Fix "always evaluate" for Logger anonymous function tuples

[Odaeus]

My colleague discovered that many of our tests were failing on a fresh build of a project due to an exception being raised when the Quantum package attempted to use the Logger:

Reason: shutdown: failed to start child: Integrations.Scheduler
  ** (EXIT) shutdown: failed to start child: Quantum. JobBroadcaster
    ** (EXIT) an exception was raised:
      ** (Protocol.UndefinedError) protocol String.Chars not implemented for type Tuple. This protocol is implemented for the following type(s): Atom, BitString, Date, DateTime, Decimal, Float, Integer, Lis

Got value:

    {"Loading Initial Jobs from Config", [node: :nonode@nohost]}

      (elixir 1.18.4) lib/string/chars.ex:3: String.Chars. impl_for!/1
      (elixir 1.18.4) lib/string/chars.ex:22: String. Chars.to_string/1
      (logger 1.18.4) lib/logger.ex:982: Logger.__do_10g__/4
      (quantum 3.5.3) lib/quantum/job_broadcaster.ex:51: Quantum.JobBroadcaster.init/1

We had recently upgraded to Elixir 1.18.4 and adjusted the logging config in our test environment to:

config :logger, always_evaluate_messages: true

We traced this down to Quantum's usage of giving anonymous functions that return a tuple to Logger calls. This is an officially supported case but the code path for forced evaluation was running the function and sending the entire result tuple as data to the Logger, which is no longer supported.

I've attempted to fix this by adjusting the forced evaluation functions to handle this case. Importantly, without accidentally adding in support for passing a tuple directly.

Please let me know if this does the job!

-- Andrew

[josevalim]

Can you please double check the metadata handling? I think the metadata is rather merged (you can check __do_log__. It is also better if we generate less code, so I would do this:

Suggested change

	{data, metadata} =
	case Logger.__evaluate_log__(unquote(data)) do
	{data, nil} -> {data, unquote(quoted_metadata)}
	{data, metadata} -> {data, Map.new(metadata)}
	end
	{data, metadata} = Logger.__evaluate_log__(unquote(data), unquote(quoted_metadata))

And then __evaluate_log__ will call Enum.into(fun_meta, quoted_metadata). Please update the test so check the metadata is merged by having it actually log and then comparing the result!

[Odaeus]

Yes, it is merged in the normal path. I've updated the handler as you suggested, which is more elegant.
I updated the tests to include a bit of additional metadata and I match the simple string but I'm not sure if there's the possibility of an ordering issue there. I think it's stable but please let me know if individual matches would be better.

[josevalim]

💚 💙 💜 💛 ❤️