On 2 May 2014 11:14, Luca Barbato <[email protected]> wrote:
> On 02/05/14 19:57, Josh Allmann wrote:
>> On 2 May 2014 10:16, Hendrik Leppkes <[email protected]> wrote:
>>> On Fri, May 2, 2014 at 5:54 PM, Josh Allmann <[email protected]> 
>>> wrote:
>>>> Since the scaler is often auto-inserted in video filtergraphs,
>>>> this avoids a runtime error when filters are selectively enabled.
>>>
>>> I dunno, just because its commonly used at runtime shouldn't mandate
>>> including it all the time.
>>> If a user selectively disables it, its their own fault, isn't it. Its
>>> default on afterall.
>>>
>>
>> In that case, then why include fifos, buffers, ... ?
>>
>> If the scale filter is not unconditionally included, then it should
>> not be auto-inserted. The user should not have to be aware of lavfi's
>> plumbing -- that's the job of the build system.
>
> The problem is _quite_ annoying, if we do as you suggest we'd have
> people screaming at us because they do not want swscale and their use
> case is spoiled by the additional memory it requires for nothing.
>
> If we do not we'll have people complaining about having an unexpected
> behaviour that forces them to rebuild the whole thing and add another.
>

If pulling in libswscale is undesirable, then the scale filter should
not be auto-inserted.

Auto-inserted scale filters don't even work that well anyway. For
example, in the overlay filter, it is only used for colorspace
conversion -- the user still has to manually resize the overlay if the
dimensions are incorrect, which adds another pass to the processing.

What's more, unless the user sets a high debug level, he's probably
not even going to realize another conversion is being done within
lavfi (I didn't...). Principle of least astonishment: violated. This
desire for granular control is really at odds with the automagic
nature of filter auto-insertion.

Josh

> We have _suggest to declare this kind of dependency but I'm not sure it
> would help in this specific case.
>
> Probably warning at the end of configure about suggested-but-disabled
> filter isn't enough =\
>
> lu
> _______________________________________________
> libav-devel mailing list
> [email protected]
> https://lists.libav.org/mailman/listinfo/libav-devel
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to