On Wed, 13 Mar 2002 11:37:00 +0100, Oliver Bleutgen <myzope@gmx.net> wrote:
- I did paste only the objects which have such a high count (next one was approx 200).
Hmm. Ill have to ponder what that means.
- This server (2.3.3) has no special products on it, just 2 Folderish ZClasses, which are heavily used (they construct the whole site).
So, can we rule out that there's a bug in the zclasses machinery which is the cause for this?
No, thats not impossible
I.e. should there be some zclass related object in the above list otherwise?
Not necessarily. The class with the bug need not leave an instance of himself tied up in his immortal cyclic trash.
Is it right that in Zope 2.3.x (and others?) usage of <dtml-var "REQUEST.set(...)"> can lead to such leakage?
Yes, thats true in all versions.
Is there a description somewhere what the basic causes of such leakages are? I.e. only bugs in python c-code/zope c-code?
No, its possible for a bug in through-the-web edited dtml to cause this.
Is there some text about what to not do in python product development to avoid introducing leakage? For example, on #zope someone pointed out to me that storing acquisition wrappers in the ZODB will lead to leakage.
In old versions (roughly before 2.3 or 2.4; I forget exactly when) it was possible to create a cyles involving a persistent object having a reference to an acquisition wrapper. Today that raises an exception. Toby Dickenson tdickenson@geminidataloggers.com