Re: [PATCH] tests: qtest: bios-tables-test depends on the unpacked edk2 ROMs

[Paolo Bonzini]

On 23/09/21 12:44, Peter Maydell wrote:
> On Thu, 23 Sept 2021 at 11:32, Paolo Bonzini <pbonzini@redhat.com> wrote:
> > On 23/09/21 12:00, Philippe Mathieu-Daudé wrote:
> > > See also "blobs: Only install required (system emulation) files":
> > > https://lore.kernel.org/qemu-devel/20210323155132.238193-1-f4bug@amsat.org/
> >
> > Nice and makes a lot of sense.  Regarding "there is probably a nicer way
> > to do this with Meson", I would do without all the variables and do
> > something like
> >
> > foreach target : target_dirs
> >     if target in ['...']
> >       blobs_ss.add('...')
> >     elif target in ['...']
> >       blobs_ss.add('...')
> >     endif
> > endforeach
> >
> > directly in pc-bios/meson.build.
>
> Is it possible also to have meson handle the "symlink the blob from
> the source dir to the build dir" which currently configure is doing ?

Yes, though I would have to check the details on how to do that best 
(for example whether to keep them at configure time or move them to make).

By the way, I think a lot of the DIRS in configure are not needed 
anymore; meson would create them anyway and they're not needed by the 
LINKS loop below.

> That would mean we could avoid having effectively two lists of blobs.
> (Somebody would need to cross-check that all the blobs the wildcards in
> configure are linking in are listed in the meson list.) I guess
> the question is whether other parts of the build system assume those
> links already exist (ie they don't explicitly declare a dependency
> on the existence of the blob and would need to change to do so).

Yeah, that's also why in my patch I didn't bother adding the roms 
dependency to bios-tables-test only.  It's less prone to failure if 
they're just built before any qtest is run.

Paolo