RE: [xml] Error handling with the XmlReader API

["Stéphane Bidoul" <stephane bidoul softwareag com>]

>   Hum, one of the key point of the error handler over SAX is that
> the user code keep the control of the processing flow. The LastError
> code seems better in this respect. However the C# binding is based on
> exception and in general it's a good thing for error handling. For C
> that's not possible. So let's keep a non-callback interface for error
> handling too at that level.
>   There is a few technical problems with the approach you suggest,
> basically the error callbacks from the parser will not be synchronous
> to the current node of the parser, hard to avoid technically. 
> When doing
> a Read() a 512 byte blob is parsed, and errors are reported at that 
> point. Second problem is that there will be multiple error callbacks
> at the SAX level for a single logical error, that can 
> probably be handled
> relatively cleanly though. Last when parsing a chunk multiple errors
> may be raised during what's viewed from the reader interface 
> as a single
> step, so getting only the last message may prove too limited 
> in practice.
>
>   It's not that simple. The second interface seems more in-line with
> the reader design, but the first interface would be simpler 
> to implement
> and probably more accurate.
>
>   I can't really select one as being better than the other at 
> this point,
> what other people think, you included ?

Here are 2 more arguments to put in the balance.

First, my proposals lack locator information. That's probably easier
to add to the callback method than the getLastError approach.

Second, C# has a mechanism to use callbacks instead of exception to report
validation errors (the user can choose). For well-formedness
errors, exceptions are always thrown.

I guess one could conceive variations on the getLastError theme
to correlate errors with nodes but that would be complex to implement
and to use, IMO.

So I would vote for the callback approach, with locator information.

typedef void (*xmlTextReaderErrorFunc)(
   void *arg, const char *msg,
   int line, int col, const char *uri, int severity);

int xmlTextReaderSetErrorHandler(
   xmlTextReaderPtr reader, xmlTextReaderErrorFunc f, void *arg);

After all, I think it's not a good idea to give the xmlTextReaderPtr to
the callback, because when the callback is invoked, the reader
is not necessary in a valid state? 
However providing a user argument is probably necessary.

-sbi