xml-dev - Re: SAX2: entity reporting madness orderingofstartDocument() &startDTD()

[David Brownell <david-b@pacbell.net>]

>  > I do care that either PE reporting should get "fixed" somehow, to
address
>  > issues like
>  >
>  > <![ %maybe; [ <!-- markupdecl --> ]]>
>  > <!ATTLIST foo %core.attrs; %i18n.attrs; %etc; >
>  >
>  > or that PE reporting be completely removed.
>
> Again, would it be sufficient if we reported only *external* PEs?

You mean, the kind that have SYSTEM identifiers?  I don't see how
that'd solve the problem; the "maybe" and "etc" (etc) PEs in those
examples can have system identifiers.


>  > Also, it seems wrong that there be a requirement that PE inclusion
>  > events be lexically scoped ("properly nested" etc) while other DTD
>  > events not be scoped in that way.  Consistency is a virtue
>  > ... either always do it, or never.
>
> Way back in once-upon-an-SGML-time, people cared about this kind of
> thing so that they would know which declarations appeared in which
> external entities (i.e. what's in the internal subset and what's in
> the external DTD subset and what's in a fragment that the DTD subset
> refers to).  If no one cares about this kind of thing any more, than
> perhaps it would be better to leave out PE reporting, or to allow it
> only for the external DTD subset with no finer-grained distinctions.
> What do people want.

I'd still like to have a clear record of the internal subset; that's not
a "once-upon-an-SGML" issue!

But I still don't see a way to report PE boundaries accurately without
providing callbacks at the level of tokens and string parsing.  Rather
than get inaccurate reports, I'd rather see none at all.

>
>  > At this point my preferred resolution would be to say (a) PE
>  > inclusion MUST NOT be reported, (b) declarations in the DTD may be
>  > reported in any order.  I could go either way on (b) but it seems
>  > least trouble overall.
>
> I'd love (b) as well.

Then let's consider (b) resolved for SAX2.

What about (a)?


>     In fact, I'd love to drop DeclHandler and
> LexicalHandler altogether, but I know that I sound like a broken
> record.

No way could I go that far!  The problems with entities derive from
XML, which doesn't have a very logical layering of abstractions in
that area.  I don't think it's possible to have an API which reports
entities sanely and is above the level of tokens and string parsing.

Drop entity reporting, and then the worst I'd say about those two
interfaces is that (a) enumerated notation attribute types aren't
really reported, (b) the root element name declaration is misfiled
as a lexical event.  Other than that, they do what folk have needed
in those areas, near as I can tell.

- Dave








***************************************************************************
This is xml-dev, the mailing list for XML developers.
To unsubscribe, mailto:majordomo@xml.org&BODY=unsubscribe%20xml-dev
List archives are available at http://xml.org/archives/xml-dev/
***************************************************************************