Re: Tutorial: adapted API parts

On Mon, Dec 04, 2000 at 10:38:09PM +0100, [email protected] wrote:
> >For what it's worth, we could make the APIs *appear* the same.
> 
> That's what I meant - just simplifying the interface by making it more
> general.
> 
> >However,
> >the programmer still needs to decide how the callback will be invoked. i
> >don't think it makes sense to autogenerate *both* styles (XS & C)
> >implicitly without asking.  And it certainly doesn't make sense to adapt
> >at run-time by detecting the caller (somehow).  Am i missing something?
> 
> Please see the "Perl watcher, C callback" chapter. There's a function
> c_callback() used as a Perl invoked watcher callback. The interface is,
> as usual (and thanks to Inline):
> 
>   void c_callback(pe_event * event)
> 
> I just read C code produced by Inline and found that - correct me if
> I'm wrong - it just builds a wrapper around a C function.

Oh!  Indeed it does.  Hm hm.

> Somewhere in
> the wrapper, the "original" function is called. So we have the wrapper
> (to invoke the callback from Perl) and *still* the callback function as
> it
> was written by the application programmer.
> 
> Now let's assume I want to have the same callback functionality for a
> watcher initialized on *C* side. (Whether this makes sense or not - it
> could be different scripts and I might want to reuse C callback code
> unchanged which was already tested successfully.)
> 
> *If* the C invokation interface would be the same
> 
>   void c_callback(pe_event * event)
> 
> , an application programmer could invoke just this same callback - from
> C side now. The difference is that no intermediate translation wrapper
> is called *but this is done behind the scenes*. I'm calling the
> "original function" in C just as the wrapper function does. For the
> application programmer who just has to write the callback code this is
> not too important, isn't it? I currently think it is not, because there
> is no further difference between the callback types except of the
> interface. (Maybe I missed something?)
> 
> Well, the wrapper would still be there (unless someone declares the
> callback as static). Is this expensive? (Real question, no rhetorics!)
> As I understand it at the moment, the wrapper would be automatically
> invoked if the callback is called on Perl side, and passed by if the
> caller resides on the C shore.

OK.  Does anyone oppose changing the C callback from:

(* (void(*)(void*,pe_event*)) ev->callback)(ev->ext_data, ev);

To:

(* (void(*)(pe_event*,void*)) ev->callback)(ev, ev->ext_data);

?

-- 
May the best description of competition prevail.
      (via, but not speaking for Deutsche Bank)