[FFmpeg-devel] [RFC] Dropping the rgb2rgb interface (was: [PATCH] make building swscale rgb template conditional)

[Ramiro Polla]

On Sun, Sep 5, 2010 at 12:07 PM, Michael Niedermayer <michaelni at gmx.at> wrote:
> On Sun, Sep 05, 2010 at 03:22:06AM -0300, Ramiro Polla wrote:
[...]
>> Using global function pointers is obviously a bad idea. I'd suggest we
>> drop exposing rgb2rgb but I faintly remember Michael being against
>> that because some programs already use that interface and we'd have to
>> bump major if we want to remove it, and also because it's useful to
>> have a simpler alternative to get rgb2rgb functions than going through
>> the swscale interface. Also I'd like to point out that on shared
>> builds these global function pointers no longer work since only
>> swscale_*, sws_*, and ff_* stuff is exported.
>>
>> If we must keep the rgb2rgb interface (not the global pointers, these
>> should definitely go away at next major bump), I suggest we either:
>
>> 1- create a simpler function like sws_getRGBContext that only takes
>> arguments relative to rgb2rgb (like src and dst pixel format), and
>> that can be used with sws_scale()
>
> i see no sense in this ...

The rgb2rgb functions could be obtained with fewer parameters (for
example there is no need to specify the algorithm for 1:1
conversions).

>> 2- create an rgb2rgb context which the user can initialize and then
>> use rgb->rgb24tobgr32() and such.
>
> maybe i dont know
>
>
>>
>> Michael, how do you suggest we proceed regarding the rgb2rgb interface?
>
> yes, what do the users of rgb2rgb think?
> do they want to keep an alternative way to access it or can the main sws
> api be used?

I searched for quite some time to try and find where rgb2rgb is being
used. I mainly used google codesearch. The search strings tried to
exclude all internal uses of FFmpeg and MPlayer, and were along the
lines of:
"rgb15tobgr15 -file:(mpng|mencoder|vf_.*|vo_.*|utils|vo_vesa|ws|skinload|cs_test|colorspace-test|swscale).c
-file:lumiere-sections.txt -file:swscale-0.def -file:index.sgml
-file:rgb2rgb(_template|_1)?.(h|c|sgml)"
"sws_rgb2rgb_init
-file:(swscale-example|mga_.*|mpng|mencoder|vf_.*|vo_.*|utils|vo_vesa|ws|skinload|cs_test|colorspace-test|swscale).c
-file:lumiere-sections.txt -file:swscale-0.def -file:index.sgml
-file:rgb2rgb(_template|_1)?.(h|c|sgml)"

There are some programs that at some point used rgb2rgb directly:
xvidcap has had it commented out forever
veejay no longer uses it
avidemux no longer uses it
xbmc cargo cults initialization of sws_rgb2rgb_init() unnecessarily

Some other programs are either forks of mplayer (mplayer-ce,
mplayerxp, syllable, warpvision) or copied some specific code inside
their libraries (for example crusher has interleaveBytes).

All other hits for any of those global pointers and sws_rgb2rgb_init()
are either inside FFmpeg/MPlayer or are in files listing the functions
from rgb2rgb (like .def or .txt files). rgb2rgb.h is not even an
installed header. I've sent a patch to mplayer-dev-eng to remove the
last uses of rgb2rgb. I think it's safe to assume that no one uses
rgb2rgb anymore.

Does anyone have anything against dropping the rgb2rgb "external"
interface (direct access to global function pointers and some other
functions in rgb2rgb.h). Please speak up now.

Ramiro Polla