Flashes are not getting reflashed on 409

[totaltrash]

https://inertiajs.com/the-protocol#asset-versioning

Finally, in the event that flash session data exists when a 409 Conflict response occurs, the server will automatically reflash this data.

[tmartin8080]

Thanks @totaltrash .

[bigx333]

fetch_flash does basically the same for redirects https://github.com/phoenixframework/phoenix/blob/master/lib/phoenix/controller.ex#L1296

Can use the same logic

[totaltrash]

ugh, sorry, it's still early here. ignore my previous comment!

[totaltrash]

I've had a nightmare with this. I can't write a failing test (I think it needs an integration test to replicate it), and I can't write an implementation... my elixir chops are not quite there.

Looking at how Phoenix handles flashes, as per your link @bigx333 , the flashes get deleted from the session before sending the response (unless a redirect). Se we'll need to store the flashes under a different session key, and merge them in upon the next request after a 409? Or something else?

[bigx333]

I've had a nightmare with this. I can't write a failing test (I think it needs an integration test to replicate it), and I can't write an implementation... my elixir chops are not quite there.

Looking at how Phoenix handles flashes, as per your link @bigx333 , the flashes get deleted from the session before sending the response (unless a redirect). Se we'll need to store the flashes under a different session key, and merge them in upon the next request after a 409? Or something else?

I didn't really have time to look into this but fetch_flash removes the session after it saves the flash on conn.private.phenix_flash, you should be able to just use put_session to persist the flash just like fetch_flash does and fetch_flash will automatically repopulate conn.private on the next request.

Something like this should work (untested)

|> check_flash()
|> send_resp(:conflict, "")

def check_flash(conn) do
  flash = conn.private.phoenix_flash
  flash_size = map_size(flash)

  if flash_size > 0 do
    put_session(conn, "phoenix_flash", flash)
  else
    conn
  end
end

Regarding the test, I'm not quite sure, normally on a phoenix test using get("/") would recycle the connection and set the session based on the cookies but here we don't have a router so this needs to be done manually.

@tmartin314 Any input on writing a test for this?

[tmartin8080]

Might be worth while to refactor the tests to use a dummy router and controller so we can test requests regularly. The controller test we have started out pretty simple and needs to cover many different scenarios now.

I'll start re-working the ConnCase.

[tmartin8080]

@totaltrash @bigx333 I was able to refactor the tests in 0.2.4. I'll take a stab at this issue as well. Thanks again for all your help.

[totaltrash]

G'day, I've pulled the latest changes including #24 into my test app. The flashes aren't getting reflashed upon a 409.

fetch_flash removes the session after it saves the flash on conn.private.phenix_flash, you should be able to just use put_session to persist the flash just like fetch_flash does and fetch_flash will automatically repopulate conn.private on the next request

I think the issue is that it clears the session before sending the response using register_before_send.

So the timing is:

1. fetch_flash plug is called by the router, loading the flashes out of the session, and a callback is registered with register_before_send
2. InertiaPhoenix.plug is called, an invalid inertia version is identified, and any existing flashes are written back to the session.
3. The register_before_send is called, which deletes the session (as the status is not a redirect) including the reflashed flashes

[tmartin8080]

@totaltrash It looks like @bigx333 tackled this issue. It's merged into v0.2.6

[totaltrash]

Yep, nailed it, nice @bigx333