[linux-dvb] Re: refactoring

[Holger Waechtler <holger@qanu.de>]

Andrew de Quincey wrote:

> On Wednesday 06 Oct 2004 15:50, Holger Waechtler wrote:
>  
> > Andrew de Quincey wrote:
> >    
> > > > Yeah, was pondering on whether you was arguing with me or against me :)
> > > > The #ifdef bloat won't be that much of a problem will it? The code in
> > > > question will just be a function with a #ifdef block for each frontend?
> > > > If there is only one frontend supported by an adapter, I can see no harm
> > > > in 'select'ing it though.
> > > >        
> > > Hi, I've just been implementing this. There is a problem with doing it
> > > using "default".
> > >
> > > If a user enables one DVB card, saves the config, and compiles the kernel,
> > > everything will be fine. However, if they then enable another DVB card
> > > requiring a different frontend, things will go wrong. The reason is that
> > > default does not work if a value has already been set for a kernel
> > > option... so when the first card was selected, all non-essential
> > > frontends will be marked as "not selected". Default will not turn them on
> > > again for the second card.
> > >
> > > I'd rather avoid this confusion completely - most users won't care that
> > > three frontend modules are being loaded when only one is really
> > > necessary. What I'm going to do is turn on all frontends by default. Only
> > > the relevant ones will actually get loaded by the cards however, so we
> > > end up in the same situation as before.
> > >
> > > Obviously if the user manually turns one off, they can be assumed to be
> > > knowing what they're doing, and can sort out any problems themselves.
> > >      
> > sounds good so far. What about adding them directly to the obj-$()-list
> > in the Makefile of the drivers? Since the code of most drivers is far
> > less than a memory page many drivers would not increase in codesize at
> > all (the demod code would just fill up the free space in the last page
> > of the driver) when the demod code gets linked directly into the card
> > driver module...
> >    
> Here are the sizes of the frontends:
> Module                  Size  Used by
> ves1820                 7428  0
> tda80xx                 9216  0
> cx22702                 6912  0
> mt352                   6788  0
> nxt6000                 7680  0
> sp887x                  8452  0
> tda1004x               12932  0
> ves1x93                 7296  0
> mt312                   9220  0
> dib3000mb              17156  0
> l64781                  8452  0
> tda8083                 7044  0
> cx24110                 8708  0
> cx22700                 6532  0
> sp8870                  8708  0
> stv0299                12164  0
> at76c651                6660  0
>
> Compiled on debian woody with GCC 2.94.5
>
> The av7110 has the most requirements: ves1820, ves1x93, stv0299, tda8083, 
> sp8870
>  
Please see my mail from two days ago for a detailed analysis with 
objdump on the example of the cx27000/alps_tdmb7 driver, only about 
1kByte from 6kB are actually code. Pathological examples like the 
stv0299 are harder to analyze, here a big part of the code are probing 
and if(tuner_type=XXX) {} handlers.

Holger