From: Edwin Fine <efine145-nospam01@...>
Date: 2006-11-24T01:45:05+09:00
Subject: Re: ruby-prof on amd-64?

Daniel,

As you probably know, most 64-bit processors use the LP64 model (i.e. 
ints are 32 bits, longs and pointers 64 bits). Many things that people 
assume are ints on a 32-bit system actually are longs on a 64 bit 
system. This is why people should really use the C types properly, in 
other words, size_t for lengths and ptrdiff_t for pointer arithmetic, 
instead of int. size_t is unsigned on 32-bits, but unsigned long on 64 
bits. Also, I think thread ids are 32-bit on 32-bit systems, but 64-bit 
on 64-bit systems. Look at the type of pthread_t in sys/types.h on your 
AMD-64 system.

So the following code from ruby_prof.c is going to present problems on 
64 bit systems:

typedef struct {
    /* Cache hash value for speed reasons. */
    st_data_t key;
    VALUE klass;
    ID mid;
    int thread_id; /* Edwin says: should be pthread_t thread_id*/
... etc ...
}

Also, this code could be an issue:

static inline prof_data_t *
stack_push(prof_stack_t *stack)
{
    if (stack->ptr == stack->end) {
	int len, new_capa; /* Edwin says: Naughty, naughty - should use 
ptrdiff_t len, not int len*/
	len = stack->ptr - stack->start; /* Edwin says: result is actually 
ptrdiff_t, not int (ptrdiff_t will be unsigned long on LP64)*/
	new_capa = (stack->end - stack->start) * 2;
	REALLOC_N(stack->start, prof_data_t, new_capa);
	stack->ptr = stack->start + len;
	stack->end = stack->start + new_capa;
    }
    return stack->ptr++;
}

Judicious changes of int to a suitable portable data type would most 
likely solve these issues.

-- 
Posted via http://www.ruby-forum.com/.