Re: [PATCH] Implement %define lr.default_rules.

[Akim Demaille]

Le 20 avr. 09 à 07:48, Joel E. Denny a écrit :

Hi Joel,

> I haven't pushed yet because I would really like to rename "default  
> rule"
> to "default reduction" throughout Bison first:
>
> 1. A rule is a production in a grammar, so a default rule sounds to me
> like it is a single property of the grammar.  When I read "perform a
> reduction by the default rule when there's no other action on the
> lookahead", I want to check the grammar to see what it specifies as  
> the
> default rule.

I agree.

> 2. A reduction is a parser action that applies a rule.  Parser states
> contain actions, so a default reduction sounds to me like it is a  
> property
> of a state, as we expect.  In other words, when I read "perform the
> default reduction when there's no other action on the lookahead", I  
> want
> to look for the reduction in the current parser state, where all the  
> other
> relevant actions are stored.

I agree too.

> Does my argument make sense to anyone else?  Is this terminology too
> standard to change now?  If you agree with the change, should it be
> "default reduction" or "default reduce"?  Thanks.

I agree with the change, and I would use "default reduction".


> +  /* We need a lookahead either to distinguish different reductions
> +     (i.e., there are two or more), or to distinguish a reduction  
> from a
> +     shift.  Otherwise, it is straightforward, and the state is
> +     `consistent'.  However, for states that have any rules, treat  
> only

I don't understand "states that have any rules".

> +     the accepting state as consistent (since there is never a  
> lookahead
> +     token that makes sense there, and so no lookahead token should  
> be
> +     read) if the user has otherwise disabled default rules.  */
>   if (rp->num > 1
> -      || (rp->num == 1 && sp->num && TRANSITION_IS_SHIFT (sp, 0)))
> +      || (rp->num == 1 && sp->num && TRANSITION_IS_SHIFT (sp, 0))
> +      || (rp->num == 1 && rp->rules[0]->number != 0
> +          && default_rule_only_for_accept))
>     n_lookahead_tokens += rp->num;
>   else
>     s->consistent = 1;



> @@ -244,6 +245,7 @@ print_reductions (FILE *out, state *s)
>   rule *default_rule = NULL;
>   size_t width = 0;
>   int i, j;
> +  bool non_default_action = false;

I'm not fond of negated Boolean variables.  Especially seems it seems  
used only negated itself :)  But then, I can understand "false" as  
"not known yet" and "true" as "known".

> +AT_BISON_CHECK([[input.y]], [[1]], [[]],
> +[[input.y:1.9-24: invalid value for %define variable  
> `lr.default_rules': `bogus'
> +]])

It should be easy to report the set of valid values too.  Sort of  
unexpected/expected :)

> +# AT_TEST_TABLES_AND_PARSE(TITLE, COND-VALUE, TEST-SPEC,
> +#                          DECLS, GRAMMAR, INPUT,
> +#                          BISON-STDERR, TABLES-OR-LAST-STATE,
> +#                          [OTHER-CHECKS],
> +#                          [PARSER-EXIT-VALUE],
> +#                          [PARSER-STDOUT], [PARSER-STDERR])

Wow, this is a m4sterpiece of 4rt.