[evolution-patches] Re: [patch] another 64-bit bug-fix

[Not Zed <notzed ximian com>]

This code should really be removed, and it should use e-memory, and the e-memory stuff fixed instead.  

On Tue, 2004-12-14 at 23:09 -0800, David Mosberger wrote:
> The patch below fixes more 64-bit problems, this time in camel-mime-parser.c.
>
> The problem there is two-fold:
>
>  (1) STRUCT_ALIGN is set to only 4.  This implies that on 64-bit
>      machines, we may end up with misaligned pointers, which is slow
>      on many RISC-type architectures and also on ia64.  The fix is to
>      change the alignment to sizeof (void *).

>  (2) Even with STRUCT_ALIGN set to sizeof (void *), we may still end
>      up with misaligned pointers because the code doesn't take into account
>      struct alignment.  To fix this, I removed the "data[1]" members from
>      MemPoolNode/MemPoolThresholdNode structures and used pointer-arithmetic
>      to calculate the starting address of the data portion.

> With these changes, there are no misaligned accesses anymore.
>
> Thanks,
>
> 	--david
>
> PS: I developed the patch for evolution v2.0.3, but the code in question
>     appears mostly unchanged in CVS and I verified that the patch applies
>     cleanly against the CVS version.
>
> --- camel-mime-parser.c~	2004-12-05 23:46:50.000000000 -0800
> +++ camel-mime-parser.c	2004-12-14 22:23:14.142731476 -0800
> @@ -57,7 +57,7 @@
>  
>  #define MEMPOOL
>  
> -#define STRUCT_ALIGN 4
> +#define STRUCT_ALIGN (sizeof (void *))
>  
>  #ifdef PURIFY
>  int inend_id = -1,
> @@ -76,12 +76,10 @@
>  	struct _MemPoolNode *next;
>  
>  	int free;
> -	char data[1];
>  } MemPoolNode;
>  
>  typedef struct _MemPoolThresholdNode {
>  	struct _MemPoolThresholdNode *next;
> -	char data[1];
>  } MemPoolThresholdNode;
>  
>  typedef struct _MemPool {
> @@ -116,10 +114,10 @@
>  	if (size>=pool->threshold) {
>  		MemPoolThresholdNode *n;
>  
> -		n = g_malloc(sizeof(*n) - sizeof(char) + size);
> +		n = g_malloc(sizeof(*n) + size);
>  		n->next = pool->threshold_blocks;
>  		pool->threshold_blocks = n;
> -		return &n->data[0];
> +		return n + 1;
>  	} else {
>  		MemPoolNode *n;
>  
> @@ -127,16 +125,16 @@
>  		while (n) {
>  			if (n->free >= size) {
>  				n->free -= size;
> -				return &n->data[n->free];
> +				return (char *) (n + 1) + n->free;
>  			}
>  			n = n->next;
>  		}
>  
> -		n = g_malloc(sizeof(*n) - sizeof(char) + pool->blocksize);
> +		n = g_malloc(sizeof(*n) + pool->blocksize);
>  		n->next = pool->blocks;
>  		pool->blocks = n;
>  		n->free = pool->blocksize - size;
> -		return &n->data[n->free];
> +		return (char *) (n + 1) + n->free;
>  	}
>  }
>

--

Michael Zucchi <notzed ximian com> "Ride, Work, Sleep. Beer." Novell's Evolution and Free Software Developer