Re: [PATCH 1/3] util & iothread: Introduce event-loop abstract class

[Stefan Hajnoczi]

On Mon, Feb 28, 2022 at 08:05:52PM +0100, Nicolas Saenz Julienne wrote:
> On Thu, 2022-02-24 at 09:48 +0000, Stefan Hajnoczi wrote:
> > On Mon, Feb 21, 2022 at 06:08:43PM +0100, Nicolas Saenz Julienne wrote:
> > > diff --git a/qom/meson.build b/qom/meson.build
> > > index 062a3789d8..c20e5dd1cb 100644
> > > --- a/qom/meson.build
> > > +++ b/qom/meson.build
> > > @@ -4,6 +4,7 @@ qom_ss.add(files(
> > >    'object.c',
> > >    'object_interfaces.c',
> > >    'qom-qobject.c',
> > > +  '../util/event-loop.c',
> > 
> > This looks strange. I expected util/event-loop.c to be in
> > util/meson.build and added to the util_ss SourceSet instead of qom_ss.
> > 
> > What is the reason for this?
> 
> Sorry I meant to move it into the qom directory while cleaning up the series
> but forgot about it.
> 
> That said, I can see how moving 'event-loop-backend' in qom_ss isn't the
> cleanest.

Yes, qom/ is meant for the QEMU Object Model infrastructure itself, not
for all the QOM classes that rely on it.

> So I tried moving it into util_ss, but for some reason nobody is calling
> 'type_init(even_loop_register_type)'. My guess is there's some compilation
> quirk I'm missing.

Maybe the issue is that libqemuutil.a (util_ss) object files are linked
on demand. If there are no symbol dependencies in the main QEMU code to
event-loop.o then it won't be linked into the executable. That may be
why event_loop_register_type() isn't being called (it's set up by an
__attribute__((constructor)) function in event-loop.o so it doesn't help
create a symbol dependency).

> Any suggestions? I wonder if util_ss is the right spot for 
> 'event-loop-backend'
> anyway, but I don't have a better idea.

What Paolo suggested sounds good: move event-loop.c next to iothread.c
in the top-level source directory.

Stefan

signature.asc
Description: PGP signature