Re: [Qemu-ppc] [RFC PATCH 1/5] ppc64: Express dependencies of 'pseries' and 'powernv' machines with kconfig

[Paolo Bonzini]

On 30/01/19 11:02, Thomas Huth wrote:
>> I think the two approaches are more or less equivalent, but
>> "#CONFIG_FOO=n" has a small advantage when the feature has a build-time
>> dependency, such as CONFIG_MILKYMIST_TMU2's dependency on OpenGL.  In
>> that case,
>>
>>    CONFIG_MILKYMIST_TMU2=y
>>
>> would report a contradiction if OpenGL is not available at build time, while
>>
>>    default y
>>    ...
>>    #CONFIG_MILKYMIST_TMU2=n
>>
>> would not.
> I somehow dislike the idea of adding lines that are commented out by
> default. For example, if someone later renames or removes the config
> switch in the Kconfig files, the comment could stay there without being
> noticed, while a CONFIG_XXX=y would cause a clean build failure if the
> option is not available anymore.

I agree.  On the other hand I don't see another solution for the
MILKYMIST_TMU2 and QXL cases, and consistency is a plus too.

How often are config switches renamed or removed, especially
user-visible optional devices though?  I see CONFIG_MEM_DEVICE,
CONFIG_PCI_APB, CONFIG_LIBDECNUMBER, CONFIG_XEN_I386,
CONFIG_MIPS_BOSTON, CONFIG_PIIX_PCI, CONFIG_ISA_MMIO, CONFIG_ICC_BUS
over the last few years but none probably would have been optional
devices in default-configs/ but rather in Kconfig files.

(Long term default-configs/ should probably be built automatically,
possibly by adding a description field to the Kconfig files, and that
would also ameliorate the issue you describe).

Paolo