Re: [rfc][patch] DRM initial function table support.

[Ian Romanick <idr@us.ibm.com>]

Christoph Hellwig wrote:
> On Tue, Aug 31, 2004 at 09:07:11AM -0700, Ian Romanick wrote:
> 
>>I think the intention is to have default functions set in the 
>>device-independent code and have the device-dependent code over-ride 
>>them.  Since the defaults may not always be NULL, doing a struct like 
>>that wouldn't really work.  I suppose we could have a struct and a 
>>device-independent function that copies the non-NULL pointers from the 
>>per-device struct.  Would that be better?
> 
> Don't copy them.  Just put
> 
> if (foo->ops->method1)
> 	foo->ops->method1(args);
> else
> 	generic_method1(args);
> 
> in your code.  It's an additional branch, but you avoid the indirect
> functioncalloverhead in exchange.

<MrHorse>No sir, I didn't like it.</MrHorse>  That would not only be 
ugly to read, but it would add maintenance burden.  If the default 
changes from NULL to non-NULL, code has to be changed from doing nothing 
in the NULL case to calling generic_method1.  The one place that we miss 
is the one place that will crash Linus' box. :)