[#COR-1097] Corrupted class index in some delete/re-add scenarios

db4o Core

Corrupted class index in some delete/re-add scenarios

Created: 15/Jan/08 06:20 PM Updated: 23/Jan/08 03:25 PM

Component/s:

None

Affects Version/s:

7.0.25

Fix Version/s:

7.1.26

Time Tracking:

Original Estimate:

Not Specified

Remaining Estimate:

Time Spent:

15h

File Attachments:

RepeatDeleteReaddTestCase.java (3 kb)

Peers:	Andrew Zhang and Rodrigo B. de Oliveira
Order:	3
Original IDS Estimate:	4 IDS
Resolution Date:	23/Jan/08 03:25 PM
First Response Date:	16/Jan/08 11:57 AM
Labels:
Participants:	Andrew Zhang, Patrick Roemer and Rodrigo B. de Oliveira
Number of Attachments:	1
Number of Comments:	6

Description

« Hide

When deleting and re-adding objects to a database, subsequent queries (constrained against the extent only) may not find all objects. The results are somewhat erratic: The number of found objects may vary on different runs, etc.

All

Comments

Work Log

Change History

Sort Order:

Ascending order - Click to sort in descending order

[ Permlink | « Hide ]

Patrick Roemer - 15/Jan/08 06:27 PM

The attached test case should be expected to generate the following output on each loop execution:

class com.db4o.db4ounit.common.defragment.RepeatDeleteReaddTestCase$ItemA/10
class com.db4o.db4ounit.common.defragment.RepeatDeleteReaddTestCase$ItemB/10

The actual output is something like:

class com.db4o.db4ounit.common.defragment.RepeatDeleteReaddTestCase$ItemA/10
class com.db4o.db4ounit.common.defragment.RepeatDeleteReaddTestCase$ItemB/10
class com.db4o.db4ounit.common.defragment.RepeatDeleteReaddTestCase$ItemA/10
class com.db4o.db4ounit.common.defragment.RepeatDeleteReaddTestCase$ItemB/10
class com.db4o.db4ounit.common.defragment.RepeatDeleteReaddTestCase$ItemA/8
! db4ounit.AssertionException: Expected '10' but was '8'

(The position where the failure occurs may vary in between runs.)

[ Permlink | « Hide ]

Rodrigo B. de Oliveira - 16/Jan/08 11:57 AM

And the more the CPU load the bigger the probability it finishes green.

[ Permlink | « Hide ]

Andrew Zhang - 18/Jan/08 10:27 AM

The test passes against Apache Harmony and JRockit.
Looks like a bug of SUN jre.
ObjectReference id_add(ObjectReference newRef) {
  newRef._idPreceding = null; // looks like these two lines are not executed.
  newRef._idSubsequent = null; // looks like these two lines are not executed.
  newRef._idSize = 1;
  return id_add1(newRef);
}

[ Permlink | « Hide ]

Andrew Zhang - 18/Jan/08 02:51 PM

The root node is not correctly initialized.

[ Permlink | « Hide ]

Patrick Roemer - 22/Jan/08 02:27 PM

Needs to be ported back to production branch.

[ Permlink | « Hide ]

Patrick Roemer - 23/Jan/08 02:32 PM

Ported back to 6.4