On Fri, Aug 04, 2017 at 06:13:42PM +0100, Daniel P. Berrange wrote:
> Currently any mockable functions are marked with attributes
> noinline, noclone and weak. This prevents the compiler from
> optimizing away the impl of these functions.
> 
> It has an unfortunate side effect with the libtool convenience
> libraries, if executables directly link to them. For example
> virlockd, virlogd both link to libvirt_util.la  When this is
> done, the linker does not consider weak symbols as being
> undefined, so it never copies them into the final executable.
> 
> In this new approach, we stop annotating the headers entirely.
> Instead we create a weak function alias in the source.
> 
>    int fooImpl(void) {
>       ..the real code..
>    }
> 
>    int foo(void) __attribute__((noinline, noclone, weak, alias("fooImpl"))
> 
> If any functions in the same file call "foo", this prevents the
> optimizer from inlining any part of fooImpl. When linking to the
> libtool convenience static library, we also get all the symbols
> present. Finally the test suite can just directly define a
> 'foo' function in its source, removing the need to use LD_PRELOAD
> (though removal of LD_PRELOADS is left for a future patch).
> 
> Signed-off-by: Daniel P. Berrange <berra...@redhat.com>

Self-NACK.

This breaks on OS-X because the linker doesn't support 'alias' or
'weak'.  For that matter it doesn't support LD_PRELOAD either, so
we need to avoid wrapping the functions on this platform.


Regards,
Daniel
-- 
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|

--
libvir-list mailing list
libvir-list@redhat.com
https://www.redhat.com/mailman/listinfo/libvir-list

Reply via email to