Why *((type **)&p) = new_pointer is better?

[harvey]

Hi,
I'm reading kernel codes and saw some strange assigning statements like:

int *p = NULL, n = xxx;
*((int **)&p) = &n;

Why not to directly assign like

p = &n;

Is there any special reason for this?

[NickJohnson]

Perhaps this is an attempt to force p to not be optimized out of the stack frame or to force a write to memory. It's particularly odd that &p is being cast to int**, considering that it already is that type. I'm assuming you found this in Linux: where did you find it? It may be part of a more elaborate hack.

[harvey]

Perhaps this is an attempt to force p to not be optimized out of the stack frame or to force a write to memory. It's particularly odd that &p is being cast to int**, considering that it already is that type. I'm assuming you found this in Linux: where did you find it? It may be part of a more elaborate hack.

I saw this somewhere in Linux kernel some time ago and I can't remember exactly where it is. I need some time to search:-)

[Jezze]

Perhaps just a very bad programmer? =)

[Solar]

Perhaps just a very bad programmer? =)

A possibility never to be ruled out easily.

[harvey]

You might be reading it wrong.

Maybe...
I'll let you know if I find the place.

[AJ]

Hi,

Perhaps just a very bad programmer? =)

For me, declarations like:

Code: Select all

int *p = NULL, n = xxx;

always ring alarm bells anyway, even if, as in your example above, it was intended that n is not a pointer.

Cheers,
Adam

[harvey]

Hi,

Perhaps just a very bad programmer? =)

For me, declarations like:

Code: Select all

int *p = NULL, n = xxx;

always ring alarm bells anyway, even if, as in your example above, it was intended that n is not a pointer.

Cheers,
Adam

yes, n is a int, so I added & and the focus is actually on the second statement.

[AJ]

Hi,

even if, as in your example above, it was intended that n is not a pointer.

I'm not saying your example got it wrong. It may just be a matter of preference, but I was just saying that declaring a type and a pointer to that type in the same statement is not always the clearest way of doing it.

Cheers,
Adam