CommonMark is a rationalized version of Markdown syntax, with a spec and BSD3-licensed reference implementations in C and JavaScript.
The C implementation provides both a shared library (libcmark) and a
standalone program cmark that converts CommonMark to HTML. It is
written in standard C99 and has no library dependencies. The parser is
very fast (see benchmarks).
It is easy to use libcmark in python, lua, ruby, and other dynamic
languages: see the wrappers/ subdirectory for some simple examples.
The JavaScript implementation provides both an NPM package and a single JavaScript file, with no dependencies, that can be linked into an HTML page. For further information, see the README in the js directory.
A note on security: Neither implementation attempts to sanitize link attributes or raw HTML. If you use these libraries in applications that accept untrusted user input, you must run the output through an HTML sanitizer to protect against XSS attacks.
Building the C program (cmark) and shared library (libcmark)
requires cmake. If you modify scanners.re, then you will also
need re2c, which is used to generate scanners.c from
scanners.re. We have included a pre-generated scanners.c in
the repository to reduce build dependencies.
If you have GNU make, you can simply make, make test, and make install. This calls cmake to create a Makefile in the build
directory, then uses that Makefile to create the executable and
library. The binaries can be found in build/src.
For a more portable method, you can use cmake manually. cmake knows how to create build environments for many build systems. For example, on FreeBSD:
mkdir build
cd build
cmake .. # optionally: -DCMAKE_INSTALL_PREFIX=path
make # executable will be created as build/src/cmark
make test
make install
Or, to create Xcode project files on OSX:
mkdir build
cd build
cmake -G Xcode ..
make
make test
make install
The GNU Makefile also provides a few other targets for developers. To run a benchmark:
make bench
To run a "fuzz test" against ten long randomly generated inputs:
make fuzztest
To run a test for memory leaks using valgrind:
make leakcheck
To reformat source code using astyle:
make astyle
To make a release tarball and zip archive:
make archive
To compile with MSVC and NMAKE:
nmake
You can cross-compile a Windows binary and dll on linux if you have the
mingw32 compiler:
make mingw
The binaries will be in build-mingw/windows/bin.
The JavaScript library can be installed through npm:
npm install commonmark
This includes a command-line converter called commonmark.
If you want to use it in a client application, you can fetch
a pre-built copy of commonmark.js from
http://spec.commonmark.org/js/commonmark.js.
For further information, see the README in the js directory.
The spec contains over 500 embedded examples which serve as conformance
tests. To run the tests using an executable $PROG:
python3 test/spec_tests.py --program $PROG
If you want to extract the raw test data from the spec without actually running the tests, you can do:
python3 test/spec_tests.py --dump-tests
and you'll get all the tests in JSON format.
The source of the spec is spec.txt. This is basically a Markdown
file, with code examples written in a shorthand form:
.
Markdown source
.
expected HTML output
.
To build an HTML version of the spec, do make spec.html. To build a
PDF version, do make spec.pdf. (Creating a PDF requires pandoc
and a LaTeX installation. Creating the HTML version requires only
libcmark and python3.)
The spec is written from the point of view of the human writer, not the computer reader. It is not an algorithm---an English translation of a computer program---but a declarative description of what counts as a block quote, a code block, and each of the other structural elements that can make up a Markdown document.
Because John Gruber's canonical syntax description leaves many aspects of the syntax undetermined, writing a precise spec requires making a large number of decisions, many of them somewhat arbitrary. In making them, we have appealed to existing conventions and considerations of simplicity, readability, expressive power, and consistency. We have tried to ensure that "normal" documents in the many incompatible existing implementations of Markdown will render, as far as possible, as their authors intended. And we have tried to make the rules for different elements work together harmoniously. In places where different decisions could have been made (for example, the rules governing list indentation), we have explained the rationale for my choices. In a few cases, we have departed slightly from the canonical syntax description, in ways that we think further the goals of Markdown as stated in that description.
For the most part, we have limited ourselves to the basic elements described in Gruber's canonical syntax description, eschewing extensions like footnotes and definition lists. It is important to get the core right before considering such things. However, we have included a visible syntax for line breaks and fenced code blocks.
There are only a few places where this spec says things that contradict the canonical syntax description:
-
It allows all punctuation symbols to be backslash-escaped, not just the symbols with special meanings in Markdown. We found that it was just too hard to remember which symbols could be escaped.
-
It introduces an alternative syntax for hard line breaks, a backslash at the end of the line, supplementing the two-spaces-at-the-end-of-line rule. This is motivated by persistent complaints about the “invisible” nature of the two-space rule.
-
Link syntax has been made a bit more predictable (in a backwards-compatible way). For example,
Markdown.plallows single quotes around a title in inline links, but not in reference links. This kind of difference is really hard for users to remember, so the spec allows single quotes in both contexts. -
The rule for HTML blocks differs, though in most real cases it shouldn't make a difference. (See the section on HTML Blocks for details.) The spec's proposal makes it easy to include Markdown inside HTML block-level tags, if you want to, but also allows you to exclude this. It is also makes parsing much easier, avoiding expensive backtracking.
-
It does not collapse adjacent bird-track blocks into a single blockquote:
> this is two > blockquotes > this is a single > > blockquote with two paragraphs -
Rules for content in lists differ in a few respects, though (as with HTML blocks), most lists in existing documents should render as intended. There is some discussion of the choice points and differences in the subsection of List Items entitled Motivation. We think that the spec's proposal does better than any existing implementation in rendering lists the way a human writer or reader would intuitively understand them. (We could give numerous examples of perfectly natural looking lists that nearly every existing implementation flubs up.)
-
The spec stipulates that two blank lines break out of all list contexts. This is an attempt to deal with issues that often come up when someone wants to have two adjacent lists, or a list followed by an indented code block.
-
Changing bullet characters, or changing from bullets to numbers or vice versa, starts a new list. We think that is almost always going to be the writer's intent.
-
The number that begins an ordered list item may be followed by either
.or). Changing the delimiter style starts a new list. -
The start number of an ordered list is significant.
-
Fenced code blocks are supported, delimited by either backticks (
```or tildes (~~~).
There is a forum for discussing CommonMark; you should use it instead of github issues for questions and possibly open-ended discussions. Use the github issue tracker only for simple, clear, actionable issues.
The spec was written by John MacFarlane, drawing on
- his experience writing and maintaining Markdown implementations in several languages, including the first Markdown parser not based on regular expression substitutions (pandoc) and the first markdown parsers based on PEG grammars (peg-markdown, lunamark)
- a detailed examination of the differences between existing Markdown implementations using BabelMark 2, and
- extensive discussions with David Greenspan, Jeff Atwood, Vicent Marti, Neil Williams, and Benjamin Dumke-von der Ehe.
John MacFarlane was also responsible for the original versions of the C and JavaScript implementations. The block parsing algorithm was worked out together with David Greenspan. Vicent Marti optimized the C implementation for performance, increasing its speed tenfold. Kārlis Gaņģis helped work out a better parsing algorithm for links and emphasis, eliminating several worst-case performance issues. Nick Wellnhofer contributed many improvements, including most of the C library's API and its test harness. Vitaly Puzrin has offered much good advice about the JavaScript implementation.