On 19 Aug 2008, at 15:29, Andrew Jaquith wrote:
I really don't know what to say about this. We have lots and lots
of unit tests for ACLs, and many web unit tests that test things
from the browser.
All of these tests work fine.
Unless we can devise some tests that allow reproduction of the
issue, every time, it's going to be hard for me to help.
One can poke around and hope...
There are a couple of things which our unit tests do *not* test for:
1) Longevity. A long-running JSPWiki instance which is tested
against a known set of pages. Webtests almost do this, but they
don't do it repeatedly. This should find caching issues. For
example, on those sites that I run, JSPWIKI-27 appears only after a
few hours. But it never appears on jspwiki.org, even though they run
exactly the same codebase.
2) Multithreading. If there are any cases where synchronization is a
problem, it could pop up here. All of our tests are single-thread
only. This means that it would appear more often under load.
My guess is that this particular Heisenbug is either of these, and it
relates to caching somehow. Florian, does setting the cache size to
1 help (don't turn it off, just set the size to 1.)?
/Janne
