View Source Sourceror (Sourceror v1.6.0)

Utilities to work with Elixir source code.

Installation

Add :sourceror as a dependency to your project's mix.exs:

defp deps do
  [
    {:sourceror, "~> 1.6"}
  ]
end

A note on compatibility

Sourceror is compatible with Elixir versions down to 1.10 and OTP 21. For Elixir versions prior to 1.13 it uses a vendored version of the Elixir parser and formatter modules. This means that for Elixir versions prior to 1.12 it will successfully parse the new syntax for stepped ranges instead of raising a SyntaxError, but everything else should work as expected.

Goals of the library

  • Be as close as possible to the standard Elixir AST.
  • Make working with comments as simple as possible.
  • No dev/prod dependencies, to simplify integration with other tools.

Sourceror's AST

Having the AST and comments as separate entities allows Elixir to expose the code formatting utilities without making any changes to it's AST, but also delegates the task of figuring out what's the most appropriate way to work with them to us.

Sourceror's take is to use the node metadata to store the comments. This allows us to work with an AST that is as close to regular elixir AST as possible. It also allows you to move nodes around without worrying about leaving a comment behind and ending up with misplaced comments.

Two metadata fields are added to the regular Elixir AST:

  • :leading_comments - holds the comments directly above the node or are in the same line as it. For example:

    test "parses leading comments" do
      quoted = """
      # Comment for :a
      :a # Also a comment for :a
      """ |> Sourceror.parse_string!()
    
      assert {:__block__, meta, [:a]} = quoted
      assert meta[:leading_comments] == [
        %{line: 1, column: 1, previous_eol_count: 1, next_eol_count: 1, text: "# Comment for :a"},
        %{line: 2, column: 4, previous_eol_count: 0, next_eol_count: 1, text: "# Also a comment for :a"},
      ]
    end
  • :trailing_comments - holds the comments that are inside of the node, but aren't leading any children, for example:

    test "parses trailing comments" do
      quoted = """
      def foo() do
      :ok
      # A trailing comment
      end # Not a trailing comment for :foo
      """ |> Sourceror.parse_string!()
    
      assert {:__block__, block_meta, [{:def, meta, _}]} = quoted
      assert [%{line: 3, text: "# A trailing comment"}] = meta[:trailing_comments]
      assert [%{line: 4, text: "# Not a trailing comment for :foo"}] = block_meta[:trailing_comments]
    end

Note that Sourceror considers leading comments to the ones that are found in the same line as a node, and trailing comments to the ones that are found before the ending line of a node, based on the end, closing or end_of_expression line. This also makes the Sourceror AST consistent with the way the Elixir formatter works, making it easier to reason about how a given AST would be formatted.

Traversing the AST

Elixir provides the Macro.prewalk, Macro.postwalk and Macro.traverse functions to traverse the AST. You can use the same functions to traverse the Sourceror AST as well, since it has the same shape as the standard Elixir AST.

Sourceror also provides the Sourceror.prewalk, Sourceror.postwalk and Sourceror.traverse variants. At the time of writing they are mostly wrappers around the standard Elixir functions for AST traversal, but they may be enhanced in the future if more AST formats are introduced.

In addition to these, Sourceror also provides a Zipper implementation for the Elixir AST. You can learn more about it in the Zippers notebook.

Patching the source code

You can use Sourceror to manipulate the AST and turn it back into human readable Elixir code, this is commonly known as writing a "codemod". For example, you can write a codemod to replace calls to String.to_atom to String.to_existing_atom:

test "updates the source code" do
  source =
    """
    String.to_atom(foo)\
    """

  new_source =
    source
    |> Sourceror.parse_string!()
    |> Macro.postwalk(fn
      {{:., dot_meta, [{:__aliases__, alias_meta, [:String]}, :to_atom]}, call_meta, args} ->
        {{:., dot_meta, [{:__aliases__, alias_meta, [:String]}, :to_existing_atom]}, call_meta, args}

      quoted ->
        quoted
    end)
    |> Sourceror.to_string()

  assert new_source ==
    """
    String.to_existing_atom(foo)\
    """
end

However, this will affect the whole source code, as we are working on the full source AST. Sourceror relies on the Elixir formatter to produce human readable code, so the original code formatting will be lost by using Sourceror.to_string. If your code is already using the Elixir formatter then this won't be an issue, but it will be an undesirable effect if you're not using it.

An alternative to this is to use Patches instead. A patch is a data structure that specifies the text range that should be replaced, and either a replacement string or a function that takes the text in that range and produces a string replacement.

Using patches, we could do the same as above, but produce a patch instead of modifying the AST. As a result, only the parts that need to be changed will be affected, and the rest of the code keeps the original formatting:

test "patches the source code" do
  source =
    """
    case foo do
      nil ->         :bar
      _ ->

          String.to_atom(foo)

          end\
    """

  {_quoted, patches} =
    source
    |> Sourceror.parse_string!()
    |> Macro.postwalk([], fn
      {{:., dot_meta, [{:__aliases__, alias_meta, [:String]}, :to_atom]}, call_meta, args} = quoted, patches ->
        range = Sourceror.get_range(quoted)
        replacement =
          {{:., dot_meta, [{:__aliases__, alias_meta, [:String]}, :to_existing_atom]}, call_meta, args}
          |> Sourceror.to_string()

        patch = %{range: range, change: replacement}
        {quoted, [patch | patches]}

      quoted, patches ->
        {quoted, patches}
    end)

  assert Sourceror.patch_string(source, patches) ==
    """
    case foo do
      nil ->         :bar
      _ ->

          String.to_existing_atom(foo)

          end\
    """
end

You have to keep in mind that:

  1. If you patch a node that has inner code, like replacing a full case, then the contents of the node will be reformatted as well.

  2. At the moment, Sourceror won't check for conflicts in the patches ranges, so care needs to be taken to not produce conflicting patches. You may need to do a number of parse -> patch -> reparse if you find yourself generating conflicting patches.

Some of the most common patching operations are available in the Sourceror.Patch module

Background

There have been several attempts at source code manipulation in the Elixir community. Thanks to its metaprogramming features, Elixir provides builtin tools that let us get the AST of any Elixir code, but when it comes to turning the AST back to code as text, we had limited options. Macro.to_string/2 is a thing, but the produced code is generally ugly, mostly because of the extra parenthesis or because it turns string interpolations into calls to erlang modules, to name some examples. This meant that, even if we could use Macro.to_string/2 to get a string and then give that to the Elixir formatter Code.format_string!/2, the output would still be suboptimal, as the formatter is not designed to change the semantics of the code, only to pretty print it. For example, call to erlang modules would be kept as is instead of being turned back to interpolations.

We also had the additional problem of comments being discarded by the tokenizer, and literals not having information like line numbers or delimiter characters. This makes the regular AST too lossy to be useful if what we want is to manipulate the source code, because we need as much information as possible to be able to stay as close to the source as possible. There have been several proposal in the past to bring all this information to the Elixir AST, but they all meant a change that would either break macros due to the addition of new types of AST nodes, or making a compromise in core Elixir itself by storing comments in the nods metadata. This discussion in the Elixir mailing list highlights the various issues faced when deciding if and how the comments would be preserved. Arjan Scherpenisse also did a talk where he discusses about the problems of using the standard Elixir AST to build refactoring tools.

Despite of all these issues, the Elixir formatter is still capable of manipulating the source code to pretty print it. Under the hood it does some neat tricks to have all this information available: on one hand, it tells the tokenizer to extract the comments from the source code and keep it at hand(not in the AST itself, but as a separate data structure), and on the other hand it tells the parser to wrap literals in block nodes so metadata can be preserved. Once it has all it needs, it can start converting the AST and comments into an algebra document, and ultimately convert that to a string. This functionality was private, and if we wanted to do it ourselves we would have to replicate or vendor the Elixir formatter with all its more than 2000 lines of code. This approach was explored by Wojtek Mach in wojtekmach/fix, but it involved vendoring the elixir Formatter code, was tightly coupled to the formatting process, and any change in Elixir would break the code.

Since Elixir 1.13 this functionality from the formatter was finally exposed via the Code.string_to_quoted_with_comments/2 and Code.quoted_to_algebra/2 functions. The former gives us access to the list of comments in a shape the Elixir formatter is able to use, and the latter lets us turn any arbitrary Elixir AST into an algebra document. If we also give it the list of comments, it will merge them together, allowing us to format AST and preserve the comments. Now all we need to care about is of manipulating the AST, and let the formatter do the rest.

Summary

Functions

Appends comments to the leading or trailing comments of a node.

Compares two positions.

Shifts the line numbers of the node or metadata by the given line_correction.

Returns the arguments of the node.

Returns the column of a node. If none is found, the default value is returned(defaults to 1).

Returns the line where the given node ends. It recursively checks for end, closing and end_of_expression line numbers. If none is found, the default value is returned(defaults to 1).

Returns the end position of the quoted expression. It recursively checks for end, closing and end_of_expression positions. If none is found, the default value is returned(defaults to [line: 1, column: 1]).

Returns the line of a node. If none is found, the default value is returned(defaults to 1).

Returns the metadata of the given node.

Gets the range used by the given quoted expression in the source code.

Returns the start metadata position of a node.

Parses a single expression from the given string. It tries to parse on a per-line basis.

Parses the source code into an extended AST suitable for source manipulation as described in Code.quoted_to_algebra/2.

Same as parse_string/1 but raises on error.

Applies one or more patches to the given string.

Performs a depth-first post-order traversal of a quoted expression.

Performs a depth-first post-order traversal of a quoted expression with an accumulator.

Prepends comments to the leading or trailing comments of a node.

Performs a depth-first pre-order traversal of a quoted expression.

Performs a depth-first pre-order traversal of a quoted expression with an accumulator.

A wrapper around Code.quoted_to_algebra/2 for compatibility with pre 1.13 Elixir versions.

A wrapper around Code.string_to_quoted_with_comments/2 for compatibility with pre 1.13 Elixir versions.

A wrapper around Code.string_to_quoted_with_comments!/2 for compatibility with pre 1.13 Elixir versions.

A convenience utility function for stripping all meta data out of a quoted expression. This is useful when debugging for readability of very large ASTs.

Converts a quoted expression to a string.

Updates the arguments for the given node.

Types

@type comment() :: %{
  line: integer(),
  previous_eol_count: integer(),
  next_eol_count: integer(),
  text: String.t()
}
@type position() :: keyword()
@type traversal_function() ::
  (Macro.t(), Sourceror.TraversalState.t() ->
     {Macro.t(), Sourceror.TraversalState.t()})

Functions

Link to this function

append_comments(quoted, comments, position \\ :leading)

View Source
@spec append_comments(
  quoted :: Macro.t(),
  comments :: [comment()],
  position :: :leading | :trailing
) :: Macro.t()

Appends comments to the leading or trailing comments of a node.

Link to this function

compare_positions(left, right)

View Source
@spec compare_positions(position(), position()) :: :gt | :eq | :lt

Compares two positions.

Returns :gt if the first position comes after the second one, and :lt for vice versa. If the two positions are equal, :eq is returned.

nil values for lines or columns are coalesced to 0 for integer comparisons.

Link to this function

correct_lines(meta, line_correction, opts \\ [])

View Source
@spec correct_lines(Macro.t() | Macro.metadata(), integer(), Macro.metadata()) ::
  Macro.t() | Macro.metadata()

Shifts the line numbers of the node or metadata by the given line_correction.

This function will update the :line, :closing, :do, :end and :end_of_expression line numbers of the node metadata if such fields are present.

@spec get_args(Macro.t()) :: [Macro.t()]

Returns the arguments of the node.

iex> Sourceror.get_args({:foo, [], [{:__block__, [], [:ok]}]})
[{:__block__, [], [:ok]}]
Link to this function

get_column(arg, default \\ 1)

View Source
@spec get_column(Macro.t(), default :: integer() | nil) :: integer() | nil

Returns the column of a node. If none is found, the default value is returned(defaults to 1).

A default of nil may also be provided if the column number is meant to be coalesced with a value that is not known upfront.

iex> Sourceror.get_column({:foo, [column: 5], []})
5

iex> Sourceror.get_column({:foo, [], []}, 3)
3
Link to this function

get_end_line(quoted, default \\ 1)

View Source
@spec get_end_line(Macro.t(), integer()) :: integer()

Returns the line where the given node ends. It recursively checks for end, closing and end_of_expression line numbers. If none is found, the default value is returned(defaults to 1).

iex> Sourceror.get_end_line({:foo, [end: [line: 4]], []})
4

iex> Sourceror.get_end_line({:foo, [closing: [line: 2]], []})
2

iex> Sourceror.get_end_line({:foo, [end_of_expression: [line: 5]], []})
5

iex> Sourceror.get_end_line({:foo, [closing: [line: 2], end: [line: 4]], []})
4

iex> """
...> alias Foo.{
...>   Bar
...> }
...> """ |> Sourceror.parse_string!() |> Sourceror.get_end_line()
3
Link to this function

get_end_position(quoted, default \\ [line: 1, column: 1])

View Source
@spec get_end_position(Macro.t(), position()) :: position()

Returns the end position of the quoted expression. It recursively checks for end, closing and end_of_expression positions. If none is found, the default value is returned(defaults to [line: 1, column: 1]).

Note that this position is the literal contents of the metadata found in the node. For accurate end positions, use get_range/1.

iex> quoted = ~S"""
...> A.{
...>   B
...> }
...> """ |>  Sourceror.parse_string!()
iex> Sourceror.get_end_position(quoted)
[line: 3, column: 1]

iex> quoted = ~S"""
...> foo do
...>   :ok
...> end
...> """ |>  Sourceror.parse_string!()
iex> Sourceror.get_end_position(quoted)
[line: 3, column: 1]

iex> quoted = ~S"""
...> foo(
...>   :a,
...>   :b
...>    )
...> """ |>  Sourceror.parse_string!()
iex> Sourceror.get_end_position(quoted)
[line: 4, column: 4]
Link to this function

get_line(arg, default \\ 1)

View Source
@spec get_line(Macro.t(), default :: integer() | nil) :: integer() | nil

Returns the line of a node. If none is found, the default value is returned(defaults to 1).

A default of nil may also be provided if the line number is meant to be coalesced with a value that is not known upfront.

iex> Sourceror.get_line({:foo, [line: 5], []})
5

iex> Sourceror.get_line({:foo, [], []}, 3)
3
@spec get_meta(Macro.t()) :: Macro.metadata()

Returns the metadata of the given node.

iex> Sourceror.get_meta({:foo, [line: 5], []})
[line: 5]
Link to this function

get_range(quoted, opts \\ [])

View Source

Gets the range used by the given quoted expression in the source code.

The quoted expression must have at least line and column metadata, otherwise it is not possible to calculate an accurate range, or to calculate it at all. Additionally, certain syntax constructs desugar into ASTs without a meaningful range. In these cases, get_range/1 returns nil.

This function is most useful when used after Sourceror.parse_string/1, before any kind of modification to the AST.

iex> quoted = ~S"""
...> def foo do
...>   :ok
...> end
...> """ |> Sourceror.parse_string!()
iex> Sourceror.get_range(quoted)
%Sourceror.Range{start: [line: 1, column: 1], end: [line: 3, column: 4]}

iex> quoted = ~S"""
...> Foo.{
...>   Bar
...> }
...> """ |> Sourceror.parse_string!()
iex> Sourceror.get_range(quoted)
%Sourceror.Range{start: [line: 1, column: 1], end: [line: 3, column: 2]}

Options

  • :include_comments - When true, it includes the comments into the range. Defaults to false.
iex> ~S"""
...> # Foo
...> :baz # Bar
...> """
...> |> Sourceror.parse_string!()
...> |> Sourceror.get_range(include_comments: true)
%Sourceror.Range{start: [line: 1, column: 1], end: [line: 2, column: 11]}
Link to this function

get_start_position(quoted, default \\ [line: 1, column: 1])

View Source
@spec get_start_position(Macro.t(), position()) :: position()

Returns the start metadata position of a node.

Note that this position is the literal contents of the metadata found in the node. For accurate start positions, use get_range/1.

iex> quoted = Sourceror.parse_string!(" :foo")
iex> Sourceror.get_start_position(quoted)
[line: 1, column: 2]

iex> quoted = Sourceror.parse_string!("\n\nfoo()")
iex> Sourceror.get_start_position(quoted)
[line: 3, column: 1]

iex> quoted = Sourceror.parse_string!("Foo.{Bar}")
iex> Sourceror.get_start_position(quoted)
[line: 1, column: 1]

iex> quoted = Sourceror.parse_string!("foo[:bar]")
iex> Sourceror.get_start_position(quoted)
[line: 1, column: 1]

iex> quoted = Sourceror.parse_string!("foo(:bar)")
iex> Sourceror.get_start_position(quoted)
[line: 1, column: 1]
Link to this function

parse_expression(string, opts \\ [])

View Source
@spec parse_expression(
  String.t(),
  keyword()
) :: {:ok, Macro.t(), String.t()} | {:error, String.t()}

Parses a single expression from the given string. It tries to parse on a per-line basis.

Returns {:ok, quoted, rest} on success or {:error, source} on error.

Examples

iex> ~S"""
...> 42
...>
...> :ok
...> """ |> Sourceror.parse_expression()
{:ok, {:__block__, [trailing_comments: [], leading_comments: [],
                    token: "42", line: 2, column: 1], [42]}, "\n:ok"}

Options

  • :from_line - The line at where the parsing should start. Defaults to 1.
@spec parse_string(String.t()) :: {:ok, Macro.t()} | {:error, term()}

Parses the source code into an extended AST suitable for source manipulation as described in Code.quoted_to_algebra/2.

Two additional fields are added to nodes metadata:

  • :leading_comments - a list holding the comments found before the node.
  • :trailing_comments - a list holding the comments found before the end of the node. For example, comments right before the end keyword.

Comments are the same maps returned by Code.string_to_quoted_with_comments/2.

@spec parse_string!(String.t()) :: Macro.t()

Same as parse_string/1 but raises on error.

Link to this function

patch_string(string, patches)

View Source
@spec patch_string(String.t(), [Sourceror.Patch.t()]) :: String.t()

Applies one or more patches to the given string.

A patch is a map containing the following values:

  • :range - a map containing :start and :end keys, whose values are keyword lists containing the :line and :column representing the boundary of the patch.
  • :change - the string being patched in or a function that takes the text of the patch range and returns the replacement.
  • :preserve_indentation (default true) - whether to automatically correct the indentation of the patch string to preserve the indentation level of the patch range (see examples below)

Note that :line and :column start at 1 and represent a cursor positioned before the targeted position. For instance, here's how you would select the string "ToBePatched" in the following example:

defmodule ToBePatched do
#         ^          ^
# col     11         22

%{range: %{start: [line: 1, column: 11], end: [line: 1, column: 22]}}

Ranges are usually derived from parsed AST nodes. See get_range/2 for more.

Patches are applied bottom-up, such that patches to the beginning of the string do not interfere with the line/column of patches that come later. However, no checks are done for overlapping ranges, so take care to pass in non-overlapping patches.

Examples

iex> original = ~S"""
...> if not allowed? do
...>   raise "Not allowed!"
...> end
...> """
iex> patch = %Sourceror.Patch{
...>   change: "unless allowed? do\n  raise \"Not allowed!\"\nend",
...>   range: %Sourceror.Range{start: [line: 1, column: 1], end: [line: 3, column: 4]}
...> }
iex> Sourceror.patch_string(original, [patch])
~S"""
unless allowed? do
  raise "Not allowed!"
end
"""

A range can also be a function, in which case the original text in the patch range will be given as an argument:

iex> original = ~S"""
...> hello :world
...> """
iex> patch = %Sourceror.Patch{
...>   change: &String.upcase/1,
...>   range: %Sourceror.Range{start: [line: 1, column: 7], end: [line: 1, column: 13]}
...> }
iex> Sourceror.patch_string(original, [patch])
~S"""
hello :WORLD
"""

By default, lines after the first line of the patch will be indented relative to the indentation level at the start of the range:

iex> original = ~S"""
...> outer do
...>   inner(foo do
...>     :original
...>   end)
...> end
...> """
iex> patch = %Sourceror.Patch{
...>   change: "bar do\n  :replacement\nend",
...>   range: %Sourceror.Range{start: [line: 2, column: 9], end: [line: 4, column: 6]}
...> }
iex> Sourceror.patch_string(original, [patch])
~S"""
outer do
  inner(bar do
    :replacement
  end)
end
"""

If you don't want this behavior, you can add :preserve_indentation: false to your patch:

iex> original = ~S"""
...> outer do
...>   inner(foo do
...>     :original
...>   end)
...> end
...> """
iex> patch = %Sourceror.Patch{
...>   change: "bar do\n  :replacement\nend",
...>   range: %Sourceror.Range{start: [line: 2, column: 9], end: [line: 4, column: 6]},
...>   preserve_indentation: false
...> }
iex> Sourceror.patch_string(original, [patch])
~S"""
outer do
  inner(bar do
  :replacement
end)
end
"""
@spec postwalk(Macro.t(), traversal_function()) :: Macro.t()

Performs a depth-first post-order traversal of a quoted expression.

See postwalk/3 for more information.

Link to this function

postwalk(quoted, acc, fun)

View Source
@spec postwalk(Macro.t(), term(), traversal_function()) :: {Macro.t(), term()}

Performs a depth-first post-order traversal of a quoted expression with an accumulator.

fun is a function that will receive the current node as a first argument and the traversal state as the second one. It must return a {quoted, state}, in the same way it would return {quoted, acc} when using Macro.postwalk/3.

The state is a map with the following keys:

  • :acc - The accumulator. Defaults to nil if none is given.
Link to this function

prepend_comments(quoted, comments, position \\ :leading)

View Source
@spec prepend_comments(
  quoted :: Macro.t(),
  comments :: [comment()],
  position :: :leading | :trailing
) :: Macro.t()

Prepends comments to the leading or trailing comments of a node.

@spec prewalk(Macro.t(), traversal_function()) :: Macro.t()

Performs a depth-first pre-order traversal of a quoted expression.

See prewalk/3 for more information.

Link to this function

prewalk(quoted, acc, fun)

View Source
@spec prewalk(Macro.t(), term(), traversal_function()) :: {Macro.t(), term()}

Performs a depth-first pre-order traversal of a quoted expression with an accumulator.

fun is a function that will receive the current node as a first argument and the traversal state as the second one. It must return a {quoted, state}, in the same way it would return {quoted, acc} when using Macro.prewalk/3.

The state is a map with the following keys:

  • :acc - The accumulator. Defaults to nil if none is given.
Link to this macro

quoted_to_algebra(quoted, opts)

View Source (macro)

A wrapper around Code.quoted_to_algebra/2 for compatibility with pre 1.13 Elixir versions.

Link to this macro

string_to_quoted(string, opts)

View Source (macro)

A wrapper around Code.string_to_quoted_with_comments/2 for compatibility with pre 1.13 Elixir versions.

Link to this macro

string_to_quoted!(string, opts)

View Source (macro)

A wrapper around Code.string_to_quoted_with_comments!/2 for compatibility with pre 1.13 Elixir versions.

@spec strip_meta(quoted :: Macro.t()) :: Macro.t()

A convenience utility function for stripping all meta data out of a quoted expression. This is useful when debugging for readability of very large ASTs.

Link to this function

to_string(quoted, opts \\ [])

View Source
@spec to_string(
  Macro.t(),
  keyword()
) :: String.t()

Converts a quoted expression to a string.

The comments line number will be ignored and the line number of the associated node will be used when formatting the code.

If you're using Elixir >=1.13, the locals_without_parens from your project formatter configuration will be used. You can pass a different set of options by using the locals_without_parens option. If you want to disable that option entirely, use locals_without_parens: [].

Options

  • :indent - how many indentations to insert at the start of each line. Note that this only prepends the indents without checking the indentation of nested blocks. Defaults to 0.

  • :indent_type - the type of indentation to use. It can be one of :spaces, :single_space or :tabs. Defaults to :spaces.

  • :format - if set to :splicing, if the quoted expression is a list, it will strip the square brackets. This is useful to print a single element of a keyword list.

  • :quoted_to_algebra - expects a function of the type (Macro.t(), keyword -> Inspect.Algebra.t()) to convert the given quoted expression to an algebra document.

For more options see Code.format_string!/1 and Code.quoted_to_algebra/2.

@spec update_args(Macro.t(), ([Macro.t()] -> [Macro.t()])) :: Macro.t()

Updates the arguments for the given node.

iex> node = {:foo, [line: 1], [{:__block__, [line: 1], [2]}]}
iex> updater = fn args -> Enum.map(args, &Sourceror.correct_lines(&1, 2)) end
iex> Sourceror.update_args(node, updater)
{:foo, [line: 1], [{:__block__, [line: 3], [2]}]}