Re: [PATCH 01/12] staging: lustre: fid: Use !x to check for kzalloc failure

[Julia Lawall <julia.lawall@lip6.fr>]

On Tue, 23 Jun 2015, Dan Carpenter wrote:

> Yes.  I know Al's thoughts and kernel style.
>
> But Alan Cox and Andreas have both said they think (x = NULL) can help
> you avoid some kind of boolean vs pointer bugs.  I've had co-workers who
> did massive seds changing !foo to foo = NULL on our code base.  But
> I've never seen a real life example of a bug this fixes.
>
> To be honest, I've never seen a real life proof that (!foo) code is less
> buggy.  I should look through the kbuild mailbox...  Hm...  But my other
> idea of setting up code style readability testing website is also a good
> one.
>
> Linux kernel style is based on Joe Perches finding that 80% of the code
> prefers one way or the other.  That's a valid method for determining
> code style.  I bet it normally picks the more readable style but it
> would be interesting to measure it more formally.

On today's linux-next, I find 3218 tests on the result of kmalloc etc
using NULL and 14429 without, making 82% without.  The complete semantic
patch is shown below.

julia

@initialize:ocaml@
@@

let withnull = ref 0
let withoutnull = ref 0

@r1 disable is_null, isnt_null1 exists@
expression x,e;
position p;
statement S1,S2;
@@

x = \(kmalloc\|kzalloc\|kcalloc\|devm_kmalloc\|devm_kzalloc\)(...)
... when != x = e
    when != &x
if@p (<+...\(x = NULL\|x != NULL\|NULL = x\|NULL != x\)...+>) S1 else S2

@r2 disable not_ptr1, not_ptr2 exists@
expression x,e;
position p;
statement S1,S2;
@@

x = \(kmalloc\|kzalloc\|kcalloc\|devm_kmalloc\|devm_kzalloc\)(...)
... when != x = e
    when != &x
if@p (<+...\(!x\|x && ...\|x || ...\)...+>) S1 else S2

@script:ocaml@
_p << r1.p;
@@
withnull := !withnull + 1

@script:ocaml@
_p << r2.p;
@@
withoutnull := !withoutnull + 1

@finalize:ocaml@
@@

Printf.printf "withnull %d withoutnull %d\n" !withnull !withoutnull