Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!henry
From: henry@utzoo.UUCP (Henry Spencer)
Newsgroups: net.micro.apple
Subject: Re: LL(1) parser for an apple. (are you - (nf)
Message-ID: <3863@utzoo.UUCP>
Date: Wed, 16-May-84 17:42:28 EDT
Article-I.D.: utzoo.3863
Posted: Wed May 16 17:42:28 1984
Date-Received: Wed, 16-May-84 17:42:28 EDT
References: <3321@fortune.UUCP>
Organization: U of Toronto Zoology
Lines: 30

Rob Warnock says, in part:

   It also happens to be the case that IF the grammar is LL(1) or even
   reasonably close, a recursive-descent parser is:

	...

	4. Capable of acceptable error-handling (assuming you want to
	   detect errors, not "fix" them)
	
This is actually an understatement.  Error-handling is much easier in
recursive-descent/LL(1) than in LALR systems like yacc.  It is possible
to build an error-handling module which fits between the parser and the
scanner which does all the work.  The parser has to be specific about
what it wants -- this just means a slightly unusual-looking "scanner"
interface for the parser -- but the parser then sees a syntactically-
correct program regardless of what the user feeds in.  The key feature
of top-down parsing, for this and other purposes, is that the software
"knows what it wants":  it has a fairly specific idea of what sort of
input it's expecting at a given instant, and can take intelligent action
if there is a mismatch.  Bottom-up methods like yacc don't give you this
sort of detailed context to work in, which is why error repair is such
a headache for them.

The above is not theory; such an error-repair system was built here at
U of T years ago for the SP/k compiler, and several similar things for
other languages have been done since.  It really works well.
-- 
				Henry Spencer @ U of Toronto Zoology
				{allegra,ihnp4,linus,decvax}!utzoo!henry