relatively little work on the toolchain, since that
code is almost completely opaque to me.  I've mostly been testing the
code written by others (Richard Henderson, to support Linux on the
Alpha; and Matt Thomas, who did the NetBSD/alpha shared lib changes),
as well as writing whatever changes to the configuration headers,
etc., that are appropriate.



chris