AFAICT, no statistics are recorded for the period between when the WebStatistics are collected on the last day of the month and when they are collected for the first day of the next month.

For example, compare Wikilearn.WebStatistics rev. 1.493 (1 Dec 2002) to rev. 1.492 (30 Nov) (and, for the previous month, rev. 1.463 to rev. 1.462). Note that the statistics for the previous month do not change between the revisions, and the statistics for the (new) current month are zero. I:

• am quite certain I viewed and updated pages in the 24 hour period between 30 Nov and 1 Dec -- in fact, at least these three pages were changed:
  • Wikilearn.ForwardingX 30 Nov 2002 - 21:20 - NEW
  • Wikilearn.ComputerThingsToAccomplish 30 Nov 2002 - 19:55 - r1.3
  • Wikilearn.GNUShortcuts 30 Nov 2002 - 16:02 - r1.10
• don't recall any prior mention or discussion of this problem so have added this report on Codev

This is occurring on twiki.org.

(Most of the text above is also on Wikilearn.WebStatisticsDiscussion)

-- RandyKramer - 01 Dec 2002

What is probably happening is that the stats script is running at 0:00 PST, i.e. 8:00 GMT, on the 30 Nov in this case. Since TWiki tends to work in GMT, the next time it runs at 8:00 is on 1 Dec, meaning it misses the 16 hours from 8:00 GMT to end of day on 30 Nov. All the times you mentioned fall within this period.

Maybe PeterThoeny can clear this up - the crontab is probably running under his twiki.sf.net userid, so I can't check it myself. Anyway, it would be wise to always run statistics from cron at 23:50 GMT (assuming it takes max 10 mins to run, should adjust for busier or quieter sites), so that it picks up the stats for that (GMT) day correctly.

-- RichardDonkin - 01 Dec 2002

The cron job was running at 5 minutes past midnight, I changed it to 10 minutes before midnight. Let me know in case this did not solve the problem.

Peter, Thanks! I probably won't be able to notice anything until the end of next month, but if it occurs again I'll let you know. (And, of course, kudos to Sven for undertaking to fix the root problem.) -- RandyKramer - 02 Dec 2002

-- PeterThoeny - 02 Dec 2002

Formally speaking, this is still a bug.... If you are running the stats script by hand you have no way of updating last months stats after the fact. Can we make it robust enough to do the previous month and the current month if the current month is new?

-- SvenDowideit - 02 Dec 2002

If you are concerned about this, just remove the link to run the stats script by hand. Of course, patches are welcome if someone would like to fix this

-- RichardDonkin - 02 Dec 2002

It's ugly, and I can see why no-one wants to play with it.. but I have done a change that works. its attached..

Should we be considering changing the way it works to better fit into the Forms/MetaData way of working? And removing the hardcoded output from the script?

-- SvenDowideit - 02 Dec 2002

Thanks for the patch, but please see PatchGuidelines for details on doing a context diff patch which can be applied automatically, as linked above - and also say which version you are diff'ing against. However, I don't really want to make significant changes before the upcoming BeijingRelease, particularly considering that running the cron job just before midnight GMT is such an easy workaround. Testing the statistics script is not that easy since we don't have proper unit tests...

-- RichardDonkin - 02 Dec 2002

oops, i just did a cvs diff on cvs head... here's a cvs diff -c

... whats the classification supposed to be? its not resolved, but there is a workaround.. and I agree, it will need testing, unless we re-write.

-- SvenDowideit - 02 Dec 2002