Re: the key_depends() issue

[Iago Toral Quiroga <itoral igalia com>]

El mar, 01-02-2011 a las 13:49 +0100, Guillaume Emont escribió:
> On 01/02/2011 13:40, Juan A. Suárez Romero wrote:
> > On Tue, 2011-02-01 at 12:23 +0100, Guillaume Emont wrote:
> >> Not needed
> >> ~~~~~~~~~~
> >>
> >> supported_keys() is not strictly needed in a resolution algorithm, and
> >> could
> >> only provide a marginal performance improvement if its result is
> >> cached.
> >>
> >
> > First of all, thanks for the email.
> >
> >
> > I agree that supported_keys() is not strictly needed to do the job. But
> > the same can be told about supported_keys() in media sources: they are
> > not a must to do the job.
> >
> > Still, I think it is worth to keep supported_keys() to tell applications
> > which keys are supported, instead of applications needing to call
> > continuosly may_resolve() to filter which keys are supported. Actually,
> > I don't see supported_keys() as a complex function.
> True, I only thought of supported_keys() from the point of view of the
> big resolve() algorithm in core, but that information can indeed be
> useful to application developer, so I guess we should keep it as it is.

That was exactly the point I was going to make. I think we should keep
supported_keys.

Other than that, congrats for the good analysis and explanation, the
proposal is sensible and solves the problems, so go ahead with the
changes :)

Iago