[FFmpeg-devel] [PATCH]lavfi/minterpolate: Allow bigger resolutions if SIZE_MAX is big

[Carl Eugen Hoyos]

> Am 31.03.2020 um 12:24 schrieb Anton Khirnov <anton at khirnov.net>:
> 
> Quoting Carl Eugen Hoyos (2020-03-31 11:56:44)
>>> Am Di., 31. März 2020 um 11:18 Uhr schrieb Anton Khirnov <anton at khirnov.net>:
>>> 
>>> Quoting Carl Eugen Hoyos (2020-03-28 13:54:22)
>>>> Hi!
>>>> 
>>>> Attached patch allows to work-around ticket #7140, tested on a system
>>>> with a lot of memory.
>>> 
>>> This looks very ad-hoc.
>> 
>> Is there another part of FFmpeg that rightfully allocates that much memory?
> 
> Something decoding really big images?

Are you sure?
A quick calculation indicated a max allocation of 1500MB to me, what did you think of?

> 2GB is not really that much these
> days.

Please describe an example.

>> 
>>> The right thing to do is fix arbitrary limits in
>>> av_malloc_array().
>> 
>>> That INT_MAX there looks wrong.
>> 
>> I wonder if it is a good idea that demuxers and decoders cannot allocate
>> random amounts of memory...
> 
> I see no valid reason why this specific function should have this
> specific limit, while other memory allocation functions have different
> limits.

I know of only one other function in FFmpeg that does this, but I believe it already listens to max_alloc.

> Beyond that, I don't think we should have any arbitrary limits
> on what allocations are reasonable and what are not. There are many
> other --- more appropriate --- mechanisms for limiting memory usage.

I tend to disagree, especially as this does not affect any normal usage.

Carl Eugen