[ZODB-Dev] Storage API corner cases

[Barry Warsaw]

On Mon, 2003-05-26 at 04:16, Toby Dickenson wrote:

> Firstly, how should getSerial behave when the current revision of an object 
> has had its creation undone? The implementation in BaseStorage suggests that 
> this case will raise a KeyError, but both FileStorage and BDBFull replace 
> this base class method with one that returns the id of the revision in which 
> its creation was undone.

The interface isn't very helpful here. :)  I think getSerial() probably
ought to raise a KeyError when called on an uncreated object.

> A comment in zeoVerify suggests that ZEO is assuming that a KeyError will be 
> raised, but its implementation will work with either behavior.

Hmm.

> Is getSerial used anywhere else? Is there a reason to require the FileStorage 
> behaviour, or should we document both behaviours as acceptable?

I don't think both behaviors should be allowed.  We should pick one,
document it, add test cases, and fix the code.

> Secondly, there are differences between storages in the list of transaction 
> returned from undoLog: FileStorage includes some transactions that can never 
> be undone because the transaction includes an object for which no previous 
> history exists. DirectoryStorage omit these.

The previous history is missing because of a pack, it's the first
creation of an object, or some other reason?

> This is no problem if undoLog is only used to populate an undo GUI, but could 
> lead to problems if undoLog is used for introspection. Is this a problem for 
> anyone?

-Barry