Re: libcodecs(3), take 4

[Alistair Crooks <agc%pkgsrc.org@localhost>]

On Tue, Oct 05, 2010 at 01:45:46AM +0000, YAMAMOTO Takashi wrote:
> hi,
> 
> >> I hope this library becomes more modular.
> > 
> > You're going to have to help me out here - I have absolutely no idea
> > what you mean.
> 
> i think you put too much into a single *.c file.

It is an easy way to force access through the library API itself,
rather than just being able to call all functions.  If you know of a
better way to do this, I'd love to hear it.
 
> >> OTOH I'm not convinced that the add operation's "usefulness".
> > 
> > The "add" operation (codecs_add()) will add a transformation to the table.
> > Is that what you mean?
> > 
> > Without it, there would be no transformations available. So still a bit
> > unclear what you mean by this.
> 
> what's the point to export it?  ie. who is expected to use it besides
> the library itself?

I think something's been lost somewhere along the way.

The codecs_add() function is used to add a function to the table of
codecs.  This table is completely under the user's control.  If the
user does not add any codecs themselves before invoking a
transformation, then the default transformations are added to the
table.  This is expected to be the default behavior.

The user can lockdown the table (in the sense that subsequent attempts
to add a transformation will fail). It's not that difficult to work
around, I admit, but the functionality is there.
 
> talking about codecs_add:
> 
> - i don't think it's a good idea to use regex here.
>   what will happen if someone does codecs_add(..., ".*", ...)?

Then every attempt to find a codec will match that codec.  The user
controls what transformations are available.  The regular expression
is the best way to govern this; I don't accept Uebayasi-san's
assertion that regular expressions should not be used since they
involve dynamic memory allocation - the memory involved in a regex_t
is extremely small in today's terms, and takes place at regex
compilation time.
 
> - the "subset" concept sounds arbitrary and i don't understand how it could
>   be useful.

Yes, this is to add the user in loading only the specific set of codecs
that are needed. Say only the hash codecs are needed, then that's all
that need to be loaded. It is a means of reducing the number of codecs
to the bare minimum, and only allowing functionality which is absolutely
essential to be available.

Regards,
Alistair