Nice! It's a bit different since it seems to be a classic style parser
library and not quite in the same vein but still very cool. I
especially like how you overrode | and + to make expressing production
rules more natural
I think what I did is closer to what your blog talks about than
classic parsers. Basically, you start with leaf "Grammar" objects and
build complex Grammar productions by combining Grammars.
If want to see the very basics of this technique, here is a stripped
down version of it with a calculator example (self contained and
executable):
class SimpleGrammar
def initialize(grammar)
@grammar =3D grammar
end
def scan(cursor,buffer)
@grammar.scan(cursor,buffer)
end
def |(other); Code.new { |cursor,buffer|
scan(cursor,buffer) || other.scan(cursor,buffer)
} end
def +(other); Code.new { |cursor,buffer|
scan(cursor,buffer) && other.scan(cursor,buffer)
} end
def filter(buf0,&block); Code.new { |cursor,buffer|
buf =3D buf0.clone
scan(cursor,buf) && buffer.concat(block[buf])
} end
def discard; Code.new { |cursor,buffer|
scan(cursor,[])
} end
class Code < SimpleGrammar
def initialize(&block)
@block =3D block
end
def scan(cursor,buffer)
@block[cursor,buffer]
end
end
class Recurse < SimpleGrammar
def initialize(&block)
@grammar =3D block[self]
end
def scan(cursor,buffer)
@grammar.scan(cursor,buffer)
end
end
class Element < SimpleGrammar
def initialize(pattern)
@pattern =3D pattern
end
def scan(cursor,buffer)
c =3D cursor.read1after
if @pattern=3D=3D=3Dc
buffer << c
cursor.skip1next
true
end
end
end
# Make methods for our classes that call new for us
constants.each { |klass|
eval("
def #{klass}(*args,&block)
#{klass}.new(*args,&block)
end
def self.#{klass}(*args,&block)
#{klass}.new(*args,&block)
end
")
}
NULL =3D Code.new { true }
end
class IO
# implement just the methods we need to look like a cursor
def read1after;c=3Dgetc;ungetc(c);c;end
def skip1next;getc&&true;end
end
class Expression < SimpleGrammar::Recurse
def initialize; super() { |expr|
digit =3D Element(?0..?9)
int =3D Recurse { |int| digit+(int|NULL) }
number =3D
(int + (
Element(?.)+int |
NULL
)).filter("") { |n| [n.to_f] }
primary =3D Recurse { |primary|
number |
Element(?-).discard + primary + Code { |_,b| b[-1]=3D-b[-1] } |
Element(?().discard + expr + Element(?)).discard
}
product =3D Recurse { |product|
primary + (
Element(?*).discard + product + Code { |_,b|
b[-2]*=3Db[-1];b.pop } |
Element(?/).discard + product + Code { |_,b|
b[-2]/=3Db[-1];b.pop } |
NULL
)
}
sum =3D Recurse { |sum|
product + (
Element(?+).discard + sum + Code { |_,b| b[-2]+=3Db[-1];b.p=
op } |
Element(?-).discard + sum + Code { |_,b| b[-2]-=3Db[-1];b.p=
op } |
NULL
)
}
} end
end
Expression.new.scan(STDIN,buf=3D[]) && p(buf[0])
In the grammar package on rubyforge, the API is very similar to above,
but I added a bunch of stuff to improve performance, usability, and
features. Notice in the above, I'm not really tied to using a
character stream (IO). You could be parsing a token stream or
whatever. You just need to implement a subset of the Cursor (another
package of mine) API. This way you can build lexers, parsers,
preprocessors, etc in the same way.
