Re: [PATCH 1/2] system: Restrict libSDL CPPFLAGS to vl.c

[Paolo Bonzini]

On 12/12/24 14:18, Richard Henderson wrote:
> On 12/12/24 03:26, Philippe Mathieu-Daudé wrote:
> > Only vl.c includes libSDL headers.
> > No need to pass them to all system_ss[] files.
> >
> > Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
> > ---
> >   system/meson.build | 5 ++++-
> >   1 file changed, 4 insertions(+), 1 deletion(-)
> >
> > diff --git a/system/meson.build b/system/meson.build
> > index 4952f4b2c7d..f7e2c8b826f 100644
> > --- a/system/meson.build
> > +++ b/system/meson.build
> > @@ -23,8 +23,11 @@ system_ss.add(files(
> >     'runstate-hmp-cmds.c',
> >     'runstate.c',
> >     'tpm-hmp-cmds.c',
> > +), libpmem, libdaxctl)
> > +
> > +system_ss.add(files(
> >     'vl.c',
> > -), sdl, libpmem, libdaxctl)
> > +), sdl)
> >   if have_tpm
> >     system_ss.add(files('tpm.c'))
>
> I'm sure Paolo will correct me, but I don't think this does what you 
> think it does.  I believe this has no change at all.

No need to correct anything! :)

> The presence of sdl within a *particular* source_set.add() call is 
> immaterial.  If the condition is true (and here, because the condition 
> is missing it is true), all of the files and dependencies get lumped 
> together.  In the end, everything gets copied into common_ss.

Yep.  You can have different flags, including library CFLAGS/CPPFLAGS, 
for specific vs system (the former ends up in the libqemu-*.a.p 
directory, vs libcommon.a.p for system), or for those sourcesets that 
are used also by user-mode emulation (IIRC libhwcore.a.p) or by tests 
(the various directories libio.a.p, libmigration.a.p).  However, here 
there is no effect at all.

If you want to do this as a cleanup to keep the libraries close to their 
users, I have no objection, but please adjust the commit message and 
squash the two patches together.

Paolo

> Both of these patches only alters the order of the items in the link 
> ordering.
>
>
> r~