On Oct 28, 2008, at 10:06 AM, Patrick R. Michaud wrote:
On Tue, Oct 28, 2008 at 01:50:42AM -0500, Chris Dolan wrote:
My goal is to build arbitrarily complex data structures from closures
fired in my grammar. Specifically, I'm trying to write a PDF
parser --
my grammar is parsing correctly now, but I'd rather not have to
write the
closures in PIR if I can help it.
Would it make sense to use action methods (the "{*}" tokens) for the
closures instead of embedding them directly into the grammar?
Pm
Taking the {*} approach with the actions written in Perl6 instead of
NQP (or PIR) turned out to be significantly harder than I expected,
largely because I'm using a straight Perl6 grammar instead of PCT and
HLLCompiler. I encountered three problems, both solved temporarily:
1) how to tell Perl6 which action class to use, 2) how to get my
$/.result_object back, and 3) how to keep <ws> from blowing away my
overridden match object.
=== begin gory details ===
The Grammar.ACCEPTS nor Code.ACCEPTS methods don't pass an action
adverb to the PGE subrule, and there doesn't seem to be a way to set
a default action on a grammar. So, I changed my test code from this:
'1 0 R' ~~ PDF__Grammar::pdf_reference;
is(~$($/).WHAT, 'PDF__Syntax__Reference'); # fails, got 'Str'
because the action is not invoked
to the following, to force the action class:
my $method = PDF__Grammar::pdf_reference;
my $m = '1 0 R'.$method(action => PDF__Grammar__Actions.new);
is(~$m.WHAT, 'PDF__Syntax__Reference'); # succeeds
where my action looks like this:
class PDF__Grammar__Actions {
method pdf_reference($m) {
my $ref = PDF__Syntax__Reference.new(objnum => +
$m<objnum>, gennum => +$m<gennum>);
$m.result_object($ref);
return;
}
}
grammar PDF__Grammar {
rule pdf_reference { <objnum> <gennum> R {*} };
}
I found that a bit confusing because: a) because I'm invoking the
subrule directly, $/ is not getting set and b) assigning back to $m
at the end apparently calls get_scalar() on the Match instance, so it
is dereferenced to get my result_object instead. Unexpected, but
convenient. I might wrap it in a superfluous $() in case that
get_scalar() is ever changed.
In retrospect, if I had been matching against the whole grammar
instead of the subrule, my second problem would not have happened
because Grammar.ACCEPTS sets $/.
The third problem was because I had whitespace between the {*} and
the terminal } of the rule. That <ws> overrode $/ until I cuddled my
rule like "rule pdf_reference {<objnum> <gennum> R {*}};" Is there
an easy way to make my white space non-capturing? I'm sure I'm just
overlooking something...
=== end gory details ===
So, I have two change proposals. I'm not sure if either of them are
good ideas...
1) Make grammar rules be type Rule instead of Method, and add a
custom ACCEPTS that behaves like Grammar.ACCEPTS.
2) Add a Parrot-specific adverb to the Perl6 "grammar" declaration
to allow programmers to specify a default action class for the whole
grammar.
And bug reports:
1) Perl6 mangles Match instances when they are assigned to scalars
2) The rules incorrectly require a closing ";" to avoid a syntax error
I'm happy to write up concise ticket for any of those if they aren't
insane or already known.
Chris
