Path: utzoo!utgpu!jarvis.csri.toronto.edu!clyde.concordia.ca!uunet!mcsun!ukc!dcl-cs!aber-cs!rupert!pcg
From: pcg@rupert.cs.aber.ac.uk (Piercarlo Grandi)
Newsgroups: comp.lang.c++
Subject: Re: g++ core dump from improper use of extern "C"
Message-ID: <PCG.90Jan15134418@rupert.cs.aber.ac.uk>
Date: 15 Jan 90 13:44:18 GMT
References: <1581@aber-cs.UUCP> <18157@umn-cs.CS.UMN.EDU>
	 <25AFE9B7.14113@paris.ics.uci.edu> <25B0E7F7.6002@paris.ics.uci.edu>
Sender: pcg@aber-cs.UUCP
Organization: Coleg Prifysgol Cymru
Lines: 93
In-reply-to: rfg@ics.uci.edu's message of 14 Jan 90 20:58:31 GMT

In article <25B0E7F7.6002@paris.ics.uci.edu> rfg@ics.uci.edu (Ron Guilmette) writes:

    In article <25AFE9B7.14113@paris.ics.uci.edu> rfg@ics.uci.edu (Ron Guilmette) writes:
    >In article <18157@umn-cs.CS.UMN.EDU> mike@umn-cs.cs.umn.edu (Mike Haertel) writes:
    >>
    >>Just to be pedantic, this is not strictly true.  It is OK to
    >>declare a pointer to an incomplete type, so this works *here*.
    >>But consider the following counterexample:
    >>
    >>/*1*/	struct b { int x; };
    >>/*2*/
    >>/*3*/	void foo() {
    >>/*4*/	    struct b;
    >>/*5*/	    struct a { struct b *b; } *a;
    >>/*6*/	    struct b { struct a *a; } *b;
    >>/*7*/
    >>/*8*/	    b = a->b;
    >>/*9*/	}
    >>
    >>In this case, line 4 is necessary to make "struct b" an incomplete type.
    >>Otherwise, the "struct b *b" in line 5 would refer to the "struct b"
    >>declared on line 1...

Well, not in C; in C line 4 is an empty declaration and the type
used is the one defined on line 1 (see K&R version 1, "The ice is
admittedly thin here" footnote). In C++ 2.0 line 4 does declare a
new type, but only because of special casing:

    >Buzzzzzzzt!  Sorry, no!  But thanks for playing our game.  Just for being
    >such a good sport you win a free copy of our home game, complete with the
    >new 2.0 Reference Manual.

    OK, before you folks send me 8 zillion letters telling me what a dumbs***
    I am for having posted this, let me just say that I know already!

The special case is this: in C++ a struct or class implies an
automatic typedef; a typedef in theory does not declare a new
type, but only names an existing one; so line 4 should just
define a new typedef called 'b' for the type defined in line 1.
This means that it is virtually impossible then to define
mutually recursive types in an inner scope if *both* of them have
the same name as one in an outer scope. It would also make for
great hidden bugs. Try the following C program:

    typedef short t;

    static struct a {struct b *b; long l[1];} a;

    main()
    {
	{
	      auto t t1;
	      typedef double t;
	      auto t t2;

	      printf(" size t1 %u, size t2 %u\n",sizeof t1, sizeof t2);
    	}

    	{
	      struct a;
	      struct c {struct a *a; long l[2];} c;
	      struct a {struct c *c; long l[3];} a;

	      printf(" size a %u, size c %u, size c.a %u\n",
		  sizeof a, sizeof c, sizeof *(c.a));
    	}

    	return 0;
    }

This should print, on your typical 32 bit byte oriented machine, something
like 8 and 2 on the first line and 16, 12, 8 on the second line.

	NOTE: it will fail to compile on some ancient C compilers that
	erroneously handle a typedef like a #define, e.g. Sun's. Actually
	the problem is even more subtle, and has to do with the fact that
	certain classes of identifiers in C/C++ have global scope even if
	defined in an inner scope; struct/union/class names used to be such,
	if memory serves me right, and one could not redefine a name for one
	in an inner scope. C++, and C++ 2.0, complicates scope rules even
	further.

I am not suprised then that in the new C++ 2.0 empty declarations
have been special cased to imply the definition of a NEW type:

    Section 9.1 of the new 2.0 manual does indeed require the declaration of
    struct b in the example above.

... but this adds a new wart. Well, happy hacking! :-(.
--
Piercarlo "Peter" Grandi           | ARPA: pcg%cs.aber.ac.uk@nsfnet-relay.ac.uk
Dept of CS, UCW Aberystwyth        | UUCP: ...!mcvax!ukc!aber-cs!pcg
Penglais, Aberystwyth SY23 3BZ, UK | INET: pcg@cs.aber.ac.uk