msvs support (was: static lib containing backends)

[Ralf Wildenhues]

Hi Peter,

* Peter Ekberg wrote on Sat, Feb 11, 2006 at 09:15:11PM CET:
> On Thu, Feb 09, 2006 at 05:08:59PM +0100, Ralf Wildenhues wrote:
> > 
> > Last time I tested Peter's patch (see an earlier post of mine in this
> > thread), it worked.  Yes, it's been quite a while.  No, I don't remember
> > all the gory details.
> 
> I am not relying on libltdl for my projects, so the only testing
> I've done regarding libltdl are those in the test suite. As far
> as I remember there is some trouble with dllexport/import that
> is a bit hairy. I think the user of libltdl has to know how
> libltdl is going to be linked with the user code, in order for
> the user code to add dllimport (or not). I.e. there is no
> way to write a libltdl header that will always do the right thing
> (without user help), since it is not known at compile time how the
> user will link with libltdl. (hmm, on reading through the message
> I'm not satisfied with this wording. oh well)

I believe that's correct.  Howard reported this.  We agreed that
removing non-function interfaces would solve this.

> > Did I just hear you volunteer for testing?  ;-)
> 
> I think I remember Ralf saying something about a list of tests that
> are needed in order for the patch to be commitable. Is this something
> I can do, or is a second opinion required/desireable here? Ralf,
> can you please provide that list?

Sure, I can try.  The patch needs some cleanup.  I'll reply on
libtool-patches with inline comments about the patch.
Then the patch needs testing with
- MSVC (cl) on MinGW
- MSVC (cccl) on MinGW
- gcc on MinGW
- gcc on Cygwin

There should not be any regressions (except maybe the ones that stem
 from not applying the testsuite changes).

BTW, there is at least one largish software package that already uses
your patch to provide its MSVC port: with cccl and an in-tree libltdl.
So I don't expect too many issues.

> I have kept quiet (and haven't worked on the patch either for that
> matter) in order to not destabilize the pending release, but there's
> been lots of water under the bridges...

Oh well.

> I believe the patch does not touch any non-win32 stuff and things
> still work for MinGW and Cygwin as far as I can tell. The only
> possible trouble is that the cccl support might be broken by the
> patch (but I don't know if cccl in fact works at the moment).

It should not be broken, though: it's very useful.

Cheers,
Ralf