[Ffmpeg-devel] dependency tracking improvements

[Aurelien Jacobs]

On Thu, 25 Jan 2007 00:02:53 +0100
Michael Niedermayer <michaelni at gmx.at> wrote:

> Hi
> 
> On Wed, Jan 24, 2007 at 11:25:53PM +0100, Aurelien Jacobs wrote:
> > On Tue, 23 Jan 2007 00:50:19 +0100
> > Aurelien Jacobs <aurel at gnuage.org> wrote:
> > 
> > > On Mon, 22 Jan 2007 09:08:00 +0000
> > > M?ns Rullg?rd <mru at inprovide.com> wrote:
> > > 
> > > > Diego Biurrun <diego at biurrun.de> writes:
> > > > 
> > > > > On Mon, Jan 22, 2007 at 01:21:21AM +0100, Aurelien Jacobs wrote:
> > > > >> On Mon, 22 Jan 2007 01:06:18 +0100
> > > > >> Diego Biurrun <diego at biurrun.de> wrote:
> > > > >> 
> > > > >> > On Mon, Jan 22, 2007 at 12:52:15AM +0100, Aurelien Jacobs wrote:
> > > > >> > > 
> > > > >> > > Here is a patch which further improve new Mans dependency tracking system.
> > > > >> > > It moves dependency information from configure to allcodecs.c/allformats.c.
> > > > >> > > Advantages are that it simplifies configure and that when you add a new
> > > > >> > > codec, you don't need to mess with configure. Just add a new line in
> > > > >> > > allcodecs.c and you're done.
> > > > >> > > Opinions ?
> > > > >> > > Can I commit this ?
> > > > >> > 
> > > > >> > Putting that info in a comment sounds very fragile to me.  This way
> > > > >> > compilation might fail if somebody changes a comment ...
> > > > >> 
> > > > >> Anyway, we already heavily rely on the structure of this file to parse
> > > > >> it with sed. It's already very easy to break.
> > > > >> But if you think using comments is bad, maybe we could do something
> > > > >> like this:
> > > > >> 
> > > > >> -  REGISTER_DECODER(AAC, aac);
> > > > >> +  REGISTER_DECODER(AAC, aac, libfaad);
> > > > >> 
> > > > >> Do you prefer it ?
> > > > >> IMO, it's as much fragile, and uglier than a comment.
> > > > >
> > > > > That's better and less fragile.
> > > > 
> > > > But it doesn't allow the current dependencies to be correctly expressed.
> > > 
> > > It does !
> > > Ok I admit that I didn't explained all the details of my idea.
> > > To express deps_any, just use:
> > > 
> > > -    REGISTER_ENCDEC (AMR_NB, amr_nb);
> > > +    REGISTER_ENCDEC (AMR_NB, amr_nb, , amr_nb amr_nb_fixed);
> > > 
> > > Let's see attached patch. Does it better suits your mind ?
> > 
> > Any thought about this patch ?
> 
> yes i dont like it
> 
> if it where something like
> 
> REGISTER_ENCDEC (AMR_NB, amr_nb);  //depends: amr_nb OR amr_nb_fixed
> 
> or
> 
> REGISTER_ENCDEC (AMR_NB, amr_nb);  //depends: amr_nb || amr_nb_fixed
> 
> or
> 
> REGISTER_ENCDEC (AMR_NB, amr_nb);  //requires: amr_nb || amr_nb_fixed
> 
> the reason for my dislike of 
> REGISTER_ENCDEC (AMR_NB, amr_nb, , amr_nb amr_nb_fixed);
> 
> is that a normal developer will not be able to make sense of it, she would
> have to look at REGISTER_ENCDEC() and then find in a comment there that the
> parameter passed to it actually has nothing to do with registering a codec
> but rather is somehow extracted by configure for checking the availability
> of dependancies IMHO code should be understandable without having to read
> too much other distant stuff

I agree and that's why my first proposition used a similar syntax (I prefer
your syntax, but it would be easy to addapt my patch for it).
Unfortunatly, some people don't like the idea of using comments to describe
dependencies.
So could we all agree on a way to describe dependencies outside of
configure ?
My voice goes to the comments way, but I'm open to anything else.

Aurel