Further unify logger and Logger

[hauleth]

• Convert usage of :error_logger to :logger
• Migrate :console backend to be defined as a formatter for logger_std_h - this will allow to use the same format as Elixir is using for other handlers as well (requires Allow binaries as parts of logger_formatter template erlang/otp#2491 - OTP 24+ and Allow using custom IO devices in logger_std_h erlang/otp#2523 - OTP 23+)
• Only attach built-in Logger.Handler if there will be backends (depends on the above)
• Expose API for module and application level verbosity control (Expose functions for setting per-module levels #9793)
• Expose and document all of Erlang built-in logger levels (Expose all logger levels #9791)
• Add support for passing map as a "message" to the Logger macros to allow structured logging (needs to be fully discussed before) (Add structured logging #9797)
• Use time field from metadata instead of generating new timestamp (Use Erlang's logger timestamp #9772)
• Make Logger.Formatter works also as a Erlang's logger formatter for use with Erlang's logger handlers
• Make translators a filter
• Document usage and configuration of Erlang's handlers with Elixir's Logger (such as the file handler with rotation)
• Add API to set the global metadata (OTP 24+) (Expose :logger's global metadata #12413)
• Expose :logger.add_handlers/1 (with an atom) (Add Logger.add_handlers/1 #12412)
• Rewrite capture_log to use Erlang handlers (Use Erlang logger handlers in ExUnit.CaptureLog #12408)
• Publish and CI :logger_backends with elixir: ~> 1.15 (https://hex.pm/packages/logger_backends/1.0.0-rc.0)

[josevalim]

Note: don't send PRs for those yet. We will probably wait until the Logger integration stabilizes a bit before going all in.

[hauleth]

Issues to track:

• According to spec format can be io:format/0 erlang/otp#2444 - accept binaries in entries using format
• Allow ignoring log message from function erlang/otp#2442 - allow ignoring messages from within the log function (by returning :ignore) no longer relevant as it was removed from Elixir as well
• Allow binaries as parts of logger_formatter template erlang/otp#2491 - allow using binaries in :logger_formatter template to simplify integration with Elixir

[hauleth]

@josevalim I slowly start working on these, so it would be at least partially ready for 1.11.

[josevalim]

Perfect, the earlier the better!

[hauleth]

Another one PR to core erlang/otp#2523 - this one is to fully support :device option in logger_std_h.

[vans163]

Can this get tagged on too, https://bugs.erlang.org/browse/ERL-1159.

Basically to not push up the REPL prompt on Logger outputs.

iex(2)> :erlang.spawn fn()-> Process.sleep(1000); IO.puts "hi" end   
#PID<0.108.0>
hi
iex(3)>

iex(3)> :erlang.spawn fn()-> Process.sleep(1000); Logger.debug "hi" end
#PID<0.112.0>
iex(4)> 
16:07:31.101 [debug] hi

Notice how iex(4)> flushes first then the debugger process prints. Vs in the IO.puts case, its synchronized.

[josevalim]

Hi @vans163, because the issue happens in both Erlang and Elixir and are not specific to the integration, it should be rather directly tracked in Erlang issues tracker. Thanks for pointing it out though!

[hauleth]

Traditionally in most languages the stdout is buffered and stderr is not, this follow libc "standard". Additionally Elixir's logger use :user as a default device instead of :standard_io and that can be an issue here. AFAIK the problem is unsolvable in general case as there is no "registry" of all defined io:device/0s (any registered process can be a device). And as :user writes directly to 0 and is not synchronised in any way to :standard_io (I am no IEx expert, but I assume that it uses :standard_io not :user as a IO device).