Charles
Thanks for your helpful advice.
1) my current grammar is http://mkrmke.org/parser/mkr_parser.y
Feel free to browse it & any files in that directory (& any files on my
website).
Any comments would be appreciated.
2) iyacc is bison, modified with -i option to produce Unicon code.
3) I have 0 reduce/reduce conflicts and 28 shift-reduce conflicts.
I have had a couple of reduce/reduce conflicts.
The last one was just yesterday, when I tried to insert an action
after "genword" in the "iteration" rule.
4) I changed some precedences (%left & %right). I'm not sure if they
are having any effect.
5) I grepped for lr.type and glr-parser in mkr_parser.icn and y.output,
and found none.
6) I get "surprising" syntax errors -- on unique tokens which begin a
proposition.
7) An example of inserting actions in middle of rules:
proposition: mkrEND phrase SEMICOLON {$$ := GROUP($1,$2,$3);
interpret_group($$)};
produced a syntax error on mkrEND. (I use mkr prefix on a number of terminals
to avoid
conflicts with Unicon terminals.) When I changed the rule to
proposition: mkrEND {$$ := $1} phrase SEMICOLON {$$ := GROUP($1,$3,$4);
interpret_group($$)};
that syntax error no longer occurred. I've done this to several other rules,
but have many more to change.
8) How should I order the nonterminal rules?
I have rearranged the rules to be more bottom-up-first. Not sure if that helped.
Dick McCullough
Context Knowledge Systems
What is your view?
> Date: Thu, 3 Jul 2014 04:29:09 -0400
> From: [email protected]
> To: [email protected]
> Subject: Re: [Unicon-group] IYACC 1.0 bug report
>
> On Wed, 2 Jul 2014 08:36:26 -0700
> "Richard H. McCullough" <[email protected]> wrote:
>
> > After looking at y.output, I don't think shift/reduce conflicts are the
> > problem.
> > I think mKR requires too much look ahead.
> >
> > Is iyacc supposed to tell me that?
>
> I have often run into problems with iyacc's limitation of 1 lookahead;
> it will report reduce-reduce conflict(s) if you exceed it.
> I have sometimes fixed problems by re-factoring my rules, adding more
> nonterminals.
> Adding multiple rules with common prefixes is sometimes helpful.
>
> One grammar I wrote gave over 2000 shift-reduce conflicts,
> but I have used it for over a decade without finding any inappropriate syntax
> errors.
>
> IIRC, grammars that are irreducibly lookahead 2+, or rarely full-LR(1)*
> grammars,
> will always give reduce-reduce errors, and any iyacc output will be useless.
> *iyacc does LALR(1), which is a subset of LR(1).
> Bison can optionally do LR(1) if lr.type is set, so check for lr.type in
> the grammar.
> Bison can do a GLR parser, so check for %glr-parser in your grammar.
> If you have no reduce-reduce errors but get incorrect results:
> I don't know if iyacc cares about the order of rules or not.
> If it does, you may need to re-order your rules or a rule's branches.
> For untested grammars, remember that the order of your nonterminals is
> critical.
>
> > Can I cure that by putting actions in the middle of rules?
>
> I am not sure what you mean here.
> Factoring similar runs of similar terminals in the middle of rules to new
> nonterminals
> lets you put actions on a terminal phrase, and this often helps.
> One often ends up with fewer total rules as well.
> I avoid mixing rules or terminals with different precedences when factoring.
> unigram.y has many examples for one to review.
> If you can find examples of rules that are NOT LL(k), study them.
>
> What we both need is a bounded-backtracking parser that can handle LR(2+) and
> produce icon code.
> The alternative is writing workarounds that are AFAIK more art than science.
> For this, classroom work that includes hand-compiling LR(1) grammars is very
> helpful.
>
> --
> Charles Evans <[email protected]>
>
------------------------------------------------------------------------------
Open source business process management suite built on Java and Eclipse
Turn processes into business applications with Bonita BPM Community Edition
Quickly connect people, data, and systems into organized workflows
Winner of BOSSIE, CODIE, OW2 and Gartner awards
http://p.sf.net/sfu/Bonitasoft
_______________________________________________
Unicon-group mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/unicon-group
