markdown-it-pyrs

Currently in Beta, feedback welcome!

A Python interface for markdown-it.rs (and plugins), using Rust for blazingly fast Markdown parsing ⚡️

The goal of this project is to provide a fast, safe, extensible, and easy-to-use Markdown parser for Python. It is complimentary to markdown-it-py, which is a pure Python implementation of markdown-it, and here we aim to follow as close as possible the API for that package.

If you care primarily about speed, this is the library for you. For example, benchmarking the two libraries when parsing the CommonMark Spec file, markdown-it-pyrs is 20x faster than markdown-it-py.

Name (time, ms)	Min	Max	Mean	Rounds
markdown-it-pyrs	5.217	7.969	5.968	85
markdown-it-py	122.696	143.246	131.431	7

The drawback is that the library vendors compiled Rust code, and so:

Parser plugins cannot currently be written in Python and added dynamically to the parser.
It can be more difficult to integrate into environments like pyiodide and py-script (but maybe not for long: https://discuss.python.org/t/support-wasm-wheels-on-pypi/21924/3).

Usage

First install the package:

pip install markdown-it-pyrs

Then use it like you would markdown-it-py:

from markdown_it_pyrs import MarkdownIt

md = MarkdownIt("commonmark").enable("table")
md.render("# Hello, world!")
# '<h1>Hello, world!</h1>\n'

markdown-it.rs does not generate a token stream, but instead directly generates a Node tree. This is similar to the markdown-it-py's SyntaxTreeNode class, although the API is not identical. (source mapping is also provided by byte-offset, rather than line only)

md = (
  MarkdownIt("commonmark")
  .enable("table")
  .enable_many(["linkify", "strikethrough"])
)
node = md.tree("# Hello, world!")
print(node.walk())
# [Node(root), Node(heading), Node(text)]
print(node.pretty(srcmap=True, meta=True))
# <root srcmap="0:15">
#   <heading srcmap="0:15">
#     level: 1
#     <text srcmap="2:15">
#       content: Hello, world!

Note: Attributes of the Node class, such as Node.attrs, return a copy of the underlying data, and so mutating it will not affect what is stored on the node, e.g.

from markdown_it_pyrs import Node
node = Node("name")
# don't do this!
node.attrs["key"] = "value"
print(node.attrs) # {}
# do this instead (Python 3.9+)
node.attrs = node.attrs | {"key": "value"}
print(node.attrs) # {"key": "value"}
# Node.children is only a shallow copy though, so this is fine
child = Node("child")
node.children = [child]
node.children[0].name = "other"
print(child.name) # "other"

Command Line Interface

A CLI is also provided, which can be used like this:

echo "# Hello, world!" | markdown-it-pyrs html -
# <h1>Hello, world!</h1>
echo "# Hello, world!" | markdown-it-pyrs ast -
# <root>
#   <heading>
#     <text>

Replace - with a filename to read from a file, and see markdown-it-pyrs --help for more options, including initial configuration and enabling plugins.

Initial Configuration

Initialising MarkdownIt("zero") will not enable any plugins, and so you can add only the ones you need.

Use MarkdownIt("commonmark") to enable all the CommonMark plugins.

Use MarkdownIt("gfm") to enable all the CommonMark plugins, plus the GitHub Flavoured Markdown plugins.

Plugins

All syntax rules in markdown-it.rs are implemented as plugins. Plugins can be added to the parser by calling enable or enable_many with the name of the plugin. The following plugins are currently supported:

CommonMark Blocks:

blockquote: Block quotes with >
code: Indented code blocks
fence: Backtick code blocks
heading: # ATX headings
hr: --- horizontal rules
lheading: --- underline setext headings
list: * unordered lists and 1. ordered lists
paragraph: Paragraphs
reference: Link reference definitions [id]: src "title"

CommonMark Inlines:

autolink: <http://example.com>
backticks: `code`
emphasis: _emphasis_, *emphasis*, **strong**, __strong__
entity: &
escape: backslash escaping \
image: ![alt](src "title")
link: [text](src "title"), [text][id], [text]
newline: hard line breaks
html_block: HTML blocks
html_inline: HTML inline

GitHub Flavoured Markdown (https://github.github.com/gfm):

table:

| foo | bar |
| --- | --- |
| baz | bim |

strikethrough: ~~strikethrough~~
tasklist: - [x] tasklist item
autolink_ext: Extended autolink detection with "bare URLs" like https://example.com and www.example.com
tagfilter: HTML tag filtering, e.g. <script> tags are removed

Others:

sourcepos: Add source mapping to rendered HTML, looks like this: <stuff data-sourcepos="1:1-2:3">, i.e. line:col-line:col
replacements: Typographic replacements, like -- to —
smartquotes: Smart quotes, like " to “
linkify: Automatically linkify URLs with https://crates.io/crates/linkify (note currently this only matches URLs with a scheme, e.g. https://example.com)
heading_anchors: Add heading anchors, with defaults like GitHub
front_matter: YAML front matter
footnote: Pandoc-style footnotes (see https://pandoc.org/MANUAL.html#footnotes)
deflist: Definition lists (see https://pandoc.org/MANUAL.html#definition-lists)

Development

I'm quite new to Rust, so if you see something that could be improved, issues and PRs are welcome!

PyO3 and Maturin are used to build the Python package, by wrapping markdown-it.rs in a Python module.

pre-commit is used to run code formatting and linting checks, and tox is used to run tests.

TODO

Improvements:

Allow to override options:
- lang_prefix: Prefix for language classes on fenced code blocks
- quotes: Quote characters, for smart quotes
Add plugins: ...
Allow options for plugins:
- heading anchors
- tasklist checkboxes to be disabled
- footnotes with options to turn on/off inline/collect/backrefs

Open issue upstream:

no text_join rule (to join adjacent text and text_special tokens)
heading_anchors plugin does not allow for e.g. GitHub format where non-uniqueness is resolved by appending -1, -2, etc, and also removal of image text
Capture reference nodes in AST
disable rules
better "cross-language" AST representation
differing behaviour of linkify and normalize_url/commonmark_extras test failures
quote characters for smart-quotes and lang_prefix for fence should both be variable at run-time? (currently they both must be compiled)
fix docstring in examples/ferris/block_rule.rs::FerrisBlockScanner::run, which currently describes the JS API not the new rust one
Capture "piece-wise" source maps for nested content, e.g. for when the source is split over multiple lines and nested in another block (could get inline here https://github.com/rlidwka/markdown-it.rs/blob/6f906b38c8ffc3cc651e67b448b3655b7d0debb3/src/parser/inline/mod.rs#L115)
easier way to get root.ext items in core rules; it seems at present you have to swap memory and reswap at the end of the rule, see e.g. the InlineParserRule
allow test_rules_at_line to parse what the calling rule is, so that other rules can decide whether to interrupt based on the calling rule (in the check function), I think this would then allow behaviour similar to what alt did (possibly needed for footnote definition parsing)
- In general though, where back-compatibility is not required, I agree with djot goal 7, i.e. that block elements should not be allowed to interrupt other block elements without a newline
The possibility to return multiple (sequential) nodes from an InlineRule.run, e.g. ((node1, length1), (node2, length2), ...)
- This would be similar to docutils
In the Node walk methods, allow the function to return something to indicate whether to continue walking the children of the node
- Also in walk, parse the "path" to get to the current node, e.g. list of parent nodes, to allow for backtracing
- similar to https://github.com/syntax-tree/unist-util-visit-parents, https://github.com/syntax-tree/unist-util-visit, etc
is there a way to use a match statement, to match a Node against a NodeValue implementation? (rather than if/else for Node.cast)
Rule priority as an integer (similar to RST transform priority)
- Currently can only specify before or after another rule or all rules
- Can feel a little unclear for plugins, when to use attrs and when to add fields to node value.

Maintenance: