3 Dec
2018
3 Dec
'18
9:36 a.m.
I did just that. The National Bureau of Standards picked it up
in NBS Handbook 131, "Using ANS FORTRAN" (1980). It is expressed
in the same formalism that Burroughs used for Algol.
Doug
3 Dec
3 Dec
11:31 a.m.
New subject: > there is no way that FORTRAN can be described in any syntax
Your message dated: Sun, 02 Dec 2018 22:36:59 -0500
--------
I did just that. The National Bureau of Standards picked it up
in NBS Handbook 131, "Using ANS FORTRAN" (1980). It is expressed
in the same formalism that Burroughs used for Algol.
Doug
a couple of more data points:
Arthur Sale wrote an article on classifying fortran statements,
Volume 14 Number 1 of the Computer Journal that describes how to
classify fortran statements into 1 of 36 types.
URL:https://www.researchgate.net/profile/Arthur_Sale/publication/220459829_…)
by using the above, you can do some preprocessing to correct spacing,
line continuations, remove sequence numbers (cols 73-80) and other
formatting to make parsing easier.
John Levine wrote a ftn77 subset parser using lex and yacc (the shar
file that I have is dated Nov 1988).
-ron
===============================================================================
Ron Young rly1(a)embarqmail.com
4 Dec
4 Dec
2:10 a.m.
New subject: > there is no way that FORTRAN can be described in any syntax
On 12/3/18, Ron Young <rly1(a)embarqmail.com> wrote:
John Levine wrote a ftn77 subset parser using lex and yacc (the shar
file that I have is dated Nov 1988).
Just curious--what parts of Fortran 77 did he leave out?
Unfortunately the Fortran language was being developed at the same
time that Chomsky was first doing his research on formal grammars.
Languages such as Algol, C, PL/I, and Pascal were designed to be
easily split into a lexical regular grammar and a context-free grammar
(the lex/yacc paradigm). Fortran has some nasty grammatical quirks
that make it a bear to parse--context-sensitive lexical analysis, for
example. Given the string:
DO10I=1,10
you don't know whether to lex it as:
<identifier> DO10I
<operator> =
<integer constant> 1
or:
<keyword> DO
<integer constant> 10
<identifier> I
<operator> =
<integer constant> 1
until you see the comma.
Yes, full Fortran can be expressed in BNF and other formal syntax
mechanisms. But I don't think you can use lex and yacc to generate a
compiler for it.
-Paul W.
2:55 a.m.
New subject: > there is no way that FORTRAN can be described in any syntax
Your message dated: Mon, 03 Dec 2018 15:10:32 -0500
--------
On 12/3/18, Ron Young <rly1(a)embarqmail.com>
wrote:
not sure... I'll check when I get home.
John Levine wrote a ftn77 subset parser using lex and yacc (the shar
file that I have is dated Nov 1988).
Just curious--what parts of Fortran 77 did he leave out?
Yes, full Fortran can be expressed in BNF and other formal syntax
mechanisms. But I don't think you can use lex and yacc to generate a
compiler for it.
agreed... but if you use Sale's classifier, you can insert
whitespace to convert the input to something that lex/yacc can
handle. Although in practice, using a hand-written lexer or
a state machine (like open-watcom IIRC) would probably be the
way to go.
-ron
-Paul W.
===============================================================================
Ron Young rly1(a)embarqmail.com
2211
days inactive
2211
days old
3 comments
3 participants
participants (3)
-
Doug McIlroy -
Paul Winalski -
Ron Young