GitHub

Announcement: https://elixir-lang.org/blog/2026/06/03/elixir-v1-20-0-released/

This release requires Erlang/OTP 27+ and is compatible with Erlang/OTP 29.

Type system improvements

Elixir's type system now understands all language constructs and can infer types for your function definitions, using typing information from Elixir's standard library and your dependencies, to find verified bugs and dead code.

This has been achieved through a series of improvements, such as type refinement across clauses, occurrence typing, typing of map keys and domains, and more.

Type inference of guards

This release also performs inference of guards! Let's see some examples:

def example(x, y) when is_list(x) and is_integer(y)

The code above correctly infers x is a list and y is an integer.

def example({:ok, x} = y) when is_binary(x) or is_integer(x)

The one above infers x is a binary or an integer, and y is a two element tuple with :ok as first element and a binary or integer as second.

def example(x) when is_map_key(x, :foo)

The code above infers x is a map which has the :foo key, represented as %{..., foo: dynamic()}. Remember the leading ... indicates the map may have other keys.

def example(x) when not is_map_key(x, :foo)

And the code above infers x does not have the :foo key (hence x.foo will raise a typing violation), which has the type: %{..., foo: not_set()}.

You can also have expressions that assert on the size of data structures:

def example(x) when tuple_size(x) < 3

Elixir will correctly track the tuple has at most two elements, and therefore accessing elem(x, 3) will emit a typing violation. In other words, Elixir can look at complex guards, infer types, and use this information to find bugs in our code, without a need to introduce type signatures (yet).

Whole-body type inference

Elixir also performs inference based on the function body itself. Take the following code:

def add_foo_and_bar(data) do
  data.foo + data.bar
end

Elixir now infers that the function expects a map as first argument, and the map must have the keys .foo and .bar whose values are either integer() or float(). The return type will be either integer() or float().

Here is another example:

def sum_to_string(a, b) do
  Integer.to_string(a + b)
end

Even though the + operator works with both integers and floats, Elixir infers that a and b must be both integers, as the result of + is given to a function that expects an integer. The inferred type information is then used during type checking to find possible typing errors. The typing inferred from your dependencies are also used to help infer more precise types for your own applications.

Typing across clauses

Elixir now infers the type of a given clause based on previous clauses. Let's see an example:

case System.get_env("SOME_VAR") do
  nil -> :not_found
  value -> {:ok, String.upcase(value)}
end

System.get_env("SOME_VAR") returns either nil or a binary(). Because the first clause matches on nil, the type system knows value can no longer be nil, and therefore it must only be a binary(), which allows the second clause to also type check without violations.

This type inference across clauses also helps the type system find redundant clauses and dead code in existing codebases. Elixir v1.20 also implements occurrence typing for cond, case, and with, providing more precise types within each clause.

Typing of atom and domain keys in maps

Maps were one of the first data-structures we implemented within the Elixir type system however, up to this point, they only supported atom keys. If they had additional keys, those keys were simply marked as dynamic().

As of Elixir v1.20, we can track all possible domains as map keys. For example, the map:

%{123 => "hello", 456.0 => :ok}

will have the type:

%{integer() => binary(), float() => :ok}

It is also possible to mix domain keys, as above, with atom keys, yielding the following:

%{integer() => integer(), root: integer()}

This system is an implementation of Typing Records, Maps, and Structs, by Giuseppe Castagna (2023).

Typing of map operations

We have typed the majority of the functions in the Map module, allowing the type system to track how keys are added, updated, and removed across all possible key types.

For example, imagine we are calling the following Map functions with a variable map, which we don't know the exact shape of, and an atom key:

Map.put(map, :key, 123)
#=> returns type %{..., key: integer()}
Map.delete(map, :key)
#=> returns type %{..., key: not_set()}

As you can see, we track when keys are set and also when they are removed.

Some operations, like Map.replace/3, only replace the key if it exists, and that is also propagated by the type system:

Map.replace(map, :key, 123)
#=> returns type %{..., key: if_set(integer())}

In other words, if the key exists, it would have been replaced by an integer value. Furthermore, whenever calling a function in the Map module and the given key is statically proven to never exist in the map, an error is emitted.

By combining full type inference with bang operations like Map.fetch!/2, Map.pop!/2, Map.replace!/3, and Map.update!/3, Elixir is able to propagate information about the desired keys. Take this module:

defmodule User do
  def name(map), do: Map.fetch!(map, :name)
end
defmodule CallsUser do
  def calls_name do
    User.name(%{})
  end
end

The code above has a type violation, which is now caught by the type system:

    warning: incompatible types given to User.name/1:
        User.name(%{})
    given types:
        %{name: not_set()}
    but expected one of:
        dynamic(%{..., name: term()})
    type warning found at:
    │
 16 │     User.name(%{})
    │         ~
    │
    └─ lib/calls_user.ex:7:5: CallsUser.calls_name/0

Acknowledgements

The type system was made possible thanks to a partnership between CNRS and Remote. The development work is currently sponsored by Fresha and Tidewave.

Compile-time improvements

Elixir's v1.20 improves compilation times once more, especially on applications with many cores.

It also introduces a new compiler option called :module_definition, which if the module definition should be :compiled (the default) or :interpreted. Note this does not affect the .beam file written to disk, only how the contents inside defmodule are executed. Using the :interpreted mode may offer better compilation times for large projects, especially on machines with high core count, however, it comes with some downsides:

  • Errors during compilation may have less precise stacktraces

  • Anonymous functions within defmodule can have only up to 20 arguments.
    If this is an issue, you can use maps or tuples to group the data.
    Note the functions themselves inside defmodule, such as the ones defined
    inside def and friends, can still have up to 255 arguments

You can enable it by setting elixirc_options: [module_definition: :interpreted] in your mix.exs.

v1.20.0 (2026-06-03)

This release requires Erlang/OTP 27+ and is compatible with Erlang/OTP 29.

1. Enhancements

EEx

  • [EEx] Optimize compiler by flattening expr list only once

Elixir

  • [Base] Optimize Base validation functions by using SWAR techniques
  • [Calendar] Optimize date_from_iso_days by using the Neri-Schneider algorithm
  • [Code] Add :dbg_callback option to eval functions
  • [Code] Add module_definition: :interpreted option to Code which allows module definitions to be evaluated instead of compiled. In some applications/architectures, this can lead to drastic improvements to compilation times. Note this does not affect the generated .beam file, which will have the same performance/behaviour as before
  • [Code] Make module purging opt-in and move temporary module deletion to the background to speed up compilation times
  • [Code.Fragment] Allow preserving sigil metadata in container_cursor_to_quoted
  • [Enum] Add Enum.min_max sorter
  • [File] Add support for [:raw] opts in File.read/2
  • [File] Skip device, named pipes, etc in File.cp_r/3 instead of erroring with reason :eio
  • [Float] Optimize Float.round/2 by avoiding big integers
  • [Inspect] Increase inspect limit to help print deeply nested data structures
  • [Inspect] Support printing Erlang records (using Erlang notation)
  • [Integer] Add Integer.ceil_div/2
  • [Integer] Add Integer.popcount/1
  • [IO] Add IO.iodata_empty?/1
  • [Kernel] Add type inference across clauses. For example, if one clause says x when is_integer(x), then the next clause may no longer be an integer
  • [Kernel] Add occurrence typing on case, cond, and with
  • [Kernel] Detect and warn on redundant clauses
  • [Kernel] Perform type inference across applications
  • [Kernel] Print intermediate results of dbg for pipes
  • [Kernel] Show undefined function errors even when missing variables (this helps debug errors caused when the developer forgets to require a macro)
  • [Kernel] Warn on unused requires
  • [List] Add List.first!/1 and List.last!/1
  • [Module] Purge and delete modules if after_compile/2 callback fails
  • [PartitionSupervisor] Support via tuples in count_children/1 and stop/3
  • [Process] Add Process.get_label/1
  • [Registry] Switc...

Read the original on github.com ↗