Re: flat, no all-load for darwin

[Peter O'Gorman]

Geoffrey Keating wrote:

> Libtool was using -all_load to get all the members of a convenience
> archive.  Unfortunately, -all_load applies to *every* library in the
> link, so this would cause it to pull in everything from, say,
> libstdc++, causing huge binary bloat and occasionally link errors
> when a library is listed twice on the link line.
>
> Also, libtool is changed to use -flat_namespace for C++.  This is
> usually what you want, especially for libstdc++ itself.
>
> I've tested a version of this patch that applied to an earlier libtool
> (1.5.2, I think) by building all of GCC with it, and running its
> testsuite.
>
> I'm pretty sure I don't have write access, so could someone commit
> this for me?

Two problems with your patch:

1) -all_load does indeed cause bloat, but there is a reason I have not 
removed it thus far. If you are building fat, then -all_load will happily 
work with the fat archive, if not then the ar x stuff to get the archive 
members will fail. Removing -all_load will cause fat convenience libraries 
to stop working iirc. However now that I have a working fat unarchive func 
in cvs HEAD I plan to backport it to branch-1-5 in the near future and then 
remove the -all_load stuff (and maybe make a 1.5.8 release).

2) If I apply the -flat_namespace thing Ben Reed will be very angry with me 
as he'll no longer be able to build KDE. While you can add 
Wl,-flat_namespace to your LDFLAGS currently, and glibtool will pass it 
through and honour it, if you have specified -flat_namespace in the 
allow_undefined_flag, but you really want twolevel_namespace, it is 
impossible to add it to the flags:
ld: can't specify both -twolevel_namespace and -flatname_space
(note the flag name in the linker error, please file a bug at Apple for me :).

If you can explain why you need the flat_namespace stuff in the 
allow_undefined_flag maybe we can figure something out. I'll have to ask Ben 
 about KDE.

Sorry,
Peter
--
Peter O'Gorman - http://www.pogma.com