Re: [BUG] Bad #define, nonportable C, missing {}

[Andreas Dilger <adilger@turbolabs.com>]

On Nov 27, 2001  16:03 -0200, vda wrote:
> On Monday 26 November 2001 18:28, Alan Cox wrote:
> > Nothing to do with me 8). I didnt write that bit of the i2o code. I agree
> > its both confusing and buggy. Send a fix ?
> 
> This is a test to be sure my replacement is equivalent:
> --------------------
> #include <stdio.h>
> #define MODINC(x,y) (x = x++ % y)
> #define MODULO_INC(x,y) ((x) = ((x)%(y))+1)

Ugh, clearly the code is broken, so we don't want equivalent code, but
correct code.  Use the unambiguous "MODINC(x,y) ((x) = ((x) + 1) % (y))"
form and not "have a bug that has properly defined behaviour under ANSI C".

Just looking at the code, it is fairly clear that the desire is to keep
q_in and q_out >= 0 and < I20_EVT_Q_LEN, which is the size of the event_q
array.  With the buggy version, it is possible that you could have q_in or
q_out == I2O_EVT_Q_LEN, which is overflowing the array.  Bad, bad, bad.

--- i2o_config.c.new	Mon Oct 22 13:39:56 2001
+++ i2o_config.c.orig	Tue Nov 27 16:03:19 2001
@@ -45,7 +45,7 @@
 static spinlock_t i2o_config_lock = SPIN_LOCK_UNLOCKED;
 struct wait_queue *i2o_wait_queue;

-#define MODINC(x,y) (x = x++ % y)
+#define MODINC(x,y) ((x) = ((x) + 1) % (y))

 struct i2o_cfg_info
 {

Cheers, Andreas
--
Andreas Dilger
http://sourceforge.net/projects/ext2resize/
http://www-mddsp.enel.ucalgary.ca/People/adilger/