On Thu, 2008-06-26 at 01:26 +0300, Tomas Winkler wrote:

On Wed, Jun 25, 2008 at 7:12 PM, Johannes Berg
[off-list ref] wrote:

quoted

On Wed, 2008-06-25 at 11:58 -0400, Pavel Roskin wrote:

quoted

quoted

+                   unsigned long mid_range = (-1) / 2 + 1;

What is that?  I guess that's where Riemann's zeta function has its
non-trivial roots :-)

quoted

The Pavel hypothesis: "The Riemann hypothesis is correct when calculated
in 32-bit integers" is obviously wrong ;)

Welcome to the Desert of the Real

Welcome to C.  mid_range is 1, as you can easily check.

quoted

quoted

quoted

+                   time_diff = jiffies - bss->last_update > mid_range ?
+                           jiffies - bss->last_update :
+                           bss->last_update - jiffies;

That's pretty hairy.  Do we really lack a function to calculate time
difference?

This is simplest as you can get under assumption that times are not
apart more then 1/2 of UL.
 It would be overkill to translate it to timeval or anything we have
diff function for.

Why would jiffies ever lag behind bss->last_update?  When would
(bss->last_update - jiffies) make sense?  If it makes sense, how about
indicating it somehow?  It would be a value that decreases over time.  

quoted

Actually, come to think of it, isn't just doing the difference as in teh
original patch correct in 32-bit unsigned integers? It'll wrap around a
bit but that's ok, no?

It would make a very recent beacon to be like half an hour old on
wraps, wouldn't it?

No.  Why would it?  Can you give an example?  If you are not increasing
precision, wraparounds should make no difference.

-- 
Regards,
Pavel Roskin