Re: [XPath] A.2.2 Parsing note

Ed, thank you for your last call comment.  It will be processed by the 
working group.

A point of clarification.  I believe you're saying that, without <> 
grouping and any lexical states or lexical lookahead, that the BNF can 
produce an unambiguous LALR parser (with a standard tool... YACC for 
instance), except for leading "/"?  I'm a little surprised to hear this, 
though I am aware that the groupings can be at least greatly reduced with 
LALR(1).  Without reserved keywords, it's hard to see how "if" can be 
unambiguously lexed and serve a useful purpose to the parser.  There would 
seem to be lots of other issues of this kind.  I do have a test LALR 
parser, but so far it is a fairly straight-forward transformation of the 
LL grammar and lexical constructs, i.e. it uses the pretty much the same 
lexer as the LL parser.  It would be a useful exercise to remove as much 
of the grouping as lex states as possible.

> All other <> grouping annotations and lexical tables should be a
> non-normative
> guide that may be of assistance to those pursuing an LL approach.

The WG has considered this in the past, at least for the lexical tables, 
and has so far concluded they should be normative.  But we'll take a fresh 
look at it... I don't think there's any question that it would be better 
if they were non-normative.  The issue is only if they carry some 
information that is not stated elsewhere.  In some ways, this has been a 
changing picture as the grammar has evolved, so it may be that they used 
to carry unique information, but no longer do.

Also, given the grammar is LL, the parsing rules should be stated in LL 
and not LALR, I think.

-scott

public-qt-comments-request@w3.org wrote on 01/27/2004 07:00:24 AM:

> 
> Hi
> 
> The tables are _not_ a declarative way to specify behaviour. They are
> a highly imperative imposition of a solution that then has to be 
disclaimed
> except in so far as it still has to be observed.
> 
> The declarative approach is to specify that the preceding BNF when
> interpreted
> at a lexical level maximising the length of tokens such as QName and 
then
> in the conventional LALR(1) shift-reduce fashion has exactly one
> shift-reduce
> conflict on a leading "/" that is to be resolved as per
> grammar-note:leading-lone-slash.
> 
> All other <> grouping annotations and lexical tables should be a
> non-normative
> guide that may be of assistance to those pursuing an LL approach.
> 
>    Regards
> 
>       Ed Willink
> 
> ------------------------------------------------------------------------
> E.D.Willink,                             Email: mailto:EdWillink@iee.org
> Thales Research and Technology (UK) Ltd, Tel:  +44 118 923 8278 (direct)
> Worton Drive,                            or  +44 118 986 8601 (ext 8278)
> Worton Grange Business Park,             Fax:  +44 118 923 8399
> Reading,   RG2 0SB
> ENGLAND          http://www.computing.surrey.ac.uk/personal/pg/E.Willink
> ------------------------------------------------------------------------
>

Received on Wednesday, 28 January 2004 09:44:34 UTC