On 09/13/2011 12:00 AM, Donnie Berkholz wrote:
> On 21:57 Mon 12 Sep , Michał Górny wrote:
>> Right now, autotools-utils.eclass punts .la files only with
>> USE=-static-libs. We'd like to broaden the range of it and strip .la
>> files when they are not necessary for static linkage as well.
>> The following patch introduces an initial support for that. It assumes
>> that the .la file can be removed if the library is mentioned in any of
>> pkg-config files installed by the package, or if doesn't specify any
>> dependency libs nor linker flags.
> If I understand correctly, this will break for any packages that don't
> use pkg-config to link. The maintainers will manually need to add
> pkg-config calls to the ebuilds of anything that could statically link
> against a library using only libtool and not pkg-config. Is that
Yes, seems accurate.
I can think of 'export PKG_CONFIG="$($(tc-getPKG_CONFIG) --static)' or
something like 'export FOO_LIBS="$($(tc-getPKG_CONFIG) --libs --static
foo)"' to accomplish getting static flags from an ebuild using
toolchain-funcs.eclass if required.
Or they do it like lvm2 and cryptsetup at upstream level and add
support for statically linking the tools in the build-system.
The .la files are not helping packages not using libtool in any case,
for example, those using cmake as build-system.
And I've yet to see a real, in portage residing, example of where this
would really break anything and when I will, I'll gladly help migrating
it to the example mentioned above... Overall, corner cases that can be
easily worked around, yet punting the *harmful* .la files.