[waffle] [PATCH 07/12] waffle: support platform-specific information

[Emil Velikov]

On 25 April 2016 at 04:30, Frank Henigman <fjhenigman at google.com> wrote:
> On Sun, Apr 24, 2016 at 5:04 PM, Emil Velikov <emil.l.velikov at gmail.com> wrote:
>> On 24 April 2016 at 20:52, Frank Henigman <fjhenigman at google.com> wrote:
>>> On Sun, Apr 24, 2016 at 6:36 AM, Emil Velikov <emil.l.velikov at gmail.com> wrote:
>>>> On 21 April 2016 at 21:26, Frank Henigman <fjhenigman at google.com> wrote:
>>>>> On Fri, Jan 8, 2016 at 7:44 AM, Emil Velikov <emil.l.velikov at gmail.com> wrote:
>>>>>> On 6 January 2016 at 21:56, Frank Henigman <fjhenigman at google.com> wrote:
>>>>>>> Add a platform hook so platform-specific information can be included
>>>>>>> by waffle_display_info_json().
>>>>>>>
>>>>>>> Signed-off-by: Frank Henigman <fjhenigman at google.com>
>>>>>>> ---
>>>>>>>  src/waffle/api/waffle_display.c  | 10 +++++++++-
>>>>>>>  src/waffle/core/wcore_platform.h |  4 ++++
>>>>>>>  2 files changed, 13 insertions(+), 1 deletion(-)
>>>>>>>
>>>>>>> diff --git a/src/waffle/api/waffle_display.c b/src/waffle/api/waffle_display.c
>>>>>>> index 7abe2ef..d6988ac 100644
>>>>>>> --- a/src/waffle/api/waffle_display.c
>>>>>>> +++ b/src/waffle/api/waffle_display.c
>>>>>>> @@ -367,8 +367,16 @@ waffle_display_info_json(struct waffle_display *self, bool platform_too)
>>>>>>>
>>>>>>>      json_appendv(jj, "{", "generic", "{", "");
>>>>>>>      add_generic_info(jj, wc_self->current_context);
>>>>>>> -    json_appendv(jj, "}", "}", "");
>>>>>>> +    json_append(jj, "}");
>>>>>>>
>>>>>>> +    if (platform_too) {
>>>>>>> +        json_appendv(jj, "platform", "{", "");
>>>>>>> +        if (api_platform->vtbl->display.info_json)
>>>>>>> +            api_platform->vtbl->display.info_json(wc_self, jj);
>>>>>> The rest of waffle tends to set UNSUPPORTED_ON_PLATFORM if the backend
>>>>>> is missing the vfunc.
>>>>>
>>>>> I'm reluctant to set an error for something that isn't clearly an
>>>>> error (it might just be that a backend doesn't have platform-specific
>>>>> details) because it could add a burden to the user to distinguish this
>>>>> case from definite errors.
>>>> With all respect I have to disagree. Error checking/handling is not a
>>>> 'burden'. Even if some choose to ignore it that doesn't make it less
>>>> relevant ;-)
>>>>
>>>> Obviously things would be way easier/cleaner if things were split -
>>>> generic info vs platform specific (or even finer). As-is, with all of
>>>> them in one piece, no error-checking or UNSUPPORTED_ON_PLATFORM one
>>>> gets badly formatted/corrupted json. Thus the user has no way of
>>>> knowing if things failed for reason some, or the setup simply lacks
>>>> the information Y that they need.
>>>> </minirant>
>>>
>>> You never get corrupt json.  If the hook isn't implemented you get
>>> different json.  In v1 you'd get an empty platform section.  In v2 the
>>> platform section (e.g. "glx" or "egl") is omitted.  This is better
>>> because:
>>> - the json consumer is in the best position to decide what to do about
>>> a missing platform section - the api shouldn't decide it's an error
>>> - the caller of waffle_display_info_json() doesn't have to check
>>> waffle error state to know if there was a "real" error, they'll know
>>> by the NULL return
>> So when someone gets an OOM (even if it's during the platform
>> specifics) NULL is returned ? That does sound a bit strange, yet again
>> I would not read too much into it.
>
> That's true - when memory runs out the whole string is lost.  I guess
> that's the trade off for the simplicity of the error handling.  I'm
> quite happy with that trade off.
>
>> As long as others are happy go with it.
>>
>>> - we don't need to implement the hook in every back end
>>>
>> Need, perhaps not. It would be pretty good though ;-)
>
> I don't want to have to try to land code (even a stub) in back ends
> which I've never even compiled (and which would be a pain in the butt
> to set up for compiling).  Waffle needs continuous integration.  (:

I could wire up waffle's github repo with Travis-CI (for Linux/MacOS
builds) and Appveyor (for Windows ones), those depend on a lengthy
series that I have on the mailing list.
For Android I'm out of ideas.

https://lists.freedesktop.org/archives/waffle/2016-January/001292.html

- Resolves issues due to old wayland/gbm on CI systems.
- Removes the leaky waffle-tests, use cmocka.
- Allows the tests to be ran independently
- Adds GBM tests.

-Emil