View Source Code (Elixir v1.17.0-rc.0)

Utilities for managing code compilation, code evaluation, and code loading.

This module complements Erlang's :code module to add behavior which is specific to Elixir. For functions to manipulate Elixir's AST (rather than evaluating it), see the Macro module.

Working with files

This module contains three functions for compiling and evaluating files. Here is a summary of them and their behavior:

  • require_file/2 - compiles a file and tracks its name. It does not compile the file again if it has been previously required.

  • compile_file/2 - compiles a file without tracking its name. Compiles the file multiple times when invoked multiple times.

  • eval_file/2 - evaluates the file contents without tracking its name. It returns the result of the last expression in the file, instead of the modules defined in it. Evaluated files do not trigger the compilation tracers described in the next section.

In a nutshell, the first must be used when you want to keep track of the files handled by the system, to avoid the same file from being compiled multiple times. This is common in scripts.

compile_file/2 must be used when you are interested in the modules defined in a file, without tracking. eval_file/2 should be used when you are interested in the result of evaluating the file rather than the modules it defines.

The functions above work with Elixir source. If you want to work with modules compiled to bytecode, which have the .beam extension and are typically found below the _build directory of a Mix project, see the functions in Erlang's :code module.

Code loading on the Erlang VM

Erlang has two modes to load code: interactive and embedded.

By default, the Erlang VM runs in interactive mode, where modules are loaded as needed. In embedded mode the opposite happens, as all modules need to be loaded upfront or explicitly.

You can use ensure_loaded/1 (as well as ensure_loaded?/1 and ensure_loaded!/1) to check if a module is loaded before using it and act.

ensure_compiled/1 and ensure_compiled!/1

Elixir also includes ensure_compiled/1 and ensure_compiled!/1 functions that are a superset of ensure_loaded/1.

Since Elixir's compilation happens in parallel, in some situations you may need to use a module that was not yet compiled, therefore it can't even be loaded.

When invoked, ensure_compiled/1 and ensure_compiled!/1 halt the compilation of the caller until the module becomes available. Note that the distinction between ensure_compiled/1 and ensure_compiled!/1 is important: if you are using ensure_compiled!/1, you are indicating to the compiler that you can only continue if said module is available.

If you are using Code.ensure_compiled/1, you are implying you may continue without the module and therefore Elixir may return {:error, :unavailable} for cases where the module is not yet available (but may be available later on).

For those reasons, developers must typically use Code.ensure_compiled!/1. In particular, do not do this:

case Code.ensure_compiled(module) do
  {:module, _} -> module
  {:error, _} -> raise ...
end

Finally, note you only need ensure_compiled!/1 to check for modules being defined within the same project. It does not apply to modules from dependencies as dependencies are always compiled upfront.

In most cases, ensure_loaded/1 is enough. ensure_compiled!/1 must be used in rare cases, usually involving macros that need to invoke a module for callback information. The use of ensure_compiled/1 is even less likely.

Compilation tracers

Elixir supports compilation tracers, which allow modules to observe constructs handled by the Elixir compiler when compiling files. A tracer is a module that implements the trace/2 function. The function receives the event name as first argument and Macro.Env as second and it must return :ok. It is very important for a tracer to do as little work as possible synchronously and dispatch the bulk of the work to a separate process. Slow tracers will slow down compilation.

You can configure your list of tracers via put_compiler_option/2. The following events are available to tracers:

  • :start - (since v1.11.0) invoked whenever the compiler starts to trace a new lexical context. A lexical context is started when compiling a new file or when defining a module within a function. Note evaluated code does not start a new lexical context (because they don't track unused aliases, imports, etc) but defining a module inside evaluated code will.

    Note this event may be emitted in parallel, where multiple files/modules invoke :start and run at the same time. The value of the lexical_tracker of the macro environment, albeit opaque, can be used to uniquely identify the environment.

  • :stop - (since v1.11.0) invoked whenever the compiler stops tracing a new lexical context, such as a new file.

  • {:import, meta, module, opts} - traced whenever module is imported. meta is the import AST metadata and opts are the import options.

  • {:imported_function, meta, module, name, arity} and {:imported_macro, meta, module, name, arity} - traced whenever an imported function or macro is invoked. meta is the call AST metadata, module is the module the import is from, followed by the name and arity of the imported function/macro. A :remote_function/:remote_macro event may still be emitted for the imported module/name/arity.

  • {:alias, meta, alias, as, opts} - traced whenever alias is aliased to as. meta is the alias AST metadata and opts are the alias options.

  • {:alias_expansion, meta, as, alias} traced whenever there is an alias expansion for a previously defined alias, i.e. when the user writes as which is expanded to alias. meta is the alias expansion AST metadata.

  • {:alias_reference, meta, module} - traced whenever there is an alias in the code, i.e. whenever the user writes MyModule.Foo.Bar in the code, regardless if it was expanded or not.

  • {:require, meta, module, opts} - traced whenever module is required. meta is the require AST metadata and opts are the require options. If the meta option contains the :from_macro, then module was called from within a macro and therefore must be treated as a compile-time dependency.

  • {:struct_expansion, meta, module, keys} - traced whenever module's struct is expanded. meta is the struct AST metadata and keys are the keys being used by expansion

  • {:remote_function, meta, module, name, arity} and {:remote_macro, meta, module, name, arity} - traced whenever a remote function or macro is referenced. meta is the call AST metadata, module is the invoked module, followed by the name and arity.

  • {:local_function, meta, name, arity} and {:local_macro, meta, name, arity} - traced whenever a local function or macro is referenced. meta is the call AST metadata, followed by the name and arity.

  • {:compile_env, app, path, return} - traced whenever Application.compile_env/3 or Application.compile_env!/2 are called. app is an atom, path is a list of keys to traverse in the application environment and return is either {:ok, value} or :error.

  • :defmodule - (since v1.16.2) traced as soon as the definition of a module starts. This is invoked early on in the module life cycle, Module.open?/1 still returns false for such traces

  • {:on_module, bytecode, _ignore} - (since v1.13.0) traced whenever a module is defined. This is equivalent to the @after_compile callback and invoked after any @after_compile in the given module. The third element is currently :none but it may provide more metadata in the future. It is best to ignore it at the moment. Note that Module functions expecting not yet compiled modules (such as Module.definitions_in/1) are still available at the time this event is emitted.

The :tracers compiler option can be combined with the :parser_options compiler option to enrich the metadata of the traced events above.

New events may be added at any time in the future, therefore it is advised for the trace/2 function to have a "catch-all" clause.

Below is an example tracer that prints all remote function invocations:

defmodule MyTracer do
  def trace({:remote_function, _meta, module, name, arity}, env) do
    IO.puts "#{env.file}:#{env.line} #{inspect(module)}.#{name}/#{arity}"
    :ok
  end

  def trace(_event, _env) do
    :ok
  end
end

Summary

Types

A list with all variables and their values.

Diagnostics returned by the compiler and code evaluation.

The line. 0 indicates no line.

The position of the diagnostic.

Functions

Appends a path to the Erlang VM code path list.

Appends a list of paths to the Erlang VM code path list.

Returns a list with all available compiler options.

Returns true if the current process can await for module compilation.

Compiles the given file.

Compiles the quoted expression.

Compiles the given string.

Gets all compilation options from the code server.

Stores all given compilation options.

Deletes a path from the Erlang VM code path list.

Deletes a list of paths from the Erlang VM code path list.

Ensures the given modules are loaded.

Same as ensure_all_loaded/1 but raises if any of the modules cannot be loaded.

Similar to ensure_compiled!/1 but indicates you can continue without said module.

Ensures the given module is compiled and loaded.

Ensures the given module is loaded.

Same as ensure_loaded/1 but raises if the module cannot be loaded.

Ensures the given module is loaded.

Returns an environment for evaluation.

Evaluates the given file.

Evaluates the given quoted contents with binding and env.

Evaluates the contents given by string.

Returns the docs for the given module or path to .beam file.

Formats the given code string.

Returns the value of a given compiler option.

Returns true if the module is loaded.

Prepends a path to the Erlang VM code path list.

Prepends a list of paths to the Erlang VM code path list.

Prints a diagnostic into the standard error.

Purge compiler modules.

Stores a compilation option.

Converts a quoted expression to an algebra document using Elixir's formatter rules.

Requires the given file.

Lists all required files.

Converts the given string to its quoted form.

Converts the given string to its quoted form.

Converts the given string to its quoted form and a list of comments.

Converts the given string to its quoted form and a list of comments.

Removes files from the required files list.

Executes the given fun and capture all diagnostics.

Types

@type binding() :: [{atom() | tuple(), any()}]

A list with all variables and their values.

The binding keys are usually atoms, but they may be a tuple for variables defined in a different context.

@type diagnostic(severity) :: %{
  :source => Path.t() | nil,
  :file => Path.t() | nil,
  :severity => severity,
  :message => String.t(),
  :position => position(),
  :stacktrace => Exception.stacktrace(),
  :span => {line :: pos_integer(), column :: pos_integer()} | nil,
  optional(:details) => term(),
  optional(any()) => any()
}

Diagnostics returned by the compiler and code evaluation.

The file and position relate to where the diagnostic should be shown. If there is a file and position, then the diagnostic is precise and you can use the given file and position for generating snippets, IDEs annotations, and so on. An optional span is available with the line and column the diagnostic ends.

Otherwise, a stacktrace may be given, which you can place your own heuristics to provide better reporting.

The source field points to the source file the compiler tracked the error to. For example, a file lib/foo.ex may embed .eex templates from lib/foo/bar.eex. A syntax error on the EEx template will point to file lib/foo/bar.eex but the source is lib/foo.ex.

@type line() :: non_neg_integer()

The line. 0 indicates no line.

@type position() :: line() | {line :: pos_integer(), column :: pos_integer()}

The position of the diagnostic.

Can be either a line number or a {line, column}. Line and columns numbers are one-based. A position of 0 represents unknown.

Functions

Link to this function

append_path(path, opts \\ [])

View Source
@spec append_path(Path.t(), [{:cache, boolean()}]) :: true | false

Appends a path to the Erlang VM code path list.

This is the list of directories the Erlang VM uses for finding module code. The list of files is managed per Erlang VM node.

The path is expanded with Path.expand/1 before being appended. It requires the path to exist. Returns a boolean indicating if the path was successfully added.

Examples

Code.append_path(".")
#=> true

Code.append_path("/does_not_exist")
#=> false

Options

  • :cache - (since v1.15.0) when true, the code path is cached the first time it is traversed in order to reduce file system operations. It requires Erlang/OTP 26, otherwise it is a no-op.
Link to this function

append_paths(paths, opts \\ [])

View Source (since 1.15.0)
@spec append_paths([Path.t()], [{:cache, boolean()}]) :: :ok

Appends a list of paths to the Erlang VM code path list.

This is the list of directories the Erlang VM uses for finding module code. The list of files is managed per Erlang VM node.

All paths are expanded with Path.expand/1 before being appended. Only existing paths are appended. This function always returns :ok, regardless of how many paths were appended. Use append_path/1 if you need more control.

Examples

Code.append_paths([".", "/does_not_exist"])
#=> :ok

Options

  • :cache - when true, the code path is cached the first time it is traversed in order to reduce file system operations. It requires Erlang/OTP 26, otherwise it is a no-op.
Link to this function

available_compiler_options()

View Source
@spec available_compiler_options() :: [atom()]

Returns a list with all available compiler options.

For a description of all options, see put_compiler_option/2.

Examples

Code.available_compiler_options()
#=> [:docs, :debug_info, ...]
Link to this function

can_await_module_compilation?()

View Source (since 1.11.0)
@spec can_await_module_compilation?() :: boolean()

Returns true if the current process can await for module compilation.

When compiling Elixir code via Kernel.ParallelCompiler, which is used by Mix and elixirc, calling a module that has not yet been compiled will block the caller until the module becomes available. Executing Elixir scripts, such as passing a filename to elixir, does not await.

Link to this function

compile_file(file, relative_to \\ nil)

View Source (since 1.7.0)
@spec compile_file(binary(), nil | binary()) :: [{module(), binary()}]

Compiles the given file.

Accepts relative_to as an argument to tell where the file is located.

Returns a list of tuples where the first element is the module name and the second one is its bytecode (as a binary). Opposite to require_file/2, it does not track the filename of the compiled file.

If you would like to get the result of evaluating file rather than the modules defined in it, see eval_file/2.

For compiling many files concurrently, see Kernel.ParallelCompiler.compile/2.

Link to this function

compile_quoted(quoted, file \\ "nofile")

View Source
@spec compile_quoted(Macro.t(), binary()) :: [{module(), binary()}]

Compiles the quoted expression.

Returns a list of tuples where the first element is the module name and the second one is its bytecode (as a binary). A file can be given as second argument which will be used for reporting warnings and errors.

Link to this function

compile_string(string, file \\ "nofile")

View Source
@spec compile_string(List.Chars.t(), binary()) :: [{module(), binary()}]

Compiles the given string.

Returns a list of tuples where the first element is the module name and the second one is its bytecode (as a binary). A file can be given as a second argument which will be used for reporting warnings and errors.

Warning: string can be any Elixir code and code can be executed with the same privileges as the Erlang VM: this means that such code could compromise the machine (for example by executing system commands). Don't use compile_string/2 with untrusted input (such as strings coming from the network).

@spec compiler_options() :: map()

Gets all compilation options from the code server.

To get individual options, see get_compiler_option/1. For a description of all options, see put_compiler_option/2.

Examples

Code.compiler_options()
#=> %{debug_info: true, docs: true, ...}
@spec compiler_options(Enumerable.t({atom(), term()})) :: %{
  optional(atom()) => term()
}

Stores all given compilation options.

Changing the compilation options affect all processes running in a given Erlang VM node. To store individual options and for a description of all options, see put_compiler_option/2.

Returns a map with previous values.

Examples

Code.compiler_options(warnings_as_errors: true)
#=> %{warnings_as_errors: false}
@spec delete_path(Path.t()) :: boolean()

Deletes a path from the Erlang VM code path list.

This is the list of directories the Erlang VM uses for finding module code. The list of files is managed per Erlang VM node.

The path is expanded with Path.expand/1 before being deleted. If the path does not exist, this function returns false.

Examples

Code.prepend_path(".")
Code.delete_path(".")
#=> true

Code.delete_path("/does_not_exist")
#=> false
Link to this function

delete_paths(paths)

View Source (since 1.15.0)
@spec delete_paths([Path.t()]) :: :ok

Deletes a list of paths from the Erlang VM code path list.

This is the list of directories the Erlang VM uses for finding module code. The list of files is managed per Erlang VM node.

The path is expanded with Path.expand/1 before being deleted. If the path does not exist, this function returns false.

Link to this function

ensure_all_loaded(modules)

View Source (since 1.15.0)
@spec ensure_all_loaded([module()]) :: :ok | {:error, [{module(), reason}]}
when reason: :badfile | :nofile | :on_load_failure

Ensures the given modules are loaded.

Similar to ensure_loaded/1, but accepts a list of modules instead of a single module, and loads all of them.

If all modules load successfully, returns :ok. Otherwise, returns {:error, errors} where errors is a list of tuples made of the module and the reason it failed to load.

Examples

iex> Code.ensure_all_loaded([Atom, String])
:ok

iex> Code.ensure_all_loaded([Atom, DoesNotExist])
{:error, [{DoesNotExist, :nofile}]}
Link to this function

ensure_all_loaded!(modules)

View Source (since 1.15.0)
@spec ensure_all_loaded!([module()]) :: :ok

Same as ensure_all_loaded/1 but raises if any of the modules cannot be loaded.

@spec ensure_compiled(module()) ::
  {:module, module()}
  | {:error, :embedded | :badfile | :nofile | :on_load_failure | :unavailable}

Similar to ensure_compiled!/1 but indicates you can continue without said module.

While ensure_compiled!/1 indicates to the Elixir compiler you can only continue when said module is available, this function indicates you may continue compilation without said module.

If it succeeds in loading the module, it returns {:module, module}. If not, returns {:error, reason} with the error reason. If the module being checked is currently in a compiler deadlock, this function returns {:error, :unavailable}. Unavailable doesn't necessarily mean the module doesn't exist, just that it is not currently available, but it (or may not) become available in the future.

Therefore, if you can only continue if the module is available, use ensure_compiled!/1 instead. In particular, do not do this:

case Code.ensure_compiled(module) do
  {:module, _} -> module
  {:error, _} -> raise ...
end

See the module documentation for more information on code loading.

Link to this function

ensure_compiled!(module)

View Source (since 1.12.0)
@spec ensure_compiled!(module()) :: module()

Ensures the given module is compiled and loaded.

If the module is already loaded, it works as no-op. If the module was not compiled yet, ensure_compiled!/1 halts the compilation of the caller until the module given to ensure_compiled!/1 becomes available or all files for the current project have been compiled. If compilation finishes and the module is not available or is in a deadlock, an error is raised.

Given this function halts compilation, use it carefully. In particular, avoid using it to guess which modules are in the system. Overuse of this function can also lead to deadlocks, where two modules check at the same time if the other is compiled. This returns a specific unavailable error code, where we cannot successfully verify a module is available or not.

See the module documentation for more information on code loading.

@spec ensure_loaded(module()) ::
  {:module, module()}
  | {:error, :embedded | :badfile | :nofile | :on_load_failure}

Ensures the given module is loaded.

If the module is already loaded, this works as no-op. If the module was not yet loaded, it tries to load it.

If it succeeds in loading the module, it returns {:module, module}. If not, returns {:error, reason} with the error reason.

See the module documentation for more information on code loading.

Examples

iex> Code.ensure_loaded(Atom)
{:module, Atom}

iex> Code.ensure_loaded(DoesNotExist)
{:error, :nofile}
Link to this function

ensure_loaded!(module)

View Source (since 1.12.0)
@spec ensure_loaded!(module()) :: module()

Same as ensure_loaded/1 but raises if the module cannot be loaded.

@spec ensure_loaded?(module()) :: boolean()

Ensures the given module is loaded.

Similar to ensure_loaded/1, but returns true if the module is already loaded or was successfully loaded. Returns false otherwise.

Examples

iex> Code.ensure_loaded?(String)
true
Link to this function

env_for_eval(env_or_opts)

View Source (since 1.14.0)

Returns an environment for evaluation.

It accepts either a Macro.Env, that is then pruned and prepared, or a list of options. It returns an environment that is ready for evaluation.

Most functions in this module will automatically prepare the given environment for evaluation, so you don't need to explicitly call this function, with the exception of eval_quoted_with_env/3, which was designed precisely to be called in a loop, to implement features such as interactive shells or anything else with multiple evaluations.

Options

If an env is not given, the options can be:

  • :file - the file to be considered in the evaluation

  • :line - the line on which the script starts

Link to this function

eval_file(file, relative_to \\ nil)

View Source
@spec eval_file(binary(), nil | binary()) :: {term(), binding()}

Evaluates the given file.

Accepts relative_to as an argument to tell where the file is located.

While require_file/2 and compile_file/2 return the loaded modules and their bytecode, eval_file/2 simply evaluates the file contents and returns the evaluation result and its binding (exactly the same return value as eval_string/3).

Link to this function

eval_quoted(quoted, binding \\ [], env_or_opts \\ [])

View Source
@spec eval_quoted(Macro.t(), binding(), Macro.Env.t() | keyword()) ::
  {term(), binding()}

Evaluates the quoted contents.

Warning: Calling this function inside a macro is considered bad practice as it will attempt to evaluate runtime values at compile time. Macro arguments are typically transformed by unquoting them into the returned quoted expressions (instead of evaluated).

See eval_string/3 for a description of binding and opts.

Examples

iex> contents = quote(do: var!(a) + var!(b))
iex> {result, binding} = Code.eval_quoted(contents, [a: 1, b: 2], file: __ENV__.file, line: __ENV__.line)
iex> result
3
iex> Enum.sort(binding)
[a: 1, b: 2]

For convenience, you can pass __ENV__/0 as the opts argument and all options will be automatically extracted from the current environment:

iex> contents = quote(do: var!(a) + var!(b))
iex> {result, binding} = Code.eval_quoted(contents, [a: 1, b: 2], __ENV__)
iex> result
3
iex> Enum.sort(binding)
[a: 1, b: 2]
Link to this function

eval_quoted_with_env(quoted, binding, env, opts \\ [])

View Source (since 1.14.0)
@spec eval_quoted_with_env(Macro.t(), binding(), Macro.Env.t(), keyword()) ::
  {term(), binding(), Macro.Env.t()}

Evaluates the given quoted contents with binding and env.

This function is meant to be called in a loop, to implement features such as interactive shells or anything else with multiple evaluations. Therefore, the first time you call this function, you must compute the initial environment with env_for_eval/1. The remaining calls must pass the environment that was returned by this function.

Options

  • :prune_binding - (since v1.14.2) prune binding to keep only variables read or written by the evaluated code. Note that variables used by modules are always pruned, even if later used by the modules. You can submit to the :on_module tracer event and access the variables used by the module from its environment.
Link to this function

eval_string(string, binding \\ [], opts \\ [])

View Source
@spec eval_string(List.Chars.t(), binding(), Macro.Env.t() | keyword()) ::
  {term(), binding()}

Evaluates the contents given by string.

The binding argument is a list of all variables and their values. The opts argument is a keyword list of environment options.

Warning: string can be any Elixir code and will be executed with the same privileges as the Erlang VM: this means that such code could compromise the machine (for example by executing system commands). Don't use eval_string/3 with untrusted input (such as strings coming from the network).

Options

Options can be:

  • :file - the file to be considered in the evaluation

  • :line - the line on which the script starts

Additionally, you may also pass an environment as second argument, so the evaluation happens within that environment.

Returns a tuple of the form {value, binding}, where value is the value returned from evaluating string. If an error occurs while evaluating string, an exception will be raised.

binding is a list with all variable names and their values after evaluating string. The binding keys are usually atoms, but they may be a tuple for variables defined in a different context. The names are in no particular order.

Examples

iex> {result, binding} = Code.eval_string("a + b", [a: 1, b: 2], file: __ENV__.file, line: __ENV__.line)
iex> result
3
iex> Enum.sort(binding)
[a: 1, b: 2]

iex> {result, binding} = Code.eval_string("c = a + b", [a: 1, b: 2], __ENV__)
iex> result
3
iex> Enum.sort(binding)
[a: 1, b: 2, c: 3]

iex> {result, binding} = Code.eval_string("a = a + b", [a: 1, b: 2])
iex> result
3
iex> Enum.sort(binding)
[a: 3, b: 2]

For convenience, you can pass __ENV__/0 as the opts argument and all imports, requires and aliases defined in the current environment will be automatically carried over:

iex> {result, binding} = Code.eval_string("a + b", [a: 1, b: 2], __ENV__)
iex> result
3
iex> Enum.sort(binding)
[a: 1, b: 2]
Link to this function

fetch_docs(module_or_path)

View Source (since 1.7.0)
@spec fetch_docs(module() | String.t()) ::
  {:docs_v1, annotation, beam_language, format, module_doc :: doc_content,
   metadata, docs :: [doc_element]}
  | {:error,
     :module_not_found
     | :chunk_not_found
     | {:invalid_chunk, binary()}
     | :invalid_beam}
when annotation: :erl_anno.anno(),
     beam_language: :elixir | :erlang | atom(),
     doc_content: %{optional(binary()) => binary()} | :none | :hidden,
     doc_element:
       {{kind :: atom(), function_name :: atom(), arity()}, annotation,
        signature, doc_content, metadata},
     format: binary(),
     signature: [binary()],
     metadata: map()

Returns the docs for the given module or path to .beam file.

When given a module name, it finds its BEAM code and reads the docs from it.

When given a path to a .beam file, it will load the docs directly from that file.

It returns the term stored in the documentation chunk in the format defined by EEP 48 or {:error, reason} if the chunk is not available.

Examples

# Module documentation of an existing module
iex> {:docs_v1, _, :elixir, _, %{"en" => module_doc}, _, _} = Code.fetch_docs(Atom)
iex> module_doc |> String.split("\n") |> Enum.at(0)
"Atoms are constants whose values are their own name."

# A module that doesn't exist
iex> Code.fetch_docs(ModuleNotGood)
{:error, :module_not_found}
Link to this function

format_file!(file, opts \\ [])

View Source (since 1.6.0)
@spec format_file!(
  binary(),
  keyword()
) :: iodata()

Formats a file.

See format_string!/2 for more information on code formatting and available options.

Link to this function

format_string!(string, opts \\ [])

View Source (since 1.6.0)
@spec format_string!(
  binary(),
  keyword()
) :: iodata()

Formats the given code string.

The formatter receives a string representing Elixir code and returns iodata representing the formatted code according to pre-defined rules.

Options

  • :file - the file which contains the string, used for error reporting

  • :line - the line the string starts, used for error reporting

  • :line_length - the line length to aim for when formatting the document. Defaults to 98. This value indicates when an expression should be broken over multiple lines but it is not guaranteed to do so. See the "Line length" section below for more information

  • :locals_without_parens - a keyword list of name and arity pairs that should be kept without parens whenever possible. The arity may be the atom :*, which implies all arities of that name. The formatter already includes a list of functions and this option augments this list.

  • :force_do_end_blocks (since v1.9.0) - when true, converts all inline usages of do: ..., else: ... and friends into do-end blocks. Defaults to false. Note that this option is convergent: once you set it to true, all keywords will be converted. If you set it to false later on, do-end blocks won't be converted back to keywords.

  • :normalize_bitstring_modifiers (since v1.14.0) - when true, removes unnecessary parentheses in known bitstring modifiers, for example <<foo::binary()>> becomes <<foo::binary>>, or adds parentheses for custom modifiers, where <<foo::custom_type>> becomes <<foo::custom_type()>>. Defaults to true. This option changes the AST.

  • :normalize_charlists_as_sigils (since v1.15.0) - when true, formats charlists as ~c sigils, for example 'foo' becomes ~c"foo". Defaults to true. This option changes the AST.

Design principles

The formatter was designed under three principles.

First, the formatter never changes the semantics of the code. This means the input AST and the output AST are almost always equivalent. The only cases where the formatter will change the AST is when the input AST would cause compiler warnings and the output AST won't. The cases where the formatter changes the AST can be disabled through formatting options if desired.

The second principle is to provide as little configuration as possible. This eases the formatter adoption by removing contention points while making sure a single style is followed consistently by the community as a whole.

The formatter does not hard code names. The formatter will not behave specially because a function is named defmodule, def, or the like. This principle mirrors Elixir's goal of being an extensible language where developers can extend the language with new constructs as if they were part of the language. When it is absolutely necessary to change behavior based on the name, this behavior should be configurable, such as the :locals_without_parens option.

Running the formatter

The formatter attempts to fit the most it can on a single line and introduces line breaks wherever possible when it cannot.

In some cases, this may lead to undesired formatting. Therefore, some code generated by the formatter may not be aesthetically pleasing and may require explicit intervention from the developer. That's why we do not recommend to run the formatter blindly in an existing codebase. Instead you should format and sanity check each formatted file.

For example, the formatter may break a long function definition over multiple clauses:

def my_function(
  %User{name: name, age: age, ...},
  arg1,
  arg2
) do
  ...
end

While the code above is completely valid, you may prefer to match on the struct variables inside the function body in order to keep the definition on a single line:

def my_function(%User{} = user, arg1, arg2) do
  %{name: name, age: age, ...} = user
  ...
end

In some situations, you can use the fact the formatter does not generate elegant code as a hint for refactoring. Take this code:

def board?(board_id, %User{} = user, available_permissions, required_permissions) do
  Tracker.OrganizationMembers.user_in_organization?(user.id, board.organization_id) and
    required_permissions == Enum.to_list(MapSet.intersection(MapSet.new(required_permissions), MapSet.new(available_permissions)))
end

The code above has very long lines and running the formatter is not going to address this issue. In fact, the formatter may make it more obvious that you have complex expressions:

def board?(board_id, %User{} = user, available_permissions, required_permissions) do
  Tracker.OrganizationMembers.user_in_organization?(user.id, board.organization_id) and
    required_permissions ==
      Enum.to_list(
        MapSet.intersection(
          MapSet.new(required_permissions),
          MapSet.new(available_permissions)
        )
      )
end

Take such cases as a suggestion that your code should be refactored:

def board?(board_id, %User{} = user, available_permissions, required_permissions) do
  Tracker.OrganizationMembers.user_in_organization?(user.id, board.organization_id) and
    matching_permissions?(required_permissions, available_permissions)
end

defp matching_permissions?(required_permissions, available_permissions) do
  intersection =
    required_permissions
    |> MapSet.new()
    |> MapSet.intersection(MapSet.new(available_permissions))
    |> Enum.to_list()

  required_permissions == intersection
end

To sum it up: since the formatter cannot change the semantics of your code, sometimes it is necessary to tweak or refactor the code to get optimal formatting. To help better understand how to control the formatter, we describe in the next sections the cases where the formatter keeps the user encoding and how to control multiline expressions.

Line length

Another point about the formatter is that the :line_length configuration indicates when an expression should be broken over multiple lines but it is not guaranteed to do so. In many cases, it is not possible for the formatter to break your code apart, which means it will go over the line length. For example, if you have a long string:

"this is a very long string that will go over the line length"

The formatter doesn't know how to break it apart without changing the code underlying syntax representation, so it is up to you to step in:

"this is a very long string " <>
   "that will go over the line length"

The string concatenation makes the code fit on a single line and also gives more options to the formatter.

This may also appear in keywords such as do/end blocks and operators, where the do keyword may go over the line length because there is no opportunity for the formatter to introduce a line break in a readable way. For example, if you do:

case very_long_expression() do
end

And only the do keyword is beyond the line length, Elixir will not emit this:

case very_long_expression()
do
end

So it prefers to not touch the line at all and leave do above the line limit.

Keeping user's formatting

The formatter respects the input format in some cases. Those are listed below:

  • Insignificant digits in numbers are kept as is. The formatter, however, always inserts underscores for decimal numbers with more than 5 digits and converts hexadecimal digits to uppercase

  • Strings, charlists, atoms and sigils are kept as is. No character is automatically escaped or unescaped. The choice of delimiter is also respected from the input

  • Newlines inside blocks are kept as in the input except for:

    1. expressions that take multiple lines will always have an empty line before and after and 2) empty lines are always squeezed together into a single empty line
  • The choice between :do keyword and do-end blocks is left to the user

  • Lists, tuples, bitstrings, maps, structs and function calls will be broken into multiple lines if they are followed by a newline in the opening bracket and preceded by a new line in the closing bracket

  • Newlines before certain operators (such as the pipeline operators) and before other operators (such as comparison operators)

The behaviors above are not guaranteed. We may remove or add new rules in the future. The goal of documenting them is to provide better understanding on what to expect from the formatter.

Multi-line lists, maps, tuples, and the like

You can force lists, tuples, bitstrings, maps, structs and function calls to have one entry per line by adding a newline after the opening bracket and a new line before the closing bracket lines. For example:

[
  foo,
  bar
]

If there are no newlines around the brackets, then the formatter will try to fit everything on a single line, such that the snippet below

[foo,
 bar]

will be formatted as

[foo, bar]

You can also force function calls and keywords to be rendered on multiple lines by having each entry on its own line:

defstruct name: nil,
          age: 0

The code above will be kept with one keyword entry per line by the formatter. To avoid that, just squash everything into a single line.

Parens and no parens in function calls

Elixir has two syntaxes for function calls. With parens and no parens. By default, Elixir will add parens to all calls except for:

  1. calls that have do-end blocks
  2. local calls without parens where the name and arity of the local call is also listed under :locals_without_parens (except for calls with arity 0, where the compiler always require parens)

The choice of parens and no parens also affects indentation. When a function call with parens doesn't fit on the same line, the formatter introduces a newline around parens and indents the arguments with two spaces:

some_call(
  arg1,
  arg2,
  arg3
)

On the other hand, function calls without parens are always indented by the function call length itself, like this:

some_call arg1,
          arg2,
          arg3

If the last argument is a data structure, such as maps and lists, and the beginning of the data structure fits on the same line as the function call, then no indentation happens, this allows code like this:

Enum.reduce(some_collection, initial_value, fn element, acc ->
  # code
end)

some_function_without_parens %{
  foo: :bar,
  baz: :bat
}

Code comments

The formatter also handles code comments in a way to guarantee a space is always added between the beginning of the comment (#) and the next character.

The formatter also extracts all trailing comments to their previous line. For example, the code below

hello #world

will be rewritten to

# world
hello

Because code comments are handled apart from the code representation (AST), there are some situations where code comments are seen as ambiguous by the code formatter. For example, the comment in the anonymous function below

fn
  arg1 ->
    body1
    # comment

  arg2 ->
    body2
end

and in this one

fn
  arg1 ->
    body1

  # comment
  arg2 ->
    body2
end

are considered equivalent (the nesting is discarded alongside most of user formatting). In such cases, the code formatter will always format to the latter.

Newlines

The formatter converts all newlines in code from \r\n to \n.

Link to this function

get_compiler_option(key)

View Source (since 1.10.0)
@spec get_compiler_option(atom()) :: term()

Returns the value of a given compiler option.

For a description of all options, see put_compiler_option/2.

Examples

Code.get_compiler_option(:debug_info)
#=> true
Link to this function

loaded?(module)

View Source (since 1.15.0)
@spec loaded?(module()) :: boolean()

Returns true if the module is loaded.

This function doesn't attempt to load the module. For such behavior, ensure_loaded?/1 can be used.

Examples

iex> Code.loaded?(Atom)
true

iex> Code.loaded?(NotYetLoaded)
false
Link to this function

prepend_path(path, opts \\ [])

View Source
@spec prepend_path(Path.t(), [{:cache, boolean()}]) :: boolean()

Prepends a path to the Erlang VM code path list.

This is the list of directories the Erlang VM uses for finding module code. The list of files is managed per Erlang VM node.

The path is expanded with Path.expand/1 before being prepended. It requires the path to exist. Returns a boolean indicating if the path was successfully added.

Examples

Code.prepend_path(".")
#=> true

Code.prepend_path("/does_not_exist")
#=> false

Options

  • :cache - (since v1.15.0) when true, the code path is cached the first time it is traversed in order to reduce file system operations. It requires Erlang/OTP 26, otherwise it is a no-op.
Link to this function

prepend_paths(paths, opts \\ [])

View Source (since 1.15.0)
@spec prepend_paths([Path.t()], [{:cache, boolean()}]) :: :ok

Prepends a list of paths to the Erlang VM code path list.

This is the list of directories the Erlang VM uses for finding module code. The list of files is managed per Erlang VM node.

All paths are expanded with Path.expand/1 before being prepended. Only existing paths are prepended. This function always returns :ok, regardless of how many paths were prepended. Use prepend_path/1 if you need more control.

Examples

Code.prepend_paths([".", "/does_not_exist"])
#=> :ok

Options

  • :cache - when true, the code path is cached the first time it is traversed in order to reduce file system operations. It requires Erlang/OTP 26, otherwise it is a no-op.
Link to this function

purge_compiler_modules()

View Source (since 1.7.0)
@spec purge_compiler_modules() :: {:ok, non_neg_integer()}

Purge compiler modules.

The compiler utilizes temporary modules to compile code. For example, elixir_compiler_1, elixir_compiler_2, and so on. In case the compiled code stores references to anonymous functions or similar, the Elixir compiler may be unable to reclaim those modules, keeping an unnecessary amount of code in memory and eventually leading to modules such as elixir_compiler_12345.

This function purges all modules currently kept by the compiler, allowing old compiler module names to be reused. If there are any processes running any code from such modules, they will be terminated too.

This function is only meant to be called if you have a long running node that is constantly evaluating code.

It returns {:ok, number_of_modules_purged}.

Link to this function

put_compiler_option(key, value)

View Source (since 1.10.0)
@spec put_compiler_option(atom(), term()) :: :ok

Stores a compilation option.

Changing the compilation options affect all processes running in a given Erlang VM node.

Available options are:

  • :docs - when true, retains documentation in the compiled module. Defaults to true.

  • :debug_info - when true, retains debug information in the compiled module. Defaults to true. This enables static analysis tools as it allows developers to partially reconstruct the original source code. Therefore, disabling :debug_info is not recommended as it removes the ability of the Elixir compiler and other tools to provide feedback. If you want to remove the :debug_info while deploying, tools like mix release already do such by default. Additionally, mix test disables it via the :test_elixirc_options project configuration option. This option can also be overridden per module using the @compile directive.

  • :ignore_already_consolidated (since v1.10.0) - when true, does not warn when a protocol has already been consolidated and a new implementation is added. Defaults to false.

  • :ignore_module_conflict - when true, does not warn when a module has already been defined. Defaults to false.

  • :relative_paths - when true, uses relative paths in quoted nodes, warnings, and errors generated by the compiler. Note disabling this option won't affect runtime warnings and errors. Defaults to true.

  • :warnings_as_errors - causes compilation to fail when warnings are generated. Defaults to false.

  • :no_warn_undefined (since v1.10.0) - list of modules and {Mod, fun, arity} tuples that will not emit warnings that the module or function does not exist at compilation time. Pass atom :all to skip warning for all undefined functions. This can be useful when doing dynamic compilation. Defaults to [].

  • :tracers (since v1.10.0) - a list of tracers (modules) to be used during compilation. See the module docs for more information. Defaults to [].

  • :parser_options (since v1.10.0) - a keyword list of options to be given to the parser when compiling files. It accepts the same options as string_to_quoted/2 (except by the options that change the AST itself). This can be used in combination with the tracer to retrieve localized information about events happening during compilation. Defaults to [columns: true]. This option only affects code compilation functions, such as compile_string/2 and compile_file/2 but not string_to_quoted/2 and friends, as the latter is used for other purposes beyond compilation.

  • :on_undefined_variable (since v1.15.0) - either :raise or :warn. When :raise (the default), undefined variables will trigger a compilation error. You may be set it to :warn if you want undefined variables to emit a warning and expand as to a local call to the zero-arity function of the same name (for example, node would be expanded as node()). This :warn behavior only exists for compatibility reasons when working with old dependencies, its usage is discouraged and it will be removed in future releases.

It always returns :ok. Raises an error for invalid options.

Examples

Code.put_compiler_option(:debug_info, true)
#=> :ok
Link to this function

quoted_to_algebra(quoted, opts \\ [])

View Source (since 1.13.0)
@spec quoted_to_algebra(
  Macro.t(),
  keyword()
) :: Inspect.Algebra.t()

Converts a quoted expression to an algebra document using Elixir's formatter rules.

The algebra document can be converted into a string by calling:

doc
|> Inspect.Algebra.format(:infinity)
|> IO.iodata_to_binary()

For a high-level function that does the same, see Macro.to_string/1.

Formatting considerations

The Elixir AST does not contain metadata for literals like strings, lists, or tuples with two elements, which means that the produced algebra document will not respect all of the user preferences and comments may be misplaced. To get better results, you can use the :token_metadata, :unescape and :literal_encoder options to string_to_quoted/2 to provide additional information to the formatter:

[
  literal_encoder: &{:ok, {:__block__, &2, [&1]}},
  token_metadata: true,
  unescape: false
]

This will produce an AST that contains information such as do blocks start and end lines or sigil delimiters, and by wrapping literals in blocks they can now hold metadata like line number, string delimiter and escaped sequences, or integer formatting (such as 0x2a instead of 47). However, note this AST is not valid. If you evaluate it, it won't have the same semantics as the regular Elixir AST due to the :unescape and :literal_encoder options. However, those options are useful if you're doing source code manipulation, where it's important to preserve user choices and comments placing.

Options

  • :comments - the list of comments associated with the quoted expression. Defaults to []. It is recommended that both :token_metadata and :literal_encoder options are given to string_to_quoted_with_comments/2 in order to get proper placement for comments

  • :escape - when true, escaped sequences like \n will be escaped into \\n. If the :unescape option was set to false when using string_to_quoted/2, setting this option to false will prevent it from escaping the sequences twice. Defaults to true.

  • :locals_without_parens - a keyword list of name and arity pairs that should be kept without parens whenever possible. The arity may be the atom :*, which implies all arities of that name. The formatter already includes a list of functions and this option augments this list.

  • :syntax_colors - a keyword list of colors the output is colorized. See Inspect.Opts for more information.

Link to this function

require_file(file, relative_to \\ nil)

View Source
@spec require_file(binary(), nil | binary()) :: [{module(), binary()}] | nil

Requires the given file.

Accepts relative_to as an argument to tell where the file is located. If the file was already required, require_file/2 doesn't do anything and returns nil.

Note that if require_file/2 is invoked by different processes concurrently, the first process to invoke require_file/2 acquires a lock and the remaining ones will block until the file is available. This means that if require_file/2 is called more than once with a given file, that file will be compiled only once. The first process to call require_file/2 will get the list of loaded modules, others will get nil. The list of required files is managed per Erlang VM node.

See compile_file/2 if you would like to compile a file without tracking its filenames. Finally, if you would like to get the result of evaluating a file rather than the modules defined in it, see eval_file/2.

Examples

If the file has not been required, it returns the list of modules:

modules = Code.require_file("eex_test.exs", "../eex/test")
List.first(modules)
#=> {EExTest.Compiled, <<70, 79, 82, 49, ...>>}

If the file has been required, it returns nil:

Code.require_file("eex_test.exs", "../eex/test")
#=> nil
Link to this function

required_files()

View Source (since 1.7.0)
@spec required_files() :: [binary()]

Lists all required files.

Examples

Code.require_file("../eex/test/eex_test.exs")
List.first(Code.required_files()) =~ "eex_test.exs"
#=> true
Link to this function

string_to_quoted(string, opts \\ [])

View Source
@spec string_to_quoted(
  List.Chars.t(),
  keyword()
) ::
  {:ok, Macro.t()}
  | {:error, {location :: keyword(), binary() | {binary(), binary()}, binary()}}

Converts the given string to its quoted form.

Returns {:ok, quoted_form} if it succeeds, {:error, {meta, message_info, token}} otherwise.

Options

  • :file - the filename to be reported in case of parsing errors. Defaults to "nofile".

  • :line - the starting line of the string being parsed. Defaults to 1.

  • :column - (since v1.11.0) the starting column of the string being parsed. Defaults to 1.

  • :columns - when true, attach a :column key to the quoted metadata. Defaults to false.

  • :unescape (since v1.10.0) - when false, preserves escaped sequences. For example, "null byte\\t\\x00" will be kept as is instead of being converted to a bitstring literal. Note if you set this option to false, the resulting AST is no longer valid, but it can be useful to analyze/transform source code, typically in combination with quoted_to_algebra/2. Defaults to true.

  • :existing_atoms_only - when true, raises an error when non-existing atoms are found by the tokenizer. Defaults to false.

  • :token_metadata (since v1.10.0) - when true, includes token-related metadata in the expression AST, such as metadata for do and end tokens, for closing tokens, end of expressions, as well as delimiters for sigils. See Macro.metadata/0. Defaults to false.

  • :literal_encoder (since v1.10.0) - how to encode literals in the AST. It must be a function that receives two arguments, the literal and its metadata, and it must return {:ok, ast :: Macro.t} or {:error, reason :: binary}. If you return anything than the literal itself as the term, then the AST is no longer valid. This option may still useful for textual analysis of the source code.

  • :static_atoms_encoder - the static atom encoder function, see "The :static_atoms_encoder function" section below. Note this option overrides the :existing_atoms_only behavior for static atoms but :existing_atoms_only is still used for dynamic atoms, such as atoms with interpolations.

  • :emit_warnings (since v1.16.0) - when false, does not emit tokenizing/parsing related warnings. Defaults to true.

Macro.to_string/2

The opposite of converting a string to its quoted form is Macro.to_string/2, which converts a quoted form to a string/binary representation.

The :static_atoms_encoder function

When static_atoms_encoder: &my_encoder/2 is passed as an argument, my_encoder/2 is called every time the tokenizer needs to create a "static" atom. Static atoms are atoms in the AST that function as aliases, remote calls, local calls, variable names, regular atoms and keyword lists.

The encoder function will receive the atom name (as a binary) and a keyword list with the current file, line and column. It must return {:ok, token :: term} | {:error, reason :: binary}.

The encoder function is supposed to create an atom from the given string. To produce a valid AST, it is required to return {:ok, term}, where term is an atom. It is possible to return something other than an atom, however, in that case the AST is no longer "valid" in that it cannot be used to compile or evaluate Elixir code. A use case for this is if you want to use the Elixir parser in a user-facing situation, but you don't want to exhaust the atom table.

The atom encoder is not called for all atoms that are present in the AST. It won't be invoked for the following atoms:

  • operators (:+, :-, and so on)

  • syntax keywords (fn, do, else, and so on)

  • atoms containing interpolation (:"#{1 + 1} is two"), as these atoms are constructed at runtime

  • atoms used to represent single-letter sigils like :sigil_X (but multi-letter sigils like :sigil_XYZ are encoded).

Link to this function

string_to_quoted!(string, opts \\ [])

View Source
@spec string_to_quoted!(
  List.Chars.t(),
  keyword()
) :: Macro.t()

Converts the given string to its quoted form.

It returns the AST if it succeeds, raises an exception otherwise. The exception is a TokenMissingError in case a token is missing (usually because the expression is incomplete), MismatchedDelimiterError (in case of mismatched opening and closing delimiters) and SyntaxError otherwise.

Check string_to_quoted/2 for options information.

Link to this function

string_to_quoted_with_comments(string, opts \\ [])

View Source (since 1.13.0)
@spec string_to_quoted_with_comments(
  List.Chars.t(),
  keyword()
) ::
  {:ok, Macro.t(), [map()]} | {:error, {location :: keyword(), term(), term()}}

Converts the given string to its quoted form and a list of comments.

This function is useful when performing textual changes to the source code, while preserving information like comments and literals position.

Returns {:ok, quoted_form, comments} if it succeeds, {:error, {line, error, token}} otherwise.

Comments are maps with the following fields:

  • :line - The line number of the source code

  • :text - The full text of the comment, including the leading #

  • :previous_eol_count - How many end of lines there are between the comment and the previous AST node or comment

  • :next_eol_count - How many end of lines there are between the comment and the next AST node or comment

Check string_to_quoted/2 for options information.

Examples

iex> Code.string_to_quoted_with_comments("""
...> :foo
...>
...> # Hello, world!
...>
...>
...> # Some more comments!
...> """)
{:ok, :foo, [
  %{line: 3, column: 1, previous_eol_count: 2, next_eol_count: 3, text: "# Hello, world!"},
  %{line: 6, column: 1, previous_eol_count: 3, next_eol_count: 1, text: "# Some more comments!"},
]}

iex> Code.string_to_quoted_with_comments(":foo # :bar")
{:ok, :foo, [
  %{line: 1, column: 6, previous_eol_count: 0, next_eol_count: 0, text: "# :bar"}
]}
Link to this function

string_to_quoted_with_comments!(string, opts \\ [])

View Source (since 1.13.0)
@spec string_to_quoted_with_comments!(
  List.Chars.t(),
  keyword()
) :: {Macro.t(), [map()]}

Converts the given string to its quoted form and a list of comments.

Returns the AST and a list of comments if it succeeds, raises an exception otherwise. The exception is a TokenMissingError in case a token is missing (usually because the expression is incomplete), SyntaxError otherwise.

Check string_to_quoted/2 for options information.

Link to this function

unrequire_files(files)

View Source (since 1.7.0)
@spec unrequire_files([binary()]) :: :ok

Removes files from the required files list.

The modules defined in the file are not removed; calling this function only removes them from the list, allowing them to be required again.

The list of files is managed per Erlang VM node.

Examples

# Require EEx test code
Code.require_file("../eex/test/eex_test.exs")

# Now unrequire all files
Code.unrequire_files(Code.required_files())

# Note that modules are still available
function_exported?(EExTest.Compiled, :before_compile, 0)
#=> true
Link to this function

with_diagnostics(opts \\ [], fun)

View Source (since 1.15.0)
@spec with_diagnostics(
  keyword(),
  (-> result)
) :: {result, [diagnostic(:warning | :error)]}
when result: term()

Executes the given fun and capture all diagnostics.

Diagnostics are warnings and errors emitted during code evaluation or single-file compilation and by functions such as IO.warn/2.

If using mix compile or Kernel.ParallelCompiler, note they already capture and return diagnostics.

Options

  • :log - if the diagnostics should be logged as they happen. Defaults to false.

Rescuing errors

with_diagnostics/2 does not automatically handle exceptions. You may capture them by adding a try/1 in fun:

{result, all_errors_and_warnings} =
  Code.with_diagnostics(fn ->
    try do
      {:ok, Code.compile_quoted(quoted)}
    rescue
      err -> {:error, err}
    end
  end)