Tomos at Wikipedia wrote:
Brion-
Thanks for the explanation. But I think the access counter is behaving a bit differently than I understood from your posting.
*When I (as a logged-on user) refresh a page, either using F5 or Refresh button on IE, it does increase the count.
I assure you it works as I described in Mozilla 1.3 for FreeBSD:
* first logged-in visit increases counter * second load does not * reload does not * force reload does * new view of page after editing does increase count * reload after does not
*When I do those things without logging-on, count never increases (because there is no change made)
Right.
So the logged-on users' part seems to be different from what you explained. Is this an error?
Yes. Either IE is sending the wrong headers, or we're not understanding what it's sending, or it's not understanding what we're sending. Could you specify exactly which version of IE, on what operating system, you're using? Hopefully this can be tracked down...
(Anyone know of a convenient tool for sniffing the headers out of an HTTP connection? I need to be able to see what PHP+Apache actually send out, as it's not always the same as we instructed it...)
I personally (not speaking on behalf of other Japanese Wikipedians) feel that counting unique views would make more sense.
Yes, it would. The page count system is poorly designed to begin with, and interacts poorly with caching at present.
If I am understanding the way things are, it would be approximately:
[# of times a server cache is sent (to logged-on or IP users)] + [# of times a page is cashed to a server (=# of changes made)]
So, for example, is it feasible to add the [# of time a server cache is sent] to [the counter value] whenever someone makes an edit to a page?
We'd have to store the value somewhere for that to happen. At that point, may as well just move the counter increment to before the cache send).
-- brion vibber ( brion @ pobox.com )