>> I've finally fixed the treatment of cached articles when calculating
actual
>> hi/lo water marks. The patch (among other stuff) can be found at the
usual
>> place:
...
>> The patch also adds code to take cached articles into account in
CMDlistgroup.
To let everyone know, it doesn't do anything for my problem where group
returns the right number/range of articles, but xover on the range returned
by group returns a fraction of the expected number.
Just to share with everyone else (I have discussed it with Tom), here is the
problem:
1. nuke xover cache
2. type:
# script
# telnet localhost nntp
group alt.binaries.warez.ibm-pc
xover lowmark-himark [as returned by group]
quit
# exit
3. Do wc -l on typescript (file created by script) to give an approximate
number of articles returned by xover; compare to the number that group
reported.
4. wait a few minutes, and repeat 2 and 3. After a while, you will find
xover returned only a small fraction of the actual number of articles in the
group (in my case, it goes from 40K down to sub-10K articles). If you nuke
the xover cache again, the problem will be resolved (for another few
minutes, until the cycle repeats).
Evan
