[Bug c/35587] New: -Warray-bounds does not work at all or does not find all trivial cases, and works only with -O2 or -O3

[Bug c/35587] New: -Warray-bounds does not work at all or does not find all trivial cases, and works only with -O2 or -O3

[karsten_burger at gmx dot de]

-Warray-bound is really a very interesting new feature, since me and some
colleagues recently spent many hours debugging a large legacy software, and
finally found C-char-array overrun errors as causes in most cases.

I used the following example:

#include <stdio.h>
int main()
{
  int a[5];
  a[-1] = 0;                  // line 5: error
  a[+1] = 0;  // ok
  a[+5] = 0;                  // line 7: error
  a[15] = 0;                  // line 8: error
  printf("%d \n", a[15]);     // line 9: error
  int i=15;
  printf("%d \n", a[i]);      // line 11: error
  return 0;
}

I used the following command line:
    /usr/local/gcc-4.3.0/bin/gcc-4.3.0  bounds.c -O2 -Warray-bounds

1. Test: 
  I get no warning at compile time. I retried it using -O0, -O1, -O3, and there
 is also no warning.

2. Test:
  I changed integer i in line 10 to a value of 16.
  Now I correctly get warnings in lines 5, 7, 8, 11, but not in line 9, stating
that the index is below/larger than the array permits.
  Interestingly, the warnings are only issued if I am using -O2 or -O3, but not
with -O0 and not with -O1.


Conclusion:
* -Warray-bound is not reliable at the moment, it may or may not issue warnings
or no warnings at all.
* When it issues warnings, at line 8 of the example it finds the incorrect
assignment to a[15], but at line 9 not the incorrect use of a[15]. 
* When it issues warnings and I change line 9 to a[16], it reports lines 5, 7,
8, 9 but not line 11.
* If this option works only with -O2 and -O3 due to technical reasons, it
should be documented clearly to avoid frustration of the user.


-- 
           Summary: -Warray-bounds does not work at all or does not find all
                    trivial cases, and works only with -O2 or -O3
           Product: gcc
           Version: 4.3.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: karsten_burger at gmx dot de
 GCC build triplet: i686-linux-gnu
  GCC host triplet: i686-linux-gnu
GCC target triplet: i686-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35587

[Bug c/35587] -Warray-bounds does not work at all or does not find all trivial cases, and works only with -O2 or -O3

[karsten_burger at gmx dot de]

------- Comment #1 from karsten_burger at gmx dot de  2008-03-14 17:07 -------
Created an attachment (id=15321)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=15321&action=view)
version which does not create any warning


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35587

[Bug c/35587] -Warray-bounds does not work at all or does not find all trivial cases, and works only with -O2 or -O3

[karsten_burger at gmx dot de]

------- Comment #2 from karsten_burger at gmx dot de  2008-03-14 17:08 -------
Created an attachment (id=15322)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=15322&action=view)
assembler output using gcc option -save-temps


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35587

[Bug c/35587] -Warray-bounds does not work at all or does not find all trivial cases, and :works only with -O2 or -O3

[karsten_burger at gmx dot de]

-- 

karsten_burger at gmx dot de changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |major


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35587

[Bug c/35587] -Warray-bounds does not work at all or does not find all trivial cases, and :works only with -O2 or -O3

[rguenth at gcc dot gnu dot org]

------- Comment #3 from rguenth at gcc dot gnu dot org  2008-03-17 14:05 -------
Hm, with gcc 4.2 I get

gcc-4.2 -c t.c -O2 -Wall
t.c: In function 'main':
t.c:5: warning: array subscript is below array bounds
t.c:7: warning: array subscript is above array bounds
t.c:8: warning: array subscript is above array bounds
t.c:9: warning: array subscript is above array bounds

(the warning in line 11 is missing as the constant is
propagated and the load from a[15] CSEd - this is a very hard
to fix implementation limit.  You'd fix the case in line 9
and then the issue in line 11 would pop up.)

with 4.3 I indeed see

gcc-4.3 -c t.c -O2 -Wall

(nothing)

because all invalid stores are deleted from the code and the only
used value (a[15]) is propagated from the store a[15] = 0 and thus
replaced by the constant 0.  Again a hard to fix implementation
issue - after all these invalid stores/loads invoke undefined
behavior which is why we just "optimized" them.

IMHO for these very simple cases the frontend should already warn
(even at -O0 and -O1) and not rely on the code surviving until VRP.
(The issue is that we in this way would also warn in dead code regions,
which may be a problem on its own).

Note this is not accepts-invalid as it is only undefined at runtime.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mueller at gcc dot gnu dot
                   |                            |org, rguenth at gcc dot gnu
                   |                            |dot org
           Severity|major                       |enhancement
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
           Keywords|accepts-invalid             |
   Last reconfirmed|0000-00-00 00:00:00         |2008-03-17 14:05:57
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35587

[Bug c/35587] -Warray-bounds does not work at all or does not find all trivial cases, and :works only with -O2 or -O3

[nikos42 at gmail dot com]

------- Comment #4 from nikos42 at gmail dot com  2008-05-03 09:43 -------
Confirmed with gcc 4.3.1. Complete version info:

Using built-in specs.
Target: i486-linux-gnu
Configured with: ../src/configure linux gnu
Thread model: posix
gcc version 4.3.1 20080401 (prerelease) (Debian 4.3.0-3) 

Regards,
Nicolas


-- 

nikos42 at gmail dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |nikos42 at gmail dot com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35587