On Mon, Jun 19, 2017 at 11:59:11AM +0100, Richard W.M. Jones wrote:
On Mon, Jun 19, 2017 at 10:25:33AM +0200, Pino Toscano wrote:
> On Friday, 16 June 2017 16:58:53 CEST Richard W.M. Jones wrote:
> > On Fri, Jun 16, 2017 at 03:24:55PM +0200, Pino Toscano wrote:
> > > On Thursday, 15 June 2017 19:05:55 CEST Richard W.M. Jones wrote:
> > > > Those cleanups which only depend on libc, gnulib or libxml2 are
split
> > > > out into a separate common/cleanups directory.
> > > > ---
> > >
> > > IMHO a single cleanups.c source should be enough, otherwise it's
overly
> > > split...
> >
> > I think you do need to split it. The reason is that if the program
> > uses libcleanups.la but doesn't link to (eg) libxml2 then the link
> > will fail. We could either force everything to link unnecessarily to
> > libxml2 or we can split the object files so that the libxml2
> > dependency is never pulled in if the main program doesn't use it.
>
> This is for the libxml2 parts though. Also, I see that the cleanups are
> split from libutils, but then
> a) libcleanups is basically used where libutils is
> b) patch #14 makes the daemon link both libcleanup and libutils
> so IMHO the libc + gnulib cleanups could simply stay where they are,
> in libutils
OK, I'll combine gnulib cleanups back into libc cleanups.
I had to split gnulib cleanups out again into a separate object file.
There are some OCaml bytecode binaries that we build for testing where
it is difficult to statically link with -lgnu (and not necessary either).
Rich.
--
Richard Jones, Virtualization Group, Red Hat
http://people.redhat.com/~rjones
Read my programming and virtualization blog:
http://rwmj.wordpress.com
virt-builder quickly builds VMs from scratch
http://libguestfs.org/virt-builder.1.html