C Preprocessor bug

[utzoo!decvax!harpo!utah-cs!utah-gr!thomas]

The symptom:
	Line numbers on compiler error messages (sometimes) do not match
the line actually in error.  Usually the reported line numbers are too
big.

The explanation:
	Invoking a macro with arguments, and putting a newline into the
argument list can cause the preprocessor and the compiler to count the
wrong number of lines.  If the actual which contains the newline is
only used once in the expansion, then the compiler will count right,
but the preprocessor will count wrong.  If the actual is not used or is
used more than once, both will count wrong, but they will disagree on
the "correct" line number.  The preprocessor counts wrong all the time
because it rescans the substituted text, thus counting each newline
multiple times.

To reproduce:
Run this input through cc (without the line numbers):

     1  #define twice(a)	a;a

     3  main()
     4  {
     5  	int a;
     6  	twice(a =
     7  		0);
     8  	while;
     9  }

Cc produces the following error message:
"tmp.c", line 9: syntax error

Note that the message has the wrong line number.  Inserting a #if 0,
#endif pair will cause the line number to be even further off (because
the preprocessor has seen 3 newlines while expanding the macro).

The fix:
The fix involves two steps:
	1. Get the preprocessor to count the newlines properly.
	2. Make the preprocessor put out exactly as many newlines in
	   the output as it saw in the input.  Since newlines are
	   exactly the same as spaces (except in quoted strings, see
	   below) turning the excess newlines into spaces will produce
	   this effect.  Bare newlines are illegal in quoted strings,
	   so this will not have any effect on LEGAL code.  Backslash
	   quoted newlines are allowed in quoted strings, but are
	   ignored, thus the preprocessor can ignore both the backslash
	   and the newline in this case.

Diff listing (please note, this is NOT my usual coding style).  The
	   line numbers will probably not match your cpp.c, we have run
	   ours through a pretty printer.  (these changes all occur in
	   the routine subst()).

*** /tmp/,RCSt1006926	Wed Apr 27 10:26:41 1983
--- cpp.c	Wed Apr 27 10:22:55 1983
***************
*** 864,865
  	char *actual[MAXFRM]; /* actual[n] is text of nth actual */
  	char acttxt[BUFSIZ]; /* space for actuals */

--- 864,866 -----
  	char *actual[MAXFRM]; /* actual[n] is text of nth actual */
+ 	char actused[MAXFRM]; /* for newline processing in actuals */
  	char acttxt[BUFSIZ]; /* space for actuals */
***************
*** 865,866
  	char acttxt[BUFSIZ]; /* space for actuals */
  

--- 866,868 -----
  	char acttxt[BUFSIZ]; /* space for actuals */
+ 	int nlines;
  
***************
*** 903,905
  				if (pa>= &actual[MAXFRM]) ppwarn(match,sp->name);
! 				else *pa++=ca;
  			}

--- 905,907 -----
  				if (pa>= &actual[MAXFRM]) ppwarn(match,sp->name);
! 				else { actused[pa-actual]=0; *pa++=ca; }
  			}
***************
*** 905,906
  			}
  		}

--- 907,911 -----
  			}
+ 			nlines = lineno[ifno] - maclin;
+ 			lineno[ifno] = maclin;	/* don't count any newlines */
+ 						/* they get counted later */
  		}
***************
*** 919,921
  				if (bob(p)) {outp=inp=p; p=unfill(p);}
! 				*--p= *ca;
  			}

--- 924,932 -----
  				if (bob(p)) {outp=inp=p; p=unfill(p);}
! 				/* 
! 				 * Actuals with newlines screw up line #s
! 				 */
! 				if (*ca == '\n' && actused[*vp-1])
! 				    if (*(ca-1) == '\\') ca--;
! 				    else *--p = ' ';
! 				else { *--p= *ca; if (*ca == '\n') nlines--; }
  			}
***************
*** 921,923
  			}
! 		} else break;
  	}

--- 932,940 -----
  			}
! 			actused[*vp-1] = 1;
! 		} else {
! 		    if (nlines > 0)
! 			while (nlines-- > 0)
! 			    *--p = '\n';
! 		    break;
! 		}
  	}