Re: [xml] Feature request: callbacks

[robert <robert xml 00008 org>]

>    + considering there is multiple callbacks, should the API
>      be at the registration of a single callback or a callback set ?

I would think a 'callback' datastructure (a 'struct') might be defined,
with which someone interested in registring callbacks can 'fill in the
blanks'. This should be forward-compatible too, since adding new callbacks
in future release will only add to the structure instead of changing it.

A small example of what I mean:
struct xmlCallbacks
{
        int     (*nodecreation)(void *context, XXX);
        int     (*nodedeletion)(void *context, XXX);
        ...
};

>    + scope of the registration, I'm leaning toward an application wide
>      registration, but some may argue that this must be done on a
>      document per document basis (and would require a structure change
>      for the document nodes :-( it also makes the lookup more costly
>      or even impossible in some obscure cases I would rather avoid this).

If you want thread-safety, it should be done (or atleast should be
_possible_ to be done) at document-basis.

As I pointed out before, some obscure projects (one of which I'm working
on ;) might not be able to do an application-wide registration which is
thread-safe (in other words, which is carried out before the process goes
multi-threaded).

>  - Callback signature:
>    a few week ago I would have used
>    void (*xmlTreeChangeCallback) (xmlNodePtr node);
>    but it's clear that Perl binding would require something like
>    void (*xmlTreeChangeCallback) (void *user_data, xmlNodePtr node);
>    is that sufficient ? I assume user_data would be registered at the
>    same time as the set of callbacks.

The latter is indeed much more usable than the former :) Because no
context-pointers can be passed by libxml's I/O callbacks, I had to resort
to gcc-specific extensions to be able to reach context-data, so I'd say new
callbacks should atleast have that possibility.

> I will let this brew till the end of the week and may spend time
> implementing it during the week-end.

All work and no play? :)

> <shame>
>  Euro's one or dollar's ?
> </shame>

Dollars are worth more ;)

                                                                      robert