Revision: 992
Author: sbigaret
Date: 2006-04-15 02:44:54 -0700 (Sat, 15 Apr 2006)
ViewCVS: http://svn.sourceforge.net/modeling/?rev=992&view=rev
Log Message:
-----------
was not under subversion control
Added Paths:
-----------
trunk/ProjectModeling/Modeling/doc/bug861048.txt
Added: trunk/ProjectModeling/Modeling/doc/bug861048.txt
===================================================================
--- trunk/ProjectModeling/Modeling/doc/bug861048.txt (rev 0)
+++ trunk/ProjectModeling/Modeling/doc/bug861048.txt 2006-04-15 09:44:54 UTC (rev 992)
@@ -0,0 +1,77 @@
+Bug #861048: Invalid FK constraints in generated DB schema
+==========================================================
+
+:Author: `S\xE9bastien Bigaret`_
+
+.. _`S\xE9bastien Bigaret`: mailto:Seb...@en...
+
+
+The bug described here cannot only be fixed with a patch against the `Modeling
+framework` because it affects database schemas that have been created by the
+framework v0.9pre16 or earlier versions. Worse, **it can lead to silent data
+loss when used with postgresql and the ``psycopg`` python adaptor**.
+
+You'll find here a full description of the bug, along with instructions on how
+existing database schemas can be fixed.
+
+
+.. _`Modeling framework`: http://modeling.sourceforge.net
+
+.. contents::
+
+Description of the bug
+----------------------
+
+Generation of the database schema sometimes produces foreign key constraints
+that are invalid wrt the model.
+
+This happens when the model has a to-one relationship pointing to an
+entity which has at least a sub-entity. Such a model::
+
+ model.version='0.1'
+ model.entities = [ # A-------- b -->B
+ Entity('A'), # / \
+ Entity('B'), # /-+-\
+ Entity('C', parent='B'), # |
+ ] # |
+ model.associations=[ # C
+ Association('A','B', relations=['b', 'as'] ),
+ ]
+
+generates a database schema which contains::
+
+ [...]
+ ALTER TABLE A ADD CONSTRAINT b FOREIGN KEY (FK_B) \
+ REFERENCES B(ID) INITIALLY DEFERRED;
+ [...]
+
+This constraint is not valid, because an instance of ``A`` could reference
+either a ``B`` or a ``C``.
+
+:Consequence:
+ Since some FK constraint is invalid, you will not be able to save changes in
+ an ``EditingContext`` when an ``A`` is in relation to a ``C``, just because
+ this violates the constraint.
+
+
+Possible data loss with ``psycopg``
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+Now if you use Postgresql and ``psycopg``, the problem is that this bug hits
+an other bug in psycopg v1.1.10 or earlier, where errors at commit time are
+silently ignored. As a consequence, if you save changes on an
+``EditingContext`` which as an ``A`` in relation to a ``C``, the postgresql
+database server will reject the commit (because of the buggy FK
+constraint). But ``psycopg`` silently ignores the exception, as a consequence
+**the changes appears to be safe but none of the changes was committed in the
+database**.
+
+For this reason, it is strongly recommended to ``psycopg`` users to upgrade
+their copy to v1.1.11 or higher
+
+For more details, please refer the corresponding thread in the `psycopg
+archives
+<http://lists.initd.org/pipermail/psycopg/2003-December/002432.html>`__.
+
+Fixing the database schema
+--------------------------
This was sent by the SourceForge.net collaborative development platform, the world's largest Open Source development site.
|
