Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!uunet!tut.cis.ohio-state.edu!ucbvax!BRL.MIL!butler
From: butler@BRL.MIL ("Lee A. Butler")
Newsgroups: comp.sys.sgi
Subject: Re: Is this a lint bug?
Message-ID: <9012142035.aa25061@WOLF.BRL.MIL>
Date: 15 Dec 90 01:35:51 GMT
Sender: daemon@ucbvax.BERKELEY.EDU
Organization: The Internet
Lines: 59

Based upon some early returns, I can see that I did not communicate well
enough in my first posting.  Allow me to attempt a more thorough description.
The lines in "test.c" which have the "/* warn */" comments generate warnings
from lint.  I don't understand why, since in each case, a value of type "char"
is what is provided.  

Script started on Fri Dec 14 19:58:03 1990
1 >cat test.c
#include <stdio.h>
char *str = "the first character";
main()
{
        char *loc = "hello world";
        char cp[64];
        
        (void)sprintf(cp, "%s=\"%c\"\n", str, *loc);            /* warn */
        (void)sprintf(cp, "%s=\"%c\"\n", str, (char)*loc);      /* warn */
        (void)sprintf(cp, "%s=\"%c\"\n", str, *((char *)loc));  /* warn */

        (void)sprintf(cp, "%s=\"%c\"\n", str, (signed char)*loc);    /* ok */
        (void)sprintf(cp, "%s=\"%c\"\n", str, *((signed char*)loc)); /* ok */

        (void)printf("%s", cp);

        return(0);
}
2 >lint test.c

test.c
==============
Warning: (8)  sprintf: (char) format, (unsigned char) arg (arg 4)
Warning: (9)  sprintf: (char) format, (unsigned char) arg (arg 4)
Warning: (10)  sprintf: (char) format, (unsigned char) arg (arg 4)
3 >exit
exit

script done on Fri Dec 14 19:59:06 1990

script done on Fri Dec 14 19:59:07 1990

While I am aware that on various architectures, variables of type "char"
behave as either "unsigned char" or "signed char", I would expect such
behavior to be consistent.  I would also expect lint to be unconcerned 
with architecture specifics such as this.  Lint should only be concerned
with a variable's declared type.  A variable of type "char" should be portably
passed to sprintf for a "%c" format irrespective of whether the local
architecture treats "char" variables as signed (implying sign extension when
passed to sprintf) or unsigned (no sign extension).

Furthermore, I believe that the "signed" keyword is an ANSI C construct.
Thus, while the type "char" may (or may not) behave as a signed char, it is
unreasonable for lint to expect me to cast all of my "char" variables as
"signed char" for use with sprintf and friends.  The "signed" keyword is an
"available" but not "required" keyword in ANSI C programs.

Lee A. Butler
SLCBR-VL-V					Internet: butler@brl.mil
Ballistic Research Laboratory			   Phone: (301) 278-9200
Aberdeen Proving Grounds, MD 21005-5066