Query results cache is not cleared after commit about pony

[kozlovsky]

Thanks, this is bug, will be fixed today.

[kozlovsky]

Here is why that happened:

Each Pony ORM session has a cache. This cache contains:

• objects loaded from the database;
• query results.

The query results cache was not cleared after commit(), that was a bug. Since you call the same query twice, the second time query result was taken from a cache.

Your last query looks the same, but its generator expression has separate Python code object, so from Pony point of view it is a different query. Because of this, the query result was taken not from the cache, but from database.

Now, after the bug is fixed, query results cache is cleared correctly after commit(), rollback() or flush().

---

Note, however, that objects cache are still not cleared after commit(), only query results are cleared:

u1 = User[123]  # The object is cached
...
commit()        # Query results cache is cleared, but u1 are still cached
print u1.email  # You can still access cached user object
rollback()      # The cache is invalidated now
print u1.email  # TransactionRolledBack: Object belongs to obsolete cache

To clear object cache you should wrap the function which works with the database with @with_transaction decorator:

@with_transaction
def create_users():
    for i in range(2):
        User(email="email%d" % i, password="password%d" % i)
        # commit() is unnecessary here

create_users()
# commit() and cache clearing will be done automatically

[iamsk]

so quick, cool, thx
sth about the cache, i got it, but there is another question
why the parent iterator have effect on the select(generator)

for i in range(2):
select()
will select form db once;

select()
select()
will select form db twice;

[kozlovsky]

When Pony caches query result it uses id of generator's codeobject as part of the cache key.

In the first case, you call select() with the same generator twice, and on the second iteration Pony uses query result from the cache.

In the second case, you call two different generators, and Python creates two different codeobjects with different ids. Currently Pony doesn't recognizes such generators represent the same query and send two separate queries to the database. In future we may change the way query key is generated and after that Pony will take result of the second select() from the cache.

The next example demonstrates that different generator codeobjects have different ids:

>>> names = [ 'John', 'Smith', 'Bob' ]
>>> def f1():
...     gen1 = (name for name in names)
...     print id(gen1.gi_frame.f_code)
...     
>>> def f2():
...     gen2 = (name for name in names)
...     print id(gen2.gi_frame.f_code)
... 
>>> f1()
54873504
>>> f1()
54873504
>>> f2()
54873144
>>> f2()
54873144
>>> 

[iamsk]

ok, thanks for you answer