Strange behaviour when trying to log in or log out

[sagikazarmark]

When I try to login/logout and I redirect the request right after doing any of these actions, nothing happens (meaning login/logout fails). If I stop the request (aka. exit) and refresh the page then it succeeds. Is it a session or an auth issue?

Example:

    public function actionLogout()
    {
        $this->getAuth()->logout();

        return \Response::forge('redirect', 'admin/login');
    }

[sagikazarmark]

The problem is that the store method on Persistence object is not called when using Redirect.

[sagikazarmark]

Well, I cannot fix it by calling directly the destructor, so might be something else.

@WanWizard any idea how to manually persist a login/logout?

[WanWizard]

Auth wasn't finished up to that point yet. I only tested the individual methods from the commandline.

In v1 sessions are written in a shutdown event, which always runs, even when you redirect. The event instance is created by the Application, but Auth is not integrated yet.

On th other hand, a class destructor should always be called. Is it called? And if so, what does it write?

[sagikazarmark]

I know that you didn't finish it, but I hope reporting bugs will help your work.

Not sure what happens here. I am using the File drivers (for storage and persistence as well). Does it use any session at all?

It seems that the File Persistence driver's destructor is NOT called when using Redirect.

[WanWizard]

No, there is no session support at all at the moment. There is only one persistence driver, with is the file driver. Which is very basic, as it stores state based on REMOTE_ADDR, or localhost if no address exists. I wrote the File drivers to be able to test the API without dependencies (to database, session, or other packages).

[sagikazarmark]

That's what I thought. I still can't decide if it is an issue with redirect or the persistence driver.

[WanWizard]

I find it very strange that in case of a redirect, the class destructor isn't called. That should be part of the PHP cleanup...

[sagikazarmark]

I'll try to debug what happens.

[sagikazarmark]

Something stupid is going on here. The problem only happens when I redirect to the same controller. If I do a redirection to any other controller and go back to the "protected" one, than it works.

[sagikazarmark]

GOT IT. Opcode cache. I also have to note that I currently use php built-in webserver for testing. Turning opcode cache off solved the problem.

It should be tested with apache as well, but I think the result will be the same.

I don't know how it could/should be solved. Probably saving data into non-php files?

[WanWizard]

It's the main reason I always have all cache switched off in development and staging environments (the other important one is being able to see the apps true performance).

Even without Opcache you can bump into issues. I spent most of last week trying to debug an issue with a task being fired by a queue worker. It wouldn't run an updated version of the task, because it still had the old class in memory.

[sagikazarmark]

Well, I always have problems with this. If I leave opcache turned on, I can solve these issues during the development, however I usually forget to blame opcache.

Anyway, I agree with your points.

Any idea for solving the issue? While I don't think File driver will be used in production environment, it should still be possible.

[WanWizard]

I don't have a clue how to solve it, if calling a class destructor is dependent on having opcache active or not, I'm inclined to say that it's a PHP bug...

[sagikazarmark]

How about using another format instead of php? Or letting the file driver choose you a file format?

[WanWizard]

Ah, you mean for the file store itself?

Sorry, just now understand what the issue is, and this can be solved quite easily, we did that for v1 too, see fuel/core@8e3f9bf and fuel/core@389f822