[
Lists Home |
Date Index |
Thread Index
]
----- Original Message -----
From: "Elliotte Rusty Harold" <elharo@metalab.unc.edu>
To: "David Megginson" <dmeggin@attglobal.net>
Cc: "XML Developers List" <xml-dev@lists.xml.org>; "SAX Developers' List"
<sax-devel@lists.sourceforge.net>
Sent: Friday, March 05, 2004 10:55 AM
> At 9:11 AM -0500 3/5/04, David Megginson wrote:
>
>
> >* 2004-0005: endDocument
> >
> >- note that a parser might not invoke endDocument after reporting a
> > fatal error (this produces the fewest incompatibilities)
>
> I'm a little uncomfortable with this "fix". I think always calling
> endDocument is the right thing to do, and it should be required in
> the spec. I don't think this would increase incompatibility in any
> significant way. It would simply encourage vendors of non-conformant
> parsers to bring their parsers into compliance, thereby increasing
> compatibility. This would allow users to depend on this behavior for
> the first time.
This discussion is not complete without considering exceptions.
A handler could throw an exception too, not just the parser.
The ability to use endDocument for "cleanup" (especially in
filter chains based on content handlers) would only be preserved
if endDocument was called even in the presence of an exception.
Although, without the ability to pass an error object down the
chain, this might not be very useful other than for resource cleanup.
> I don't think a parser suddenly changing from not calling endDocument
> to calling endDocument is likely to cause major problems.
Well, I would say, even that could cause a problem. If you have
some code in endDocument and it gets called when you don't
expect it anymore ...
> Fixing
> this in the direction you propose would simply bake in the existing
> incompatibility. One way or the other, this should be nailed down.
> Either parser all parsers should call endDocument after a
> well-formedness error or none should. Letting it go either way is the
> real problem.
I would phrase it this way:
Case 1) There is no fatal error and no exceptions.
Case 2) Fatal error, no exception.
Case 3) Exception thrown.
All three cases should have a defined, non-optional, behaviour,
but only in case 1) does this seem clear at this point.
Karl
|
