[MPlayer-dev-eng] state of realvideo?

[Kees Cook]

On Tue, May 21, 2002 at 11:40:32AM +0200, Arpi wrote:
> > void (*rv20_init)(unsigned int * info, void ** handle);
> > unsigned int (*rv20_hive)(unsigned int * info, void * handle);
> > unsigned int (*rv20_free)(void * handle);
> > unsigned int (*rv20_custom)(unsigned int * info, void * handle);
> 
> are you sure in that handle is the last ?

Yeah.  Basically, what I did was I studied the disassembly for Init until
I could see what was being returned better, and then had realplay make
calls into my stubs.  On a successful Init, it would call Hive with that
returned handle value.  The handle stuff is just about the only thing I'm
sure of.  :)

> > unsigned int (*rv20_transform)
> >               (
> >                 unsigned int input_size,  // +8       read size?
> >                 unsigned int output_size,  // +0       write size?
> >                 unsigned char * input,// +10      pointer test, readable
> >                 unsigned char * output, // +14      pointer test, writable
> >                 void * handle  // +18 esi  source? handle?
> >               );
> 
> imho the order of parameters is reversed?

Well, I'm not sure about this one.  Handle is definitely last, and "input"  
and "output" are both NULL-tested, and "output" gets written to a little
(3 leading zeros?) before making lots of internal calls.  What I don't get
is that what I'm calling "*_size" don't really get examined, they just get
copied into the stack.

> imho most of the stuff in the info struct come from the .rm file header.
> at least all win32 codec i know do this - vfw, dshow and quicktime too

Is there a binary "grep"?  Something I could do like:

bgrep "0x02 0x01 0x0 0x01" whatever.rm

and it'd spit out where it found that byte pattern?



Also, I found the return values in the SDK includes.  In "pnresult.h" is a 
set of PN_E* #define's that match to error results returned from all the 
functions.

I'm using the latest .so files from the linux "realone" beta release.  It
looks like they got touched up a little: the __pure_virtual crap is gone
(possibly due to them using gcc 2.95 on the earlier stuff), and now Init
returns a value.

Oh, and I checked: while there are 4 Custom calls for every Transform, 
there are 25 Transforms a second getting called.  :)

-- 
Kees Cook                                            @outflux.net