connect_params are lost when navigating

[randycoulman]

I'm working on a feature that relies on a value being passed from the browser (app.js) to my LiveView app via the connect_params.

When I test in the browser, all works as expected.

In my tests, I inject a value using Phoenix.LiveViewTest.put_connect_params/2.

When I first visit a live view, the connect_param is present when I call LiveView.get_connect_params/1](https://hexdocs.pm/phoenix_live_view/Phoenix.LiveView.html#get_connect_params/1).

However, if I then perform an action that does a live navigation to another LiveView, the connect_params are no longer present.

In the browser, the connect_params persist across all of the live navigation operations as expected.

I believe that the root cause is this call to ConnHandler.recycle_all_headers/1. The recycle it does seems to lose the connect_params.

I haven't tested this with "raw" LiveViewTest, but I see that its follow_redirect implementation (which we also call here) calls ensure_recycled, so it may suffer from the same problem.

Is this something that it makes sense to fix here? Or should I report it upstream to Phoenix instead/as well? Even if it gets fixed there, we'll still have a problem here and vice versa, I think.

I'm not sure of the implications, but I was wondering if the conn should be recycled at all in the case of a :live_redirect, but only in the case of a :redirect. Thoughts?

[randycoulman]

I did some more experimentation and I think the problem runs deeper than this. I temporarily commented out both of the recycle calls (the one here in PhoenixTest and the one in LiveViewTest). I verified that conn.private[:live_view_connect_params] are preserved in that case (though the change probably breaks other things).

However, even with that, get_connect_params/1 does not return my injected test params. It turns out that there is more recycling going on even deeper in the stack.

The fundamental problem is that recycle does not preserve anything in conn.private (which is where put_connect_params puts things).

I've opened an issue against LiveView, but I'll keep this one open as well. Depending on how the upstream issue is resolved, it might involve changes here.

[germsvel]

Thanks so much for opening the issue here and in LiveView Randy! I'll keep an eye on the LiveView issue to see what they say.