|
From: Tom C. <su...@ic...> - 2009-12-07 17:35:37
Attachments:
inheritance.patch
|
Hi all,
I am doing some tests with InheritableSQLobject and have run into a problem
which I hope to get some feedback on.
The problem is that when creating a new object based on an
InheritableSQLObject class, and the creation of the child object fails after
a new parent was created, the parent record will remain in the database
without the relevant child.
Here is a short example:
~~~~~~~~~~ file: inheritTest3.py ~~~~~~~~~~~~
from sqlobject import *
from sqlobject.inheritance import InheritableSQLObject
__connection__ = "sqlite:/:memory:"
class Vehicle(InheritableSQLObject):
vid = StringCol(notNone=True, default=None, unique=True)
wheels = IntCol()
class Bicycle(Vehicle):
seats = IntCol(notNone=True, default=None)
class Car(Vehicle):
doors = IntCol(notNone=True, default=None)
def createTables():
Vehicle.createTable()
Bicycle.createTable()
Car.createTable()
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Here is a sample session showing the problem when trying to create a bicycle
but only supplying the args required for the Vehicle part of the object:
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In [1]: from inheritTest3 import *
In [2]: createTables(True)
In [3]: b = Bicycle(vid="id1", wheels=4)
ERROR: An unexpected error occurred while tokenizing input
The following traceback may be corrupted or invalid
The error message is: ('EOF in multi-line statement', (22, 0))
---------------------------------------------------------------------------
IntegrityError Traceback (most recent call last)
/home/tomc/mobilereporter.repo/main/app/<ipython console> in <module>()
/usr/lib/python2.5/site-packages/sqlobject/main.pyc in __init__(self, **kw)
1221 id = None
1222
-> 1223 self._create(id, **kw)
1224
.... snipping lots of traceback parts ............
/usr/lib/python2.5/site-packages/sqlobject/sqlite/sqliteconnection.pyc in
_executeRetry(self, conn, cursor, query)
187 raise DuplicateEntryError(msg)
188 else:
--> 189 raise IntegrityError(msg)
190 except self.module.InternalError, e:
191 raise InternalError(ErrorMessage(e))
IntegrityError: bicycle.seats may not be NULL
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
If you now supply all required args for the bicycle, using the same field
values for the vid and wheels as before, a duplicate entry error is raised:
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In [4]: b = Bicycle(vid="id1", wheels=4, seats=2)
ERROR: An unexpected error occurred while tokenizing input
The following traceback may be corrupted or invalid
The error message is: ('EOF in multi-line statement', (22, 0))
---------------------------------------------------------------------------
DuplicateEntryError Traceback (most recent call last)
/home/tomc/mobilereporter.repo/main/app/<ipython console> in <module>()
/usr/lib/python2.5/site-packages/sqlobject/main.pyc in __init__(self, **kw)
1221 id = None
1222
-> 1223 self._create(id, **kw)
1224
1225 for func in post_funcs:
.... snipping lots of traceback parts ............
/usr/lib/python2.5/site-packages/sqlobject/sqlite/sqliteconnection.pyc in
_executeRetry(self, conn, cursor, query)
185 msg = ErrorMessage(e)
186 if msg.startswith('column') and msg.endswith('not
unique'):
--> 187 raise DuplicateEntryError(msg)
188 else:
189 raise IntegrityError(msg)
DuplicateEntryError: column vid is not unique
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The duplicate entry is because of the parent vehicle record was created when
the first bicycle creation failed:
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In [5]: q = "SELECT * from vehicle"
In [6]: Vehicle._connection.queryAll(q)
Out[6]: [(1, 'id1', 4, 'Bicycle')]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
So, the problem is that if a new child object is created, but fails after the
parent has been created, the parent is left in tact in the database.
I'm attaching a patch as a possible fix for this, but I am not sure under
which other circumstances this patch will cause other problems. Maybe a flag
is needed inside the block that creates the child (parentCreatedHere=True),
and only if this flag is set, should the parent be deleted in exception if
the child creation failed?
I would appreciate if someone with more insight into this can look at the
patch an modify where neccessary.
BTW: This patch is against SQLObject 0.11.0.
Cheers,
Tom
|
|
From: Oleg B. <ph...@ph...> - 2009-12-07 18:28:34
|
Hi!
On Mon, Dec 07, 2009 at 06:41:54PM +0200, Tom Coetser wrote:
> The problem is that when creating a new object based on an
> InheritableSQLObject class, and the creation of the child object fails after
> a new parent was created, the parent record will remain in the database
> without the relevant child.
A valid concern, sure. I never stumbled upon the problem because I
always use transactions, and on any exception I just ABORT the transaction;
on ABORT the DB removes the half created parent.
> I'm attaching a patch as a possible fix for this
Thank you!
> - super(InheritableSQLObject, self)._create(id, **kw)
> + # TC: Create this record and catch all exceptions in order to destroy
> + # TC: the parent if the child can not be created.
> + try:
> + super(InheritableSQLObject, self)._create(id, **kw)
> + except:
> + # TC: If we are the child, destroy the parent
> + if self.sqlmeta.parentClass:
> + self._parent.destroySelf()
> + self._parent = None
> + # TC: Reraise the original exception
> + raise
> +
The patch looks rather simple and straightforward. I'll test it.
Oleg.
--
Oleg Broytman http://phd.pp.ru/ ph...@ph...
Programmers don't die, they just GOSUB without RETURN.
|
|
From: Oleg B. <ph...@ph...> - 2010-02-08 18:17:59
|
Hi! I applied the patch with an additional test - that the connection is not
a transaction and is in autocommit mode. A transaction or non-autocommit
connection must be rolled back after a exception but the place for the
rollback is application-dependent so I didn't include a rollback in the
code. Inside a transaction it's also usually impossible to remove the
parent row(s) after an error - the only thing that's allowed is a rollback.
Committed in the revisions 4101-4103 (branches 0.11, 0.12 ad the trunk).
Thank you!
On Mon, Dec 07, 2009 at 06:41:54PM +0200, Tom Coetser wrote:
> I am doing some tests with InheritableSQLobject and have run into a problem
> which I hope to get some feedback on.
>
> The problem is that when creating a new object based on an
> InheritableSQLObject class, and the creation of the child object fails after
> a new parent was created, the parent record will remain in the database
> without the relevant child.
>
> Here is a short example:
>
> ~~~~~~~~~~ file: inheritTest3.py ~~~~~~~~~~~~
> from sqlobject import *
> from sqlobject.inheritance import InheritableSQLObject
>
> __connection__ = "sqlite:/:memory:"
>
> class Vehicle(InheritableSQLObject):
> vid = StringCol(notNone=True, default=None, unique=True)
> wheels = IntCol()
>
> class Bicycle(Vehicle):
> seats = IntCol(notNone=True, default=None)
>
> class Car(Vehicle):
> doors = IntCol(notNone=True, default=None)
>
> def createTables():
> Vehicle.createTable()
> Bicycle.createTable()
> Car.createTable()
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>
> Here is a sample session showing the problem when trying to create a bicycle
> but only supplying the args required for the Vehicle part of the object:
>
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> In [1]: from inheritTest3 import *
>
> In [2]: createTables(True)
>
> In [3]: b = Bicycle(vid="id1", wheels=4)
> ERROR: An unexpected error occurred while tokenizing input
> The following traceback may be corrupted or invalid
> The error message is: ('EOF in multi-line statement', (22, 0))
>
> ---------------------------------------------------------------------------
> IntegrityError Traceback (most recent call last)
>
> /home/tomc/mobilereporter.repo/main/app/<ipython console> in <module>()
>
> /usr/lib/python2.5/site-packages/sqlobject/main.pyc in __init__(self, **kw)
> 1221 id = None
> 1222
> -> 1223 self._create(id, **kw)
> 1224
>
> .... snipping lots of traceback parts ............
>
> /usr/lib/python2.5/site-packages/sqlobject/sqlite/sqliteconnection.pyc in
> _executeRetry(self, conn, cursor, query)
> 187 raise DuplicateEntryError(msg)
> 188 else:
> --> 189 raise IntegrityError(msg)
> 190 except self.module.InternalError, e:
> 191 raise InternalError(ErrorMessage(e))
>
> IntegrityError: bicycle.seats may not be NULL
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>
> If you now supply all required args for the bicycle, using the same field
> values for the vid and wheels as before, a duplicate entry error is raised:
>
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>
> In [4]: b = Bicycle(vid="id1", wheels=4, seats=2)
> ERROR: An unexpected error occurred while tokenizing input
> The following traceback may be corrupted or invalid
> The error message is: ('EOF in multi-line statement', (22, 0))
>
> ---------------------------------------------------------------------------
> DuplicateEntryError Traceback (most recent call last)
>
> /home/tomc/mobilereporter.repo/main/app/<ipython console> in <module>()
>
> /usr/lib/python2.5/site-packages/sqlobject/main.pyc in __init__(self, **kw)
> 1221 id = None
> 1222
> -> 1223 self._create(id, **kw)
> 1224
> 1225 for func in post_funcs:
>
> .... snipping lots of traceback parts ............
>
> /usr/lib/python2.5/site-packages/sqlobject/sqlite/sqliteconnection.pyc in
> _executeRetry(self, conn, cursor, query)
> 185 msg = ErrorMessage(e)
> 186 if msg.startswith('column') and msg.endswith('not
> unique'):
> --> 187 raise DuplicateEntryError(msg)
> 188 else:
> 189 raise IntegrityError(msg)
>
> DuplicateEntryError: column vid is not unique
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>
> The duplicate entry is because of the parent vehicle record was created when
> the first bicycle creation failed:
>
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> In [5]: q = "SELECT * from vehicle"
>
> In [6]: Vehicle._connection.queryAll(q)
> Out[6]: [(1, 'id1', 4, 'Bicycle')]
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>
> So, the problem is that if a new child object is created, but fails after the
> parent has been created, the parent is left in tact in the database.
>
> I'm attaching a patch as a possible fix for this, but I am not sure under
> which other circumstances this patch will cause other problems. Maybe a flag
> is needed inside the block that creates the child (parentCreatedHere=True),
> and only if this flag is set, should the parent be deleted in exception if
> the child creation failed?
>
> I would appreciate if someone with more insight into this can look at the
> patch an modify where neccessary.
>
> BTW: This patch is against SQLObject 0.11.0.
>
>
>
> Cheers,
> Tom
> --- inheritance/__init__.py.orig 2009-12-07 14:47:31.000000000 +0000
> +++ inheritance/__init__.py 2009-12-07 16:35:39.000000000 +0000
> @@ -384,7 +384,19 @@
>
> id = self._parent.id
>
> - super(InheritableSQLObject, self)._create(id, **kw)
> + # TC: Create this record and catch all exceptions in order to destroy
> + # TC: the parent if the child can not be created.
> + try:
> + super(InheritableSQLObject, self)._create(id, **kw)
> + except:
> + # TC: If we are the child, destroy the parent
> + if self.sqlmeta.parentClass:
> + self._parent.destroySelf()
> + #TC: Do we need to do this??
> + self._parent = None
> + # TC: Reraise the original exception
> + raise
> +
>
> def _findAlternateID(cls, name, dbName, value, connection=None):
> result = list(cls.selectBy(connection, **{name: value}))
Oleg.
--
Oleg Broytman http://phd.pp.ru/ ph...@ph...
Programmers don't die, they just GOSUB without RETURN.
|
