[Mesa-dev] [PATCH] android: amd/common: fix sid_tables.h generation rules

[Rob Herring]

On Fri, Aug 11, 2017 at 10:10 AM, Mauro Rossi <issor.oruam at gmail.com> wrote:
> 2017-08-11 16:23 GMT+02:00 Rob Herring <robh at kernel.org>:
>> On Fri, Aug 11, 2017 at 9:02 AM, Mauro Rossi <issor.oruam at gmail.com> wrote:
>>> Current generation rules rely on LOCAL_PATH variable,
>>> which may be undefined when dependencies are expanded;
>>> move to using MESA_TOP variable to define sid_tables.py script path
>>
>> I count roughly 67 occurrences of pointing to python scripts using
>> LOCAL_PATH. Presumably they all need to be fixed or this isn't really
>> the problem.
>>
>>> Fixes the following building error:
>>>
>>> external/mesa/src/gallium/drivers/radeonsi/si_debug.c:30:10: fatal error: 'sid_tables.h' file not found
>>>          ^
>>> 1 error generated.
>>>
>>> Fixes: 730574c58e "android: amd/common: add support for libmesa_amd_common"
>>
>> Why do I not see this error?
>
>
> I was also suprised to see the error,
> it started to appear persistently when building nougat-x86 from scratch.
>
> As a similar case I saw this one:
> https://cgit.freedesktop.org/mesa/mesa/commit/?id=c1a29e104cc585ad3219b12d09f532a129d68dad
>
> and in general I empirically saw it is unsafe to use $(LOCAL_PATH) in
> generated files dependencies rules.
> Chih-Wei may know better the reason.

The discussion on this concluded that LOCAL_PATH as a rule dependency
is okay. LOCAL_PATH in the recipe for the rule is not.

>
> Added him in Cc:
>
> Mauro
>
> PS: If it is a false positive and not needed in 17.2 and mesa-dev
> please Rob, Chih-Wei just tell me,
> but in any case 17.1 branch requires to add:
>
> +LOCAL_STATIC_LIBRARIES := libmesa_amd_common
>
> in https://cgit.freedesktop.org/mesa/mesa/tree/src/gallium/drivers/radeonsi/Android.mk?h=17.1

Not sure, but probably needed as radeonsi was not in good shape in 17.1.

Rob