Re: [patch] read-rtx.c: validate constants

Re: [patch] read-rtx.c: validate constants

[DJ Delorie]

> On Mon, Nov 19, 2001 at 01:19:00AM -0500, DJ Delorie wrote:
> > +       fprintf (stderr, "invalid decimal constant \"%s\"\n", string);
> > +       abort();
> 
> Use fatal_with_file_and_line.  Otherwise ok.

Thanks!  Applied.

Re: [patch] read-rtx.c: validate constants

[Richard Henderson]

On Mon, Nov 19, 2001 at 01:19:00AM -0500, DJ Delorie wrote:
> +       fprintf (stderr, "invalid decimal constant \"%s\"\n", string);
> +       abort();

Use fatal_with_file_and_line.  Otherwise ok.


r~

[patch] read-rtx.c: validate constants

[DJ Delorie]

It would be better to allow hex constants (which is a task unto
itself), but we'd still need to validate that the given string is a
valid constant.  Bootstrapped on ix86-linux.

2001-11-19  DJ Delorie  <dj@redhat.com>

	* read-rtl.c (ISDIGIT, ISSPACE): Make sure we have these.
	(validate_const_int): New.
	(read_rtx): Validate constant integers.
	* config/i386/i386.md (pmulhrwv4hi3): Use decimal constants.

Index: read-rtl.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/read-rtl.c,v
retrieving revision 1.6
diff -p -3 -r1.6 read-rtl.c
*** read-rtl.c	2001/10/11 03:15:55	1.6
--- read-rtl.c	2001/11/19 05:51:49
*************** Software Foundation, 59 Temple Place - S
*** 25,30 ****
--- 25,36 ----
  #include "obstack.h"
  #include "hashtab.h"
  
+ #ifndef ISDIGIT
+ #include <ctype.h>
+ #define ISDIGIT isdigit
+ #define ISSPACE isspace
+ #endif
+ 
  #define	obstack_chunk_alloc	xmalloc
  #define	obstack_chunk_free	free
  
*************** static void read_escape		PARAMS ((struct
*** 41,46 ****
--- 47,53 ----
  static unsigned def_hash PARAMS ((const void *));
  static int def_name_eq_p PARAMS ((const void *, const void *));
  static void read_constants PARAMS ((FILE *infile, char *tmp_char));
+ static void validate_const_int PARAMS ((const char *));
  
  /* Subroutines of read_rtx.  */
  
*************** traverse_md_constants (callback, info)
*** 494,499 ****
--- 501,530 ----
      htab_traverse (md_constants, callback, info);
  }
  
+ static void
+ validate_const_int (string)
+      const char *string;
+ {
+   const char *cp;
+   int valid = 1;
+ 
+   cp = string;
+   while (*cp && ISSPACE(*cp))
+     cp++;
+   if (*cp == '-' || *cp == '+')
+     cp++;
+   if (*cp == 0)
+     valid = 0;
+   for (; *cp; cp++)
+     if (! ISDIGIT (*cp))
+       valid = 0;
+   if (!valid)
+     {
+       fprintf (stderr, "invalid decimal constant \"%s\"\n", string);
+       abort();
+     }
+ }
+ 
  /* Read an rtx in printed representation from INFILE
     and return an actual rtx in core constructed accordingly.
     read_rtx is not used in the compiler proper, but rather in
*************** again:
*** 699,704 ****
--- 730,736 ----
  
        case 'w':
  	read_name (tmp_char, infile);
+ 	validate_const_int(tmp_char);
  #if HOST_BITS_PER_WIDE_INT == HOST_BITS_PER_INT
  	tmp_wide = atoi (tmp_char);
  #else
*************** again:
*** 720,725 ****
--- 752,758 ----
        case 'i':
        case 'n':
  	read_name (tmp_char, infile);
+ 	validate_const_int(tmp_char);
  	tmp_int = atoi (tmp_char);
  	XINT (return_rtx, i) = tmp_int;
  	break;
Index: config/i386/i386.md
===================================================================
RCS file: /cvs/gcc/gcc/gcc/config/i386/i386.md,v
retrieving revision 1.314
diff -p -3 -r1.314 i386.md
*** i386.md	2001/11/16 19:58:40	1.314
--- i386.md	2001/11/19 05:51:51
***************
*** 19676,19685 ****
  	            (sign_extend:V4SI
  		       (match_operand:V4HI 2 "nonimmediate_operand" "ym")))
  	      (vec_const:V4SI
! 	         (parallel [(const_int 0x8000)
! 			    (const_int 0x8000)
! 			    (const_int 0x8000)
! 			    (const_int 0x8000)])))
  	   (const_int 16))))]
    "TARGET_3DNOW"
    "pmulhrw\\t{%2, %0|%0, %2}"
--- 19676,19685 ----
  	            (sign_extend:V4SI
  		       (match_operand:V4HI 2 "nonimmediate_operand" "ym")))
  	      (vec_const:V4SI
! 	         (parallel [(const_int 32768)
! 			    (const_int 32768)
! 			    (const_int 32768)
! 			    (const_int 32768)])))
  	   (const_int 16))))]
    "TARGET_3DNOW"
    "pmulhrw\\t{%2, %0|%0, %2}"

Re: [patch] read-rtx.c: validate constants

[Kaveh R. Ghazi]

 > 2001-11-19  DJ Delorie  <dj@redhat.com>
 > 
 > 	* read-rtl.c (ISDIGIT, ISSPACE): Make sure we have these.
 > 
 > Index: read-rtl.c
 > ===================================================================
 > RCS file: /cvs/gcc/gcc/gcc/read-rtl.c,v
 > retrieving revision 1.6
 > diff -p -3 -r1.6 read-rtl.c
 > *** read-rtl.c	2001/10/11 03:15:55	1.6
 > --- read-rtl.c	2001/11/19 05:51:49
 > *************** Software Foundation, 59 Temple Place - S
 > *** 25,30 ****
 > --- 25,36 ----
 >   #include "obstack.h"
 >   #include "hashtab.h"
 >   
 > + #ifndef ISDIGIT
 > + #include <ctype.h>
 > + #define ISDIGIT isdigit
 > + #define ISSPACE isspace
 > + #endif

I'm pretty sure this hunk is not needed, you should rely on
safe-ctype.h (from system.h) here.  Gcc arranges for a "build" copy of
safe-ctype.o to be linked in to the "generator" programs too.

		--Kaveh
--
Kaveh R. Ghazi			Engagement Manager / Project Services
ghazi@caip.rutgers.edu		Qwest Internet Solutions

Re: [patch] read-rtx.c: validate constants

[DJ Delorie]

> I'm pretty sure this hunk is not needed, you should rely on
> safe-ctype.h (from system.h) here.  Gcc arranges for a "build" copy of
> safe-ctype.o to be linked in to the "generator" programs too.

Another one of those?  I guess a build-libiberty should be next on my
to-do list, then ;-)