Xref: utzoo comp.lang.c:30540 comp.os.msdos.programmer:247 alt.msdos.programmer:1959
Path: utzoo!utgpu!news-server.csri.toronto.edu!rutgers!cs.utexas.edu!swrinde!zaphod.mps.ohio-state.edu!samsung!noose.ecn.purdue.edu!iuvax!purdue!haven!mimsy!chris
From: chris@mimsy.umd.edu (Chris Torek)
Newsgroups: comp.lang.c,comp.os.msdos.programmer,alt.msdos.programmer
Subject: Re: Suspicious pointer conversion warning in Turbo C 2.0
Message-ID: <25725@mimsy.umd.edu>
Date: 26 Jul 90 06:04:03 GMT
References: <1990Jul25.230836.2442@Octopus.COM>
Organization: U of Maryland, Dept. of Computer Science, Coll. Pk., MD 20742
Lines: 61

In article <1990Jul25.230836.2442@Octopus.COM> stever@Octopus.COM
(Steve Resnick) writes:
>As far as I know, when I provide a prototype for a function which accepts
>a void pointer, the compiler should not bitch about type conversions between
>pointers. The code fragment below generates a warning message ....

The ANSI standard only says when diagnostics are required; it makes no
prohibitions about extra diagnostics.  In other words, compilers are free
to warn about anything they like (`foo.c, line 354: warning: code written
at 4 AM').

Of course, some warnings are rather less useful than others (although the
above might well be particularly good, actually).

>The parameter bitched about is the second pointer.
>The first pointer in both cases is not complained about.

[everything but the relevant bits trimmed:]
>void AddLNode(void ** Hptr, void * Nptr);
>	Tlist *Head, *Walker;
>	AddLNode(&Head,Walker);	/* This statement generates a warning */

Now this is exactly backwards: a good compiler should complain about
the first parameter, not the second.  Argument passing is semantically
equivalent to assignment, so the commented line is rather like writing:

	void **Hptr;
	void *Nptr;

	Hptr = &Head;
	Nptr = Walker;

The first assignment expands to

	<object, pointer to pointer to void, Hptr> =
		&<object, pointer to Tlist, Head>;

which becomes

	<object, void **, Hptr> = <value, Tlist **, &Head>;

which is an attempt to stuff a `Tlist **' into a `void **'.  The C
standard guarantees that any pointer type (e.g., Tlist **) must fit
into a `void *'; it makes no such guarantee about a `void **'.  In
other words, this need not work, and the compiler ought to complain.

The second assignment expands to

	<object, pointer to void, Nptr> =
		&<object, pointer to Tlist, Walker>;

which becomes

	<object, void *, Nptr> = <value, Tlist **, Walker>;

which (because Tlist** fits in void*) has to `work' (the compiler is
free to complain, but this should at least be an option, or customers
will turn away in droves).
-- 
In-Real-Life: Chris Torek, Univ of MD Comp Sci Dept (+1 301 454 7163)
Domain:	chris@cs.umd.edu	Path:	uunet!mimsy!chris