Re: [PATCH v7 00/11] kconfig: Add support for conflict resolution

[Ole Schuerks <ole0811sch@gmail.com>]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 2269 bytes --]

On Feb 11, 2025 at 01:46 AM, Masahiro Yamada wrote:
> On Tue, Feb 11, 2025 at 12:43 AM Luis Chamberlain <mcgrof@kernel.org> wrote:
>>
>> On Mon, Feb 10, 2025 at 02:00:52PM +0900, Masahiro Yamada wrote:
>>> Thanks for this, but I have no plans to merge the SAT solver.
>>>
>>> The reason is that my future plan is to move toolchain selection
>>> to the Kconfig stage instead of specifying it statically from the command line.
>>
>> That makes sense.
>>
>>> This approach was suggested by Linus [1], and to achieve that,
>>> the shell evaluation must be dynamically re-evaluated [2].
>>
>> Sure.
>>
>>> The SAT solver would likely conflict with this plan. At least due to the
>>> significant amount of additional code, which would be an obstacle.
>>
>> I can't see how the toolchain selection, if set on Kconfig can't be
>> leveraged later to enable / disable the SAT solver, however I can
>> see the amount of code shuffling incurred to be an extra hurdle to
>> address and a preference to leave that for later.
>>
>> In other words, I susepct it is still possible to evaluate to
>> add support for the SAT solver post toolchain kconfig integration.
>>
>> Thoughts?
>
>
> It depends on how the dynamic shell evaluation is implemented.
> This is not limited to bool/tristate, but SAT solver only works for
> those two types.


I just wanted to clarify, configfix already supports string/int/hex values.
Configfix uses SAT-variables that correspond to a kconfig-symbol being set
to a specific value (e.g., one such variable would be `NR_CPUS=1`). The
sample values for which such a variable is created are currently collected
from default and range properties.

This collecting of sample values is going to be the main challenge of
integrating configfix with dynamic shell evaluations. I believe the
integration is feasible, but even if serious problems should arise, a
decent alternative is to have configfix simply ignore the possibility of
changing the symbol values that are inputs to or results of dynamic shell
evaluations. That should amount to roughly the same capabilities as the
current version without dynamic shell evaluations.

Best regards,
Ole Schuerks