Re: [PATCH] hw/pci-host/grackle: Verify PIC link is properly set

[BALATON Zoltan]

On Tue, 20 Oct 2020, Markus Armbruster wrote:
> Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk> writes:
>
> > One thing I have thought about is being able to mark a link property
> > as mandatory so if a value hasn't been set before realize then you
>
> A non-null value, I presume.

Do you mean something like distinguish between NULL and INVALID_VALUE 
where setting the latter as initial value means property is mandatory?

> > receive a fatal error. This would be for cases like this where 2
> > internal devices are connected together without any formal interface,
> > i.e. in cases where -device wouldn't work anyway.
>
> Moves the check from code one step closer to data: from the realize
> method to the object_property_add_link() call.
>
> I like doing things in data, because data is easier to reason about than
> code.

Except when object initialisation is scattered around in boiler plate code 
as in QEMU where it may not be obvious why a realize method fails due to 
something set/documented in a class_init method elsewhere, whereas an 
assert in the realize method is quite self evident and would document the 
same requirements.

Regards.
BALATON Zoltan