[Libva] [PATCH v2 3/5] API: backend: rename dri_state to drm_state.

[Gwenole Beauchesne]

Hi,

2012/7/10 ykzhao <yakui.zhao at intel.com>:
> On Sun, 2012-07-08 at 23:43 -0600, Gwenole Beauchesne wrote:
>> Hi,
>>
>> 2012/7/9 ykzhao <yakui.zhao at intel.com>:
>> > On Sun, 2012-07-08 at 23:06 -0600, Gwenole Beauchesne wrote:
>> >> 2012/7/9 ykzhao <yakui.zhao at intel.com>:
>> >> > On Sun, 2012-07-08 at 22:28 -0600, Gwenole Beauchesne wrote:
>> >> >> > There is no problem when the variable of dri_state is renamed to
>> >> >> > drm_state.
>> >> >> > But it seems a little confusing as the patch 2 already defines the
>> >> >> > structure type of drm_state.
>> >> >>
>> >> >> Both patches could be merged altogether, if this is what you suggest.
>> >> >
>> >> > No. What I mean is that the name seems a little confusing. And the
>> >> > readability of ctx->drm_state is decreased as the patch 2 defines the
>> >> > data structure of drm_state.
>> >>
>> >> What are you really talking about? And what's your suggestion then?
>> >
>> > It is OK for me to rename the "dri_state" to drm_state. But to keep the
>> > readability of "ctx->dri_state/drm_state", I suggest that the drm_state
>> > defined in patch 2 can use another name. (For example: drm_base_state)
>>
>> I am sorry but I still don't see your point and how this affects
>> readability. drm_state is the base class for dri_state, which is a
>> DRI-authenticated drm_state + data for its housekeeping. It's pretty
>> clear already. Your suggestion actually decreases readability when it
>> comes to the VA/DRM case whereby you don't care of extra data, so the
>> base drm_state is already sufficient.
>
> Of course your patch is ok. The readability I mentioned is the
> following:
>      >The drm_state is defined in the structure of VADriverContext,
> which is parsed as struct dri_state.
>      >The patch 2 defines the structure of struct drm_state.
>     In such case maybe the "drm_state" defined in dri_state will be
> misregarded as the type of "struct drm_state".

In VA/X11 implementation, you can obviously use
VADriverContext.drm_state as a struct drm_state or a struct dri_state.
So, that's still valid, and desired for common code. However, the most
relevant object there remains a struct dri_state, which is allocated
on the libVA side, so there is no further ambiguity either.

I believe we could also document further the structures. Let me send
another iteration.

> Not sure whether the below is helpful to improve the readability?
>     a. the struct drm_state in patch 2 is renamed as drm_base_state
>     b. add one typedef definition to define the struct dri_state as
> drm_state.

The problem with (b) is that you see lots of dri related definitions,
and renaming it to drm_state could be confusing.

Regards,
Gwenole.