shared stat cache

[tridge at samba.org]

Volker,

I've now confirmed that the windows behaviour is just due to slow
flush of an internal cache. I changed your test code to look at the
write time once a second after a write, and it seems to take about 3
to 6 seconds before it shows up. This might imply a 10 second flush
cycle.

Here is the output of the new test on win2003:

Initial write time Fri Oct 15 11:08:58 2004 EST
write time Fri Oct 15 11:08:58 2004 EST
write time Fri Oct 15 11:08:58 2004 EST
write time Fri Oct 15 11:08:58 2004 EST
write time Fri Oct 15 11:08:58 2004 EST
write time Fri Oct 15 11:08:58 2004 EST
write time Fri Oct 15 11:09:06 2004 EST
write time Fri Oct 15 11:09:06 2004 EST
write time Fri Oct 15 11:09:06 2004 EST

the lines are printed at one second intervals. You can see that
win2003 updated the stat info after 6 seconds. The smallest delay I've
seen is 2 seconds.

I'd be surprised if this is critcal to Excel behaviour, but it is
possible. For it to be critical windows would have to be guaranteeing
some minimum delay in the update to the write time (otherwise excel
would sometimes break on windows). 

I don't have Excel 2000, but I'd suggest you use the "cifs" backend in
Samba4 to setup the following:

 excel on windows client -> samba4-proxy -> win2003

then modify the proxy to update the write time immediately by
remembering the last write time. Then see if Excel misbehaves with
this setup. If it does then you know this behaviour is critical,
otherwise you know that the problem is caused by something else.

Cheers, Tridge