Issues with sessions about lapis

[leafo]

I'm aware of the session removal issue. I've been using the false assignment workaround in my sites, but I think it's time to make it a little bit more proper and detect nil assignments.

As for __index being a function, I've never seen that happen before, but I see no issue with adding a check there.

[catwell]

Thank you. Regarding __index being a function I will try to make a simple test case that exhibits the problem this weekend.

[catwell]

Actually the 2nd workaround does not work, it did only because I had commented out code (using false instead of nil does). And I couldn't reproduce __index being a function so disregard that for now.

[iandyh]

@leafo I am currently experiencing the __index being a function issue. Completely have no idea why this would happen. I have dig around but found no solutions.

So I use cookies instead now.

[catwell]

I have had a look and here's where I think the issue with __index being a function comes from. Well, that's if I have understood all this meta-magic correctly :)

OK, so the session is an auto_table. An auto_table is a table with a __index metamethod by default (so "__index is a function").

The reason this bug does not happen all the time is that actually the __index metamethod is a closure, and when it is called it replaces itself with the result of the enclosed fn (a table). So if you access a field in the session before trying to iterate __index it works, but not if you don't.

[leafo]

Ah good catch, thanks for looking into it.

[kotedo]

Question: Has the session issue been fixed? I am using the latest github checked out version and I am still having the aforementioned session issues.

Thank you!

[leafo]

Still haven't changed the delete issue. The __index being a function has been fixed.

[kotedo]

Right, that's what I noticed. It renders sessions almost unusable ... and I was hoping to use Lapis for an important site that requires sessions.

[leafo]

I wouldn't say they're unusable. Set to false if you want to delete. I'll try to get patch out asap.

[leafo]

So I've taken care of the assigning nil issue, but I'm not satisfied with the overall result. You will still run into trouble if you have a table serialized into the session and you modify it, the code won't be able to detect the change in the child object. I'm wondering if I should just change the interface completely, have some method that marks the session as dirty and forces it to be written, or write something that wraps the objects that come from the session so I can tell if any nested object is changed.

[catwell]

That last solution looks right. What can be stored in sessions exactly? Userdata, functions? If neither can then just adding metatables to tables stored in sessions so that they propagate the "modified" state upwards sounds good.

[leafo]

Anything that can be serialized by lua cjson. I should mention that somewhere.