Utilities to work with Elixir source code.

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

defp deps do
  [
    {:sourceror, "~> 1.0"}
  ]
end

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.

• 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.

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.

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.

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

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.

You can find Sourceror documentation on Hex Docs.

You can find usage examples in the examples folder. You can run them by cloning the repo and running elixir examples/<example_file>.exs.

You can also find documented examples you can run with Livebook in the notebooks folder.

If you want to contribute to Sourceror, please check our Contributing section for pointers.

If you have any questions about Sourceror or need assistance, please open a thread in the Discussions section.

Copyright (c) 2021 [email protected]

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.