Re: [PATCH RFC 1/3] iio: Add symlink to triggers in the device's trigger folder

[Lars-Peter Clausen <lars@metafoo.de>]

On 05/12/2015 09:06 PM, Jonathan Cameron wrote:
> On 12/05/15 17:56, Lars-Peter Clausen wrote:
>> On 05/08/2015 05:11 PM, Jonathan Cameron wrote:
>>> On 16/04/15 05:01, Robert Dolca wrote:
>>>> This patch adds a new function called iio_trigger_register_with_dev
>>>> which is a wrapper for iio_trigger_register. Besides the iio_trigger
>>>> struct this function requires iio_dev struct. It adds the trigger in
>>>> the device's trigger list and saves a reference to the device in the
>>>> trigger's struct.
>>>>
>>>> When the device is registered, in the trigger folder of the device
>>>> (where current_trigger file resides) a symlink is being created for
>>>> each trigger that was registered width iio_trigger_register_with_dev.
>>>>
>>>>    # ls -l /sys/bus/iio/devices/iio:device0/trigger/
>>>> total 0
>>>> -rw-r--r--    1 root     root          4096 Apr 16 08:33 current_trigger
>>>> lrwxrwxrwx    1 root     root             0 Apr 16 08:33 trigger0 -> ../../trigg
>>>> er0
>>>>
>>>> This should be used for device specific triggers. Doing this the user space
>>>> applications can figure out what if the trigger registered by a specific device
>>>> and what should they write in the current_trigger file. Currently some
>>>> applications rely on the trigger name and this does not always work.
>>>>
>>>> This implementation assumes that the trigger is registered before the device is
>>>> registered. If the order is not this the symlink will not be created but
>>>> everything else will work as before.
>>>>
>>>> Signed-off-by: Robert Dolca <robert.dolca@intel.com>
>>> I was rather hoping we'd get a few more comments on this.
>>> In principle I like the idea, but it's new ABI and does make life
>>> a tiny bit more complex, so what do people think?
>>>
>>> Few trivial code comments inline.
>>
>> I don't think it adds more information. Both the trigger and the
>> device get registered for the same parent device, so you can already
>> easily find the trigger for a device by going to the parent device
>> and taking a look at the triggers registered by the parent device.
> I had the same thought. The question is whether the slightly gain in
> simplicity for userspace is worth it...  I'm undecided at the moment.

As you may have guessed by now I'm always quite conservative when it comes 
to introducing new ABI. Simply because we have to maintain it forever, the 
less stuff to maintain forever the better.

Hence I think all new ABI needs a compelling reason, e.g. like a improvement 
in performance. And of course this patch slightly simplifies things, but in 
my opinion not enough to justify a ABI extension. We can always find ways to 
simplify the interface, but the metric for ABI should be whether the 
simplification actually matters. In this case I don't think it does, finding 
the trigger for a device is not really hot-path. The amount of time saved 
will be disappear in the noise.

And in my opinion applications shouldn't directly use the low-level ABI but 
rather use middle-ware libraries/frameworks, like e.g. libiio, and that's 
where you'd hide the complexities of a operation.

- Lars