lclint-interest message 194

From evs Sat Feb 21 18:55:44 1998
Date: Sat, 21 Feb 98 18:31:00 -0500
From: evs (David Evans)
To: 100045.1545@compuserve.com, lclint-interest@larch.lcs.mit.edu,
        weco@csi.com
In-Reply-To: David Evans's message of Fri, 20 Feb 98 19:43:43 -0500
Subject: re: signal function interface


I was confused in my last message.  In fact, the problems with lclint
are not as serious as I had thought.

> I tried to typedef the signal handler function and to use it:
> 
> > typedef void SignalHandlerT (int);
> > void signal (int, SignalHandlerT *);
> > #define SIG_ERR (SignalHandlerT *) -1

This is indeed a bug in the type system, resulting from lots of
confusion about when there is an is not an implicit function pointer.  

LCLint will work correctly if you do,

	typedef void (*SignalHandlerT) (int);

instead.  

I still find the C syntax here particularily confusing.  Perhaps one of
the C standard experts on the list can help clear this up for me:

typedef int func1 (int);
typedef int (*func2) (int);

int tf (int (p1)(int), int (*p2) (int), func1, func1 *, func2);

int f1 (int);

int f (void)
{
  func1 fa;
  func2 fb;
  func1 *fap;

  fa (2);
  fb (3);
  fap (3);

  return tf (f1, f1, f1, f1, f1);
}

Is this a conforming program?  (According to all the compilers I try
there is no type error in it.  LCLint incorrectly reports 2 type errors
because of the same problem with implicit function pointers.)  

Its seems quite strange to me where you get implicit function pointers
(i.e., why func1 and func1 * are the same), and where you don't.  (Is
this explained clearly in the standard somewhere?)

> > LcLint declares `signal'
> > 
> > >  void signal (int, void (*) (int));
> > 
> > to be a function taking a signal number and a pointer to a signal handler
> > and RETURNING NOTHING.  What about checking the successfull installation
> > of signal handlers?  ANSI C declares signal
> > 
> > > void (*signal (int sig, void (*func) (int)))(int);
> 
> This is also a bug!  The standard library (ansi.h) declares it
> correctly, but there is a bug in lclint that prevents it from reporting
> ignored return values for functions returning functions that return
> void.

Opps, its not a bug in the checking.  The standard library is incorrect,
and should declare it as Hermann suggests.

--- Dave