Path: utzoo!mnetor!uunet!husc6!bloom-beacon!gatech!mcnc!decvax!ucbvax!hplabs!hp-pcd!uoregon!omepd!pcm
From: pcm@iwarpo3.intel.com (Phil C. Miller)
Newsgroups: comp.lang.fortran
Subject: Re: Wanted: Fortran Specification
Message-ID: <2785@omepd>
Date: 2 Feb 88 21:32:00 GMT
References: <6592@oberon.USC.EDU>
Sender: news@omepd
Reply-To: pcm@iwarpo3.UUCP (Phil C. Miller)
Distribution: na
Organization: Intel Corp., Hillsboro
Lines: 31
Keywords: Fortran grammar

In article <6592@oberon.USC.EDU> jchen@pollux.usc.edu (Joseph S. Chen) writes:
>   Does anyone have or know how to get a copy of Fortran-77 lexical
>definition in regular expression form and syntatic definition in BNF?
>It's fine if the specifications are not "lex" and "yacc" compatible.
>Thanks in advance.
 
Unfortunately, Fortran is a bit too convoluted for lexing with regular
expression-based lexers (this point is actually mentioned in the document
describing Lex).

The problem is that determining whether a token, say "DO", is a keyword or
a variable depends on the surrounding context.  Therefore, a regular expression
cannot capture the essential information about a token.

Additionally, since Fortran ignores blanks, it is sometimes difficult to
determine whether a token has ended.  For example,

	  DO 10 i = 1, 100

is equivalent to

	  DO10i=1,100

It is not possible to determine  whether "DO" is a keyword or a variable until
the comma is scanned (the fifth following token).

As a result, lexical analyzers for Fortran have to be "hand-rolled".

Phil Miller
{...}!tektronix!omepd!omepd!iwarp!pcm
My employer is not responsible for my opinions.