Earmark—A Pure Elixir Markdown Processor

N.B.

This README contains the docstrings and doctests from the code by means of extractly and the following code examples are therefore verified with ExUnit doctests.

Table Of Content

Table Of Content
Options
Contributing
Author

Options

Earmark.Cli.Implementation

Functional (with the exception of reading input files with Earmark.File) interface to the CLI returning the device and the string to be output.

Earmark.Options

This is a superset of the options that need to be passed into Earmark.Parser.as_ast/2

The following options are proper to Earmark only and therefore explained in detail

compact_output: boolean indicating to avoid indentation and minimize whitespace
eex: Allows usage of an EEx template to be expanded to markdown before conversion
file: Name of file passed in from the CLI
line: 1 but might be set to an offset for better error messages in some integration cases
smartypants: boolean use Smarty Pants in the output
ignore_strings, postprocessor and registered_processors: processors that modify the AST returned from

Earmark.Parser.as_ast/2 before rendering (post because preprocessing is done on the markdown, e.g. eex) Refer to the moduledoc of Earmark.Transform for details

All other options are passed onto Earmark.Parser.as_ast/2

Earmark.Options.make_options/1

Make a legal and normalized Option struct from, maps or keyword lists

Without a param or an empty input we just get a new Option struct

iex(1)> { make_options(), make_options(%{}) } { {:ok, %Earmark.Options{}}, {:ok, %Earmark.Options{}} }

The same holds for the bang version of course

iex(2)> { make_options!(), make_options!(%{}) } { %Earmark.Options{}, %Earmark.Options{} }

We check for unallowed keys

iex(3)> make_options(no_such_option: true) {:error, [{:warning, 0, "Unrecognized option no_such_option: true"}]}

Of course we do not let our users discover one error after another

iex(4)> make_options(no_such_option: true, gfm: false, still_not_an_option: 42) {:error, [{:warning, 0, "Unrecognized option no_such_option: true"}, {:warning, 0, "Unrecognized option still_not_an_option: 42"}]}

And the bang version will raise an Earmark.Error as excepted (sic)

iex(5)> make_options!(no_such_option: true, gfm: false, still_not_an_option: 42) ** (Earmark.Error) [{:warning, 0, "Unrecognized option no_such_option: true"}, {:warning, 0, "Unrecognized option still_not_an_option: 42"}]

Some values need to be numeric

iex(6)> make_options(line: "42") {:error, [{:error, 0, "line option must be numeric"}]}

iex(7)> make_options(%Earmark.Options{footnote_offset: "42"}) {:error, [{:error, 0, "footnote_offset option must be numeric"}]}

iex(8)> make_options(%{line: "42", footnote_offset: nil}) {:error, [{:error, 0, "footnote_offset option must be numeric"}, {:error, 0, "line option must be numeric"}]}

Earmark.Options.relative_filename/2

Allows to compute the path of a relative file name (starting with "./") from the file in options and return an updated options struct

iex(9)> options = %Earmark.Options{file: "some/path/xxx.md"} ...(9)> options_ = relative_filename(options, "./local.md") ...(9)> options_.file "some/path/local.md"

For your convenience you can just use a keyword list

iex(10)> options = relative_filename([file: "some/path/_.md", breaks: true], "./local.md") ...(10)> {options.file, options.breaks} {"some/path/local.md", true}

If the filename is not absolute it just replaces the file in options

iex(11)> options = %Earmark.Options{file: "some/path/xxx.md"} ...(11)> options_ = relative_filename(options, "local.md") ...(11)> options_.file "local.md"

And there is a special case when processing stdin, meaning that file: nil we replace file verbatim in that case

iex(12)> options = %Earmark.Options{} ...(12)> options_ = relative_filename(options, "./local.md") ...(12)> options_.file "./local.md"

Earmark.Options.with_postprocessor/2

A convenience constructor

Earmark.Internal

All public functions that are internal to Earmark, so that only external API functions are public in Earmark

Earmark.Internal.as_ast!/2

A wrapper to extract the AST from a call to Earmark.Parser.as_ast if a tuple {:ok, result, []} is returned, raise errors otherwise

    iex(1)> as_ast!(["Hello %% annotated"], annotations: "%%")
    [{"p", [], ["Hello "], %{annotation: "%% annotated"}}]

    iex(2)> as_ast!("===")
    ** (Earmark.Error) [{:warning, 1, "Unexpected line ==="}]

Earmark.Internal.from_file!/2

This is a convenience method to read a file or pass it to EEx.eval_file if its name ends in .eex

The returned string is then passed to as_html this is used in the escript now and allows for a simple inclusion mechanism, as a matter of fact an include function is passed

Earmark.Internal.include/2

A utility function that will be passed as a partial capture to EEx.eval_file by providing a value for the options parameter

    EEx.eval(..., include: &include(&1, options))

thusly allowing

  <%= include.(some file) %>

where some file can be a relative path starting with "./"

Here is an example using these fixtures

    iex(3)> include("./include/basic.md.eex", file: "test/fixtures/does_not_matter")
    "# Headline Level 1\n"

And here is how it is used inside a template

    iex(4)> options = [file: "test/fixtures/does_not_matter"]
    ...(4)> EEx.eval_string(~s{<%= include.("./include/basic.md.eex") %>}, include: &include(&1, options))
    "# Headline Level 1\n"

Earmark.Transform

Structure Conserving Transformers

For the convenience of processing the output of Earmark.Parser.as_ast we expose two structure conserving mappers.

`map_ast`

Traverses an AST using a mapper function.

The mapper function will be called for each node including text elements unless map_ast is called with the third positional parameter ignore_strings, which is optional and defaults to false, set to true.

Depending on the return value of the mapper function the traversal will either

{new_tag, new_atts, ignored, new_meta}

just replace the tag, attribute and meta values of the current node with the values of the returned quadruple (ignoring ignored for facilitating nodes w/o transformation) and then descend into the original content of the node.
{:replace, node}

replaces the current node with node and does not descend anymore, but continues traversal on sibblings.
{new_function, {new_tag, new_atts, ignored, new_meta}}

just replace the tag, attribute and meta values of the current node with the values of the returned quadruple (ignoring ignored for facilitating nodes w/o transformation) and then descend into the original content of the node but with the mapper function new_function used for transformation of the AST.

N.B. The original mapper function will be used for transforming the sibbling nodes though.

takes a function that will be called for each node of the AST, where a leaf node is either a quadruple like {"code", [{"class", "inline"}], ["some code"], %{}} or a text leaf like "some code"

The result of the function call must be

for nodes → as described above
for strings → strings or nodes

As an example let us transform an ast to have symbol keys

      iex(1)> input = [
      ...(1)> {"h1", [], ["Hello"], %{title: true}},
      ...(1)> {"ul", [], [{"li", [], ["alpha"], %{}}, {"li", [], ["beta"], %{}}], %{}}]
      ...(1)> map_ast(input, fn {t, a, _, m} -> {String.to_atom(t), a, nil, m} end, true)
      [ {:h1, [], ["Hello"], %{title: true}},
        {:ul, [], [{:li, [], ["alpha"], %{}}, {:li, [], ["beta"], %{}}], %{}} ]

N.B. If this returning convention is not respected map_ast might not complain, but the resulting transformation might not be suitable for Earmark.Transform.transform anymore. From this follows that any function passed in as value of the postprocessor: option must obey to these conventions.

`map_ast_with`

this is like map_ast but like a reducer an accumulator can also be passed through.

For that reason the function is called with two arguments, the first element being the same value as in map_ast and the second the accumulator. The return values need to be equally augmented tuples.

A simple example, annotating traversal order in the meta map's :count key, as we are not interested in text nodes we use the fourth parameter ignore_strings which defaults to false

       iex(2)>  input = [
       ...(2)>  {"ul", [], [{"li", [], ["one"], %{}}, {"li", [], ["two"], %{}}], %{}},
       ...(2)>  {"p", [], ["hello"], %{}}]
       ...(2)>  counter = fn {t, a, _, m}, c -> {{t, a, nil, Map.put(m, :count, c)}, c+1} end
       ...(2)>  map_ast_with(input, 0, counter, true)
       {[ {"ul", [], [{"li", [], ["one"], %{count: 1}}, {"li", [], ["two"], %{count: 2}}], %{count: 0}},
         {"p", [], ["hello"], %{count: 3}}], 4}

Let us describe an implementation of a real world use case taken from Elixir Forum

Simplifying the exact parsing of the text node in this example we only want to replace a text node of the form #elixir with a link to the Elixir home page but only when inside a {"p",....} node

We can achieve this as follows

      iex(3)> elixir_home = {"a", [{"href", "https://elixir-lang.org"}], ["Elixir"], %{}}
      ...(3)> transformer = fn {"p", atts, _, meta}, _ -> {{"p", atts, nil, meta}, true}
      ...(3)>                  "#elixir", true -> {elixir_home, false}
      ...(3)>                  text, _ when is_binary(text) -> {text, false}
      ...(3)>                  node, _ ->  {node, false} end
      ...(3)> ast = [
      ...(3)>  {"p", [],[ "#elixir"], %{}}, {"bold", [],[ "#elixir"], %{}},
      ...(3)>  {"ol", [], [{"li", [],[ "#elixir"], %{}}, {"p", [],[ "elixir"], %{}}, {"p", [], ["#elixir"], %{}}], %{}}
      ...(3)> ]
      ...(3)> map_ast_with(ast, false, transformer)
      {[
       {"p", [],[{"a", [{"href", "https://elixir-lang.org"}], ["Elixir"], %{}}], %{}}, {"bold", [],[ "#elixir"], %{}},
       {"ol", [], [{"li", [],[ "#elixir"], %{}}, {"p", [],[ "elixir"], %{}}, {"p", [], [{"a", [{"href", "https://elixir-lang.org"}], ["Elixir"], %{}}], %{}}], %{}}
      ], false}

An alternate, maybe more elegant solution would be to change the mapper function during AST traversal as demonstrated here

Postprocessors and Convenience Functions

These can be declared in the fields postprocessor and registered_processors in the Options struct, postprocessor is prepened to registered_processors and they are all applied to non string nodes (that is the quadtuples of the AST which are of the form {tag, atts, content, meta}

All postprocessors can just be functions on nodes or a TagSpecificProcessors struct which will group function applications depending on tags, as a convienience tuples of the form {tag, function} will be transformed into a TagSpecificProcessors struct.

    iex(4)> add_class1 = &Earmark.AstTools.merge_atts_in_node(&1, class: "class1")
    ...(4)> m1 = Earmark.Options.make_options!(postprocessor: add_class1) |> make_postprocessor()
    ...(4)> m1.({"a", [], nil, nil})
    {"a", [{"class", "class1"}], nil, nil}

We can also use the registered_processors field:

    iex(5)> add_class1 = &Earmark.AstTools.merge_atts_in_node(&1, class: "class1")
    ...(5)> m2 = Earmark.Options.make_options!(registered_processors: add_class1) |> make_postprocessor()
    ...(5)> m2.({"a", [], nil, nil})
    {"a", [{"class", "class1"}], nil, nil}

Knowing that values on the same attributes are added onto the front the following doctest demonstrates the order in which the processors are executed

    iex(6)> add_class1 = &Earmark.AstTools.merge_atts_in_node(&1, class: "class1")
    ...(6)> add_class2 = &Earmark.AstTools.merge_atts_in_node(&1, class: "class2")
    ...(6)> add_class3 = &Earmark.AstTools.merge_atts_in_node(&1, class: "class3")
    ...(6)> m = Earmark.Options.make_options!(postprocessor: add_class1, registered_processors: [add_class2, {"a", add_class3}])
    ...(6)> |> make_postprocessor()
    ...(6)> [{"a", [{"class", "link"}], nil, nil}, {"b", [], nil, nil}]
    ...(6)> |> Enum.map(m)
    [{"a", [{"class", "class3 class2 class1 link"}], nil, nil}, {"b", [{"class", "class2 class1"}], nil, nil}]

We can see that the tuple form has been transformed into a tag specific transformation only as a matter of fact, the explicit definition would be:

    iex(7)> m = make_postprocessor(
    ...(7)>   %Earmark.Options{
    ...(7)>     registered_processors:
    ...(7)>       [Earmark.TagSpecificProcessors.new({"a", &Earmark.AstTools.merge_atts_in_node(&1, target: "_blank")})]})
    ...(7)> [{"a", [{"href", "url"}], nil, nil}, {"b", [], nil, nil}]
    ...(7)> |> Enum.map(m)
    [{"a", [{"href", "url"}, {"target", "_blank"}], nil, nil}, {"b", [], nil, nil}]

We can also define a tag specific transformer in one step, which might (or might not) solve potential performance issues when running too many processors

    iex(8)> add_class4 = &Earmark.AstTools.merge_atts_in_node(&1, class: "class4")
    ...(8)> add_class5 = &Earmark.AstTools.merge_atts_in_node(&1, class: "class5")
    ...(8)> add_class6 = &Earmark.AstTools.merge_atts_in_node(&1, class: "class6")
    ...(8)> tsp = Earmark.TagSpecificProcessors.new([{"a", add_class5}, {"b", add_class5}])
    ...(8)> m = Earmark.Options.make_options!(
    ...(8)>       postprocessor: add_class4,
    ...(8)>       registered_processors: [tsp, add_class6])
    ...(8)> |> make_postprocessor()
    ...(8)> [{"a", [], nil, nil}, {"c", [], nil, nil}, {"b", [], nil, nil}]
    ...(8)> |> Enum.map(m)
    [{"a", [{"class", "class6 class5 class4"}], nil, nil}, {"c", [{"class", "class6 class4"}], nil, nil}, {"b", [{"class", "class6 class5 class4"}], nil, nil}]

Of course the mechanics shown above is hidden if all we want is to trigger the postprocessor chain in Earmark.as_html, here goes a typical example

    iex(9)> add_target = fn node -> # This will only be applied to nodes as it will become a TagSpecificProcessors
    ...(9)>   if Regex.match?(~r{\.x\.com\z}, Earmark.AstTools.find_att_in_node(node, "href", "")), do:
    ...(9)>     Earmark.AstTools.merge_atts_in_node(node, target: "_blank"), else: node end
    ...(9)> options = [
    ...(9)> registered_processors: [{"a", add_target}, {"p", &Earmark.AstTools.merge_atts_in_node(&1, class: "example")}]]
    ...(9)> markdown = [
    ...(9)>   "http://hello.x.com",
    ...(9)>   "",
    ...(9)>   "[some](url)",
    ...(9)>  ]
    ...(9)> Earmark.as_html!(markdown, options)
    "<p class=\"example\">\n<a href=\"http://hello.x.com\" target=\"_blank\">http://hello.x.com</a></p>\n<p class=\"example\">\n<a href=\"url\">some</a></p>\n"

Use case: Modification of Link Attributes depending on the URL

This would be done as follows

        Earmark.as_html!(markdown, registered_processors: {"a", my_function_that_is_invoked_only_with_a_nodes})

Use case: Modification of the AST according to Annotations

N.B. Annotation are an experimental feature in 1.4.16-pre and are documented here

By annotating our markdown source we can then influence the rendering. In this example we will just add some decoration

    iex(10)> markdown = [ "A joke %% smile", "", "Charming %% in_love" ]
    ...(10)> add_smiley = fn {_, _, _, meta} = quad, _acc ->
    ...(10)>                case Map.get(meta, :annotation) do
    ...(10)>                  "%% smile"   -> {quad, "\u1F601"}
    ...(10)>                  "%% in_love" -> {quad, "\u1F60d"}
    ...(10)>                  _            -> {quad, nil}
    ...(10)>                end
    ...(10)>                text, nil -> {text, nil}
    ...(10)>                text, ann -> {"#{text} #{ann}", nil}
    ...(10)>              end
    ...(10)> Earmark.as_ast!(markdown, annotations: "%%") |> Earmark.Transform.map_ast_with(nil, add_smiley) |> Earmark.transform
    "<p>\nA joke  ὠ1</p>\n<p>\nCharming  ὠd</p>\n"

Structure Modifying Transformers

For structure modifications a tree traversal is needed and no clear pattern of how to assist this task with tools has emerged yet.

Earmark.Restructure.walk_and_modify_ast/4

Walks an AST and allows you to process it (storing details in acc) and/or modify it as it is walked.

items is the AST you got from Earmark.Parser.as_ast()

acc is the initial value of an accumulator that is passed to both process_item_fn and process_list_fn and accumulated. If your functions do not need to use or store any state, you can pass nil.

The process_item_fn function is required. It takes two parameters, the single item to process (which will either be a string or a 4-tuple) and the accumulator, and returns a tuple {processed_item, updated_acc}. Returning the empty list for processed_item will remove the item processed the AST.

The process_list_fn function is optional and defaults to no modification of items or accumulator. It takes two parameters, the list of items that are the sub-items of a given element in the AST (or the top-level list of items), and the accumulator, and returns a tuple {processed_items_list, updated_acc}.

This function ends up returning {ast, acc}.

Here is an example using a custom format to make <em> nodes and allowing commented text to be left out

    iex(1)> is_comment? = fn item -> is_binary(item) && Regex.match?(~r/\A\s*--/, item) end
    ...(1)> comment_remover =
    ...(1)>   fn items, acc -> {Enum.reject(items, is_comment?), acc} end
    ...(1)> italics_maker = fn
    ...(1)>   item, acc when is_binary(item) ->
    ...(1)>     new_item = Restructure.split_by_regex(
    ...(1)>       item,
    ...(1)>       ~r/\/([[:graph:]].*?[[:graph:]]|[[:graph:]])\//,
    ...(1)>       fn [_, content] ->
    ...(1)>         {"em", [], [content], %{}}
    ...(1)>       end
    ...(1)>     )
    ...(1)>     {new_item, acc}
    ...(1)>   item, "a" -> {item, nil}
    ...(1)>   {name, _, _, _}=item, _ -> {item, name}
    ...(1)> end
    ...(1)> markdown = """
    ...(1)> [no italics in links](http://example.io/some/path)
    ...(1)> but /here/
    ...(1)>
    ...(1)> -- ignore me
    ...(1)>
    ...(1)> text
    ...(1)> """
    ...(1)> {:ok, ast, []} = Earmark.Parser.as_ast(markdown)
    ...(1)> Restructure.walk_and_modify_ast(ast, nil, italics_maker, comment_remover)
    {[
      {"p", [],
        [
          {"a", [{"href", "http://example.io/some/path"}], ["no italics in links"],
          %{}},
          "\nbut ",
          {"em", [], ["here"], %{}},
          ""
        ], %{}},
        {"p", [], [], %{}},
        {"p", [], ["text"], %{}}
      ], "p"}

Earmark.Restructure.split_by_regex/3

Utility for creating a restructuring that parses text by splitting it into parts "of interest" vs. "other parts" using a regular expression. Returns a list of parts where the parts matching regex have been processed by invoking map_captures_fn on each part, and a list of remaining parts, preserving the order of parts from what it was in the plain text item.

      iex(2)> input = "This is ::all caps::, right?"
      ...(2)> split_by_regex(input, ~r/::(.*?)::/, fn [_, inner|_] -> String.upcase(inner) end)
      ["This is ", "ALL CAPS", ", right?"]

Contributing

Pull Requests are happily accepted.

Please be aware of one caveat when correcting/improving README.md.

The README.md is generated by Extractly as mentioned above and therefore contributors shall not modify it directly, but README.md.eex and the imported docs instead.

You need to run mix xtra after getting the dependencies to generate the README.md file. Thank you all who have already helped with Earmark, your names are duly noted in RELEASE.md.