Re: [PATCH 28/35] acpi: pvpanic-isa: use AcpiDevAmlIfClass:build_dev_aml to provide device's AML

[Igor Mammedov]

On Wed, 18 May 2022 12:29:25 -0400
"Michael S. Tsirkin" <mst@redhat.com> wrote:

> On Tue, May 17, 2022 at 10:13:51AM +0200, Gerd Hoffmann wrote:
> > That problem isn't new and we already have a bunch of aml_* stubs
> > because of that.  I expect it'll work just fine, at worst we'll
> > have to add a stub or two in case some calls are not covered yet.  
> 
> Right but adding these stubs is a bother, we keep missing some.
> If possible I'd like the solution to be cleaner than the status quo.
> Is adding a wrapper instead of setting a method directly such
> a big problem really?

Stubs are the bother but not much compared to alternatives.
I can't recall missing stubs recently (it's hard to miss them
as it's build time failure that won't pass CI).

However wrapper would introduce ifdeffenry instead of a stub.
And my understanding was that it's not acceptable and stubs are
what consensus approach is/was to eliminate/minimize ifdefs
in the code.

Also adding wrapper won't help anything, we also need to
decouple AML code into separate source files to avoid
dependency on AML routines and that is a bigger crunch
that includes not only new source files but spreading
CONFIG_APCI all over the tree, so I'm not sure if end
result won't be worse compared to stubs. Stubs are not
the cleanest ways around the issue but they would be
simpler to maintain in the end.