Markdown parser and generator written in Kotlin
intellij-markdown is a fast and extensible markdown processor. It is aimed to suit the following needs:
- Use one code base for both client and server-side processing;
- Support different flavours;
- Be easily extensible.
Since the parser is written in Kotlin, it can be compiled to both JS and Java bytecode thus can be used everywhere.
One of the goals of this project is to provide flexibility in terms of the tasks being solved. Markdown plugin for JetBrains IDEs is an example of usage when markdown processing is done in several stages:
- Parse block structure without parsing inlines to provide lazy parsable blocks for IDE;
- Quickly parse inlines of a given block to provide faster syntax highlighting update;
- Generate HTML for preview.
These tasks may be completed independently according to the current needs.
Simple html generation (Kotlin)
val src = "Some *Markdown*" val flavour = CommonMarkFlavourDescriptor() val parsedTree = MarkdownParser(flavour).buildMarkdownTreeFromString(src) val html = HtmlGenerator(src, parsedTree, flavour).generateHtml()
Simple html generation (Java)
final String src = "Some *Markdown*"; final MarkdownFlavourDescriptor flavour = new GFMFlavourDescriptor(); final ASTNode parsedTree = new MarkdownParser(flavour).buildMarkdownTreeFromString(text); final String html = new HtmlGenerator(src, parsedTree, flavour, false).generateHtml()
The parsing process is held in two logical parts:
- Splitting the document into the blocks of logical structure (lists, blockquotes, paragraphs, etc.);
- Parsing the inline structure of the resulted blocks.
This is the same way as the one being proposed in Commonmark spec.
Building the logical structure
Each (future) node (list, list item, blockquote, etc.) is associated with the so-called marker block. The rollback-free parsing algorithm is processing every token in the file, one by one. Tokens are passed to the opened marker blocks, and each block chooses whether to:
- do nothing
- drop itself
- complete itself
The marker processor stores the blocks, executes the actions chosen by the blocks, and, possibly, adds some new ones.
For the sake of speed and parsing convenience the text is passed to the lexer at first. Then the resulting set of tokens is processed in the special way.
Since the inline constructs in markdown have some priorities
(i.e. if two different ones overlap, the parsing result depends on the their types, not their positions,
*code, `not* emph` and
`code, *not` emph* are both code spans + literal asterisks), normal
recursive parsing is inapplicable.
Still the parsing of the inlines is quite straightforward. For each inline construct there is a particular sequential parser which accepts some input text and returns:
- The parsed ranges found in this text;
- The sub-text(s), which are to be passed to the subsequent inline parsers.
After building logical structure and parsing inlines a set of ranges corresponding to some markdown entities (i.e. nodes) is given. In order to work with the results effectively it ought to be converted to the AST.
For a given AST root a special visitor to generate the resulting HTML is created. Using a given mapping from IElementType to the HTML generating provider it processes the parsed tree in Depth-First order, generating HTML pieces for on each node visit.
Extending the parser
Many routines in the above process can be extended or redefined by creating a different markdown flavours. The minimal default is CommonMark which is implemented in this project. Github Flavoured Markdown (which is also implemented) is an example of extending CommonMark flavour. It can be used as a reference for implementing your own markdown features.
Below is a tree representing a big part of the API.
MarkdownFlavourDescriptoris a base class for extending markdown parser.
markerProcessorFactoryis responsible for block structure customization.
stateInfovalue allows to use a state during document parsing procedure.
updateStateInfo(pos: LookaheadText.Position)is called at the beginning of each position processing
populateConstraintsTokensis called to create nodes for block structure markers at the beginning of the lines (for example,
>characters constituting blockquotes)
getMarkerBlockProvidersis a place to (re)define types of block structures
getParserSequencedefines inlines parsing procedure. The method must return a list of SequentialParsers where the earliest parsers have the biggest operation precedence. For example, to parse code spans and emph with correct priority, the list should be [CodeSpanParser, EmphParser] but not the opposite.
SequentialParserhas only one method:
parse(tokens: TokensCache, rangesToGlue: List<IntRange>): ParsingResult
tokensis a special holder for the tokens returned by lexer
rangesToGlueis a list of ranges in the document which are to be searched for the structures in question.
Considering the input:
A * emph `code * span` b * cfor the emph parser ranges [
A * emph,
b * c] mean that emph must be searched in the input
A * emph | b * c.
The method must essentially return the parsing result (nodes for the found structures) and the parts of the text to be given to the next parsers.
Considering the same input for the code span parser the result would be
`code * span`of the type "code span" and the delegate pieces would be [
A * emph,
b * c].
createInlinesLexershould return the lexer to split the text to the tokens before inline parsing procedure run.
createHtmlGeneratingProviders(linkMap: LinkMap, baseURI: URI?)is the place where generated HTML is customized. This method should return a map which defines how to handle the particular kinds of the nodes in the resulting tree.
linkMaphere is a precalculated information about the links defined in the document with the means of link definition.
baseURIis the URI to be considered the base path for the relative links resolving. For example, given
foo/bar.pngshould be transformed to the
Each returned provider must implement
processNode(visitor: HtmlGenerator.HtmlGeneratingVisitor, text: String, node: ASTNode)where
textis the whole document being processed,
nodeis the node being given to the provider,
visitoris a special object responsible for the HTML generation. See
GeneratingProviders.ktfor the samples.