On 09/06/2010 04:25 PM, Amos Jeffries wrote:
On Mon, 06 Sep 2010 09:41:15 -0600, Alex Rousskov
<[email protected]> wrote:
On 09/06/2010 01:15 AM, Amos Jeffries wrote:
Inspired from the SuSE patch submitted by Christian. This is what I
think should be happening with real ptr maths instead of obsolete
integer math with potential rounding and endian errors.
void
memNodeWriteComplete(void* d)
{
mem_node* n = (mem_node*)(d - _mem_node_data_offset);
GCC: pointer of type void * used in arithmetic
The value of a (pointer + n) expression depends on what pointer is
pointing to. If you want byte-size increments, you have to cast to char*
or similar.
Meh, one cast too many.
Using ptrdiff_t instead of int is the right thing to do. I do not thing
you need to cast to it in makeMemNodeDataOffset. You may clarify the
intent by writing:
// calculate data member offset; p is not dereferenced here
mem_node *p = 0;
return reinterpret_cast<char*>(&p->data) - reinterpret_cast<char*>(p);
The point of using ptrdiff_t with p==0L is to reduce multiple casting to
non-pointer types.
Sorry, I do not understand what you mean. You have to use char* pointers
for the pointer difference to compile and for the value to be correct.
I am _for_ using ptrdiff_t to store that value. I was just pointing out
that you do not have to explicitly cast to it.
Writing reinterpret_cast<char*>() is the same as writing (char*)(). Just
slightly safer when types change and more search-friendly.
Hope this clarifies,
Alex.
