Support Erlang 21 new features

[josevalim]

To be implemented when Erlang 21 is released (part of v1.7):

• Support new handle_continue/2 gen_server callback
• Support the __STACKTRACE__ variable in catch and rescue and Erlang's new try/catch clauses
• Deprecate System.stacktrace outside of catch and rescue
• Integrate with the new logger API (done on v1.6.6)

To be implemented when we require Erlang 21 exclusively:

• Unify :logger and Logger metadata (done in Use Erlang's logger as main logging implementation #9333)
• Remove assert_no_null_byte from file operations and get_env/put_env (find_executable and cmd still need it)
• Optimize elixir_erl_try.erl by removing old stacktrace logic and unifying all clauses into a single one with map_get
• Add is_struct/1 and is_struct/2
• Add is_map_key/2

[OvermindDL1]

Use the new compile_info compiler option to attach Elixir compiler metadata

No way of doing that when building on OTP21 as it stands before it is required?

[josevalim]

We need to check the version because otherwise the data ends up in compile_info but in the wrong place (under options).

[michalmuskala]

I wonder about support for map.field in guards - this requires introducing the map_get guard to Elixir. In the past we haven't introduced new guard functions until all supported versions had them (see ceil/1 and floor/1 we still don't support because we support OTP 19).

[gerhard]

Could this error be related? rabbitmq/rabbitmq-server-boshrelease@0d23f39

We're hitting it on OTP 21.0-rc1 with RabbitMQ 3.7.4 which is using Elixir 1.6.0

cc @hairyhum @michaelklishin @dumbbell

[michalmuskala]

Yes. Elixir does not support OTP 21 yet, some failures are expected.

[michalmuskala]

An additional thing is map iterators - many functions in the erlang's maps module accept either a map or a map iterator. Should we extend our Map module in the same way?

[josevalim]

@michalmuskala that's a good point. I think we can wait for some usage patterns to emerge before actually adding it to the language.

[josevalim]

Relevant discussion here: https://elixirforum.com/t/discussion-incorporating-erlang-otp-21-map-guards-in-elixir/14816

[Jancd]

When I try to run Elixir from source,an error occured after running make test:

{"init terminating in do_boot",{'cannot get bootfile','no_dot_erlang.boot'}}
init terminating in do_boot ({cannot get bootfile,no_dot_erlang.boot})

Crash dump is being written to: erl_crash.dump...done
make: *** [lib/elixir/src/elixir_parser.erl] Error 1

Erlang/OTP version:Erlang/OTP 21 [erts-10.0] [source] [64-bit] [smp:4:4] [ds:4:4:10] [async-threads:1] [hipe],and OS :CentOS7.

[josevalim]

@7ethan how have you installed Erlang? Also see this issue: #7794 (comment)

[Jancd]

oh,got it.

[ekremkaraca]

Hey guys,

Since Erlang 22.0 was released, how about closing this issue and moving tasks to #8414?

[josevalim]

@ekremkaraca we still support Erlang/OTP 20, so while we support it, we can't work on the pending tasks here. We will drop Erlang/OTP 20 once Elixir v1.10 is out (January 2020), then this issue can be addressed.

[wojtekmach]

With map_get and is_map_key there could be a MapSet.is_member guard:

  defguard is_member(map_set, value)
           when :erlang.is_map_key(value, :erlang.map_get(:map, map_set))

Curious if there's any interest in that and/or a better implementation.

[ericmj]

Should it also check :erlang.map_get(:__struct__, map_set) == MapSet?

There's a possibility dialyzer will complain about this guard because we are accessing internal fields of an opaque type.

[wojtekmach]

Good point about checking __struct__. I checked with dialyxir and didn't see a warning.

[legoscia]

Would the Enumerable implementation for Map use iterators to implement reduce instead of :maps.to_list once depending on Erlang 21?

[michalmuskala]

I did a test some time ago and unfortunately using iterators made a big performance difference - about 15%. The primary reason is that iterators are prepared for the erlang-style /2 functions, while we have /1 taking a 2-tuple, so using iterators means unpacking and repacking tuples all the time.

[josevalim]

This is ready to be tackled.

[eksperimental]

"Add is_struct/1" can be ticked off, and "and is_struct/2" crossed out.

[eksperimental]

I would like to work on is_map_key/2. Is anybody already working on it?

[josevalim]

Sorry, I am almost done with is_map_key but I am calling it a night. :) I will push it tomorrow morning. -- *José Valimwww.plataformatec.com.br <http://www.plataformatec.com.br/>Founder and Director of R&D*

[hauleth]

Closed 🎉