SessionWrapper.pop does not set modified flag

[Ben Slavin <benjamin.slavin@…>]

In #4338 a pop method was added to the SessionWrapper class.

This method alters the state of the underlying Session object but does not set the modified flag. This prevents the change from being persisted to the database unless SESSION_SAVE_EVERY_REQUEST is set or another modification is made to the Session.

This bug was introduced in [5306].

[Ben Slavin <benjamin.slavin@…>]

Sets SessionWrapper.modified when SessionWrapper.pop is called

[Chris Beaven]

Well, technically pop() doesn't guarantee that we are modifying the session. .pop('nonexistantkey', None) isn't doing any modification (just like get() isn't setting self.accessed).

Perhaps it would be better if SessionWrapper._session_cache was a subclass of dict which handled .accessed and .modified` itself...

[Ben Slavin <benjamin.slavin@…>]

You're right that .pop() doesn't guarantee modification. This was something that I had considered, but I think it's better to err on the side of safety. I do like your suggestion of subclassing dict, but I don't know if it's necessary.

Re: your comment about .accessed... this actually isn't true. .get() doesn't need to modify .accessed since ._get_session() handles that already.

[Collin Grady <cgrady@…>]

self.modified = key in self._session ? :)

[Ben Slavin <benjamin.slavin@…>]

Sets SessionWrapper.modified when SessionWrapper.pop finds an item

[Ben Slavin <benjamin.slavin@…>]

Collin... bah! ;-)

I've attached a new patch with self.modified = self.modified or key in self._session

[Chris Beaven]

Is it worth fixing .get in the same ticket or should we open a new one?

[Ben Slavin <benjamin.slavin@…>]

If we decide to change .get we actually have to make a number of changes... __getitem__, keys, items, get. We'd also have to change the functionality of _get_session and would require anyone accessing ._session elsewhere to manually set .accessed (I don't know if this is done in practice anywhere).

Because of all the items touched, I think that it might be better to open another ticket, but I'm not convinced that the current behavior of ._get_session is problematic.

[Chris Beaven]

with tests

[Chris Beaven]

Actually, I was getting confused. .get does set .accessed, since it uses the ._session property.

There's not much to decide upon in this ticket, it's an obvious problem. I'm promoting to checkin

[Adrian Holovaty]

(In [5592]) Fixed #4729 -- SessionWrapper.pop now sets modified flag if necessary. Thanks, Ben Slavin, SmileyChris and Collin Grady

[Ben Slavin]

Updated to new session implementation

[Ben Slavin]

[6333] reintroduced this issue.

Patch has been included.

[Gary Wilson]

The tests also need some alteration so that there is a failure seen before the application of this patch.

[Ben Slavin]

Patch for new session implementation, including tests

[Ben Slavin]

Attached a new patch.

Promoting to 'ready for checkin' because the tests existed prior to [6333] but were erroneously deleted in the new session implementation (which is why this problem reappeared).

[Thomas Güttler <hv@…>]

Tests look good. Please commit.

[Malcolm Tredinnick]

(In [6558]) Fixed #4729 -- Restored functionality to the Session class so that popping a
value marks it as modified. This was accidentally lost in the changes in
[6333]. Thanks, hawkeye.