[ZODB-Dev] Re: [Zope-Coders] Zope test results

[Toby Dickenson]

On Thu, 2 May 2002 15:40:46 +0100, Toby Dickenson
<tdickenson@geminidataloggers.com> wrote:

>The test uses a set of ten objects, a cache size of five. It checks that=
 the=20
>least-recently-used ten objects get deactivated by the cache, and the=20
>least-recently-used five do not.
>
>However, that relies on usage order *within* *one* *transaction*. I =
wonder=20
>whether the transaction commit process might not always touch objects in=
 the=20
>same order?

Yes, thats definitely it.

The order these objects were being registered with the transaction
depends on their order in the root object's dictionary, as seen when
pickling it. Im not exactly sure why that leads to undefined
behaviour, but it definitely means the test was bogus.

I have comitted a new test that uses several transactions, without
relying on the order in which objects are accessed within one
transaction.=20

Toby Dickenson
tdickenson@geminidataloggers.com