Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!ames!ig!arizona!gudeman
From: gudeman@arizona.edu (David Gudeman)
Newsgroups: comp.lang.misc
Subject: Re: Language Design
Message-ID: <9868@megaron.arizona.edu>
Date: 24 Mar 89 08:24:23 GMT
Organization: U of Arizona CS Dept, Tucson
Lines: 67

In article  <9170@claris.com> hearn@claris.com (Bob Hearn) writes:
]In article <9861@megaron.arizona.edu> gudeman@arizona.edu (David Gudeman) writes:
]>
]>One inaccuracy is in the production
]>
]>	typedef-name : identifier
]>
]>I don't know if there are others.  "typedef-name" occurs in the production
]>
]>	type-specifier: ... | typedef-name
]>
]>and a mechanical reading of this implies that _any_ identifier can
]>occur as a type-specifier.  This is not the case.  In fact, an
]>identifier can only be a type-specifier if it has previously been
]>declared with a typedef.
]
]Yes, this is another big misunderstanding in this discussion... I would
]say that the *grammar* stipulates that a type-specifier can be an
]identifier, but the *semantics* specifies that this results in a valid
]program only when the identifier is a valid type name.  This is really
]the same problem as  [undeclared identifier]

It sounds as though you are defining the semantics of a language as
any part that cannot be described with a context free grammar.  Of
course this makes you trivially correct, but does not affect the
opinions of people who criticize C based on a different definition of
syntax and semantics.  Also, you cannot treat the C grammar as a
formal grammar that is only meant to accept strings from a formal
language.  The grammar is intended to show syntactic categories with
semantic attachments.

The problem of undeclared identifiers is common to many programming
languages, but the typedef problem is more serious.  An identifier is
always parsed into category "identifier" regardless of whether it has
been declared.  However, typedefs are ambigous.  For example:

typedef int foo;
int i;

void f()
{ foo *i;
  ...
}

The K&R grammar is ambigous on how to parse this, while any C
programmer would expect "i" to be a local pointer to an int.  Once an
identifier has been declared as a type, it should no longer be an
identifier, but rather have a different syntactic class.  No, this
does not affect the language accepted by the grammar.  But it does
affect the syntactic category of the whole string "foo *i;".

Because of the differences in syntactic categories, most people would
say that

static foo x;

should be a syntax error if foo has not been typedefed.  You can argue
that, but like I said before, you are only defining terms to fit your
views.

>In any case, this discussion is getting old... I vote we talk about something
>new.  This is comp.lang.misc, not comp.lang.C.grammar, right? :-)

The following should not be taken as a flame, I'm just making an
observation: If you want to stop the discussion, you don't have to
reply.  It doesn't work to write an article and then suggest that no
one respond to it after you have gotten in the last word.