[ZODB-Dev] FileStorage iteration data records missing data_txn

[Christian Theune]

Hi,

On Fri, 2009-01-09 at 11:39 -0500, Jim Fulton wrote:
> On Jan 9, 2009, at 7:41 AM, Christian Theune wrote:
> 
> > Hi,
> >
> > in ZEORaid we switched from using `restore` to `store` for replaying
> > transactions from another storage and found that the data records that
> > FileStorage returns aren't filling in the `data_txn` fields correctly,
> > but end up writing `None` all the time.
> 
> What makes you think this is incorrect?
> 
> data_txn should:
> 
> 1. Not be in the interface.  It is an optimization that depends on a  
> peculiarity of FileStorage of a case that only arises (today) with  
> undo.  I'm inclined to remove this optimization. In fact, in the  
> presence of blobs, the optimization is dubious.  This implies that the  
> prev_txn argument to restore should be ignored.
> 
> 2. data_txn should only be other than None for a record that was part  
> of an undo.  It is used to avoid copying data. Rather than copying  
> data, we simply record the position of the earlier record that  
> contains the data.

My thought was based on the presumption that the DataRecord interface was
described correctly (which it isn't as you point out below).

> > I'm not exactly sure how to fix this, but attached is a patch that  
> > makes
> > it work for the cases I could come up with.
> >
> > I'd like to add this change to the trunk (I'll write tests if this
> > change gets accepted).
> 
> 
> Your patch is incorrect.  data_txn is not the previous record  
> transaction id.  It is the transaction id of the transaction  
> containing data when the current record doesn;t contain any.  It is  
> used as an optimization by referring to data stored in an earlier  
> record rather than copying data to the current record.
> 
> The relevant interface didn't originally mention this field. The  
> documentation you added is incorrect. I'm going to remove the field  
> (along with version and tid) from the interface.

Ah. That's the source of the misunderstanding, thanks for clearing that
up. This sounds like I should also clean up the DataRecord in the
BaseStorage because that also exposes data_txn, and storage iteration
protocol on ZEO transports this attribute as well.

Looks like I've got to use loadBefore() to find the transaction that an
object's modification was based on.

Christian

-- 
Christian Theune · ct at gocept.com
gocept gmbh & co. kg · forsterstraße 29 · 06112 halle (saale) · germany
http://gocept.com · tel +49 345 1229889 7 · fax +49 345 1229889 1
Zope and Plone consulting and development
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: This is a digitally signed message part
Url : http://mail.zope.org/pipermail/zodb-dev/attachments/20090110/0752c41e/attachment.bin