On Wed, 27 Sep 2000, Frederick Bruckman wrote:

> On Wed, 27 Sep 2000, Bill Studenmund wrote:
> 
> > Plese immediatly revert this. You can NOT just rip out shared libraries.

Note: deleting a shared library is effectivly bumping its major
number. You didn't change the version # on the package which needs to
happen in such cases.

> The shared libraries have been ripped out on ELF since 1.19, a year
> ago this week.

Shit. This really sucks as I had them working when I was compiling the
tools themselves.

> > Also, plese actually talk to someone who has worked on this before making
> > changes. We actually have shared libraries for xforms in our packages, and
> > if anything needs ripping out, it's the sillyness of trying to make shread
> > libraries from static ones, not the shared library itself.
> 
> Please refer to the discussion appended to the PR. It is very unfair
> of you to accuse me of taking unilateral action without doing so.

Did you talk to anyone who actually made one of the distribution files?
You talked to a lot of people, yes. But no one who's beein in the source.

> The only "on-the-record" motivation I can find for the orginal libtool
> sillyness is a discussion on tech-pkg, which you were a party to,
> which suggested using it to move the forms shared library version away
> from the xforms package version.

Right. And I support the idea of our library # being different from
Zhao's. That way we can bump major numbers if needed.

I was not aware that this change would be implimented by taking the
static library appart to make the dynamic one. If I'd known that, I'd have
said, "STOP." I realize you are not the person who did this. But you are
treating a very real problem in what I think is the wrong manner - we
should just kill the libtool part rather than delete the libraries.

Please test that. It shouldn't take more than an evening to test, and it
will either work or not. If ripping out the libtool part fixes things,
a.out has shared libs. Otherwise we are where we were - no shared libs
anywhere.

> "libtool" complicates matters, but it's not even the problem. The
> problem is simply that -lforms should really pull in -lGL. [This would
> have been obvious if xforms source were available, but in fact, since
> the source for gl.c is part of the distribution... it _is_ obvious.]
> _In_ _fact_, libtool could be used to add a run-time dependency on
> -lGL, and also add DEPENDS=Mesa-* to the xforms package, but this has
> an obvious drawback, since evidently none of the present packages
> which require xforms use any Mesa functions (else they wouldn't even
> link static without -lGL).

I'm still double-checking, but I thought I had it working w/o -lGL on
ELF. Some programs had problems, some didn't.

> The remaining alternative would be to duplicate the i386/a.out trick
> of not resolving all symbols in shared libraries, on ELF and other
> a.out platforms. I'm not sure that would be a good thing.

That's not for this discussion, but I think it would be a good
thing. Other OS's do it. You just have those symbols resolve to something
which sends a sigbus to the program or something like that.

Take care,

Bill