TypeScript icon, indicating that this package has built-in type declarations

2.1.5 • Public • Published



Defines the shape of Yozora Tokenizer and life cycle methods, as well as some utility functions to assist in resolving tokens.


  • npm

    npm install --save @yozora/core-tokenizer
  • yarn

    yarn add @yozora/core-tokenizer


According to the Parse Strategy, there are two types of tokenizers: Block Tokenizer and Inline tokenizer.

Block Tokenizer

The parsing steps of the block tokenizer are divided into three life cycles:

  • match-block: match a block node and get a BlockToken

  • parse-block: Parse BlockTokens into a YAST nodes

match-block phase

In the process of parsing block nodes, the content is read line by line. The block-level node has a nested structure:

> This is a blockquote
> - This is a list item in blockquote
> - # This is a setext heading in the list item of the blockquote
> - > ...

As shown in the second line of the above code, when parsing ListItem, it cannot get the first character in the original document line, but wait for its ancestor elements along the existing nesting structure (such as the above Blockquote) to complete the matching, and then gets a matching opportunity. In order to make the tokenizers work with each other transparently, when designing the life cycle methods of the block-level tokenizer in the match-block stage, the parsing logic of the nested structure lifted into @yozora/core-parser , and use a special data structure called PhrasingContentLine as the actual parsing unit of a line:

export interface PhrasingContentLine {
   * Start index of interval in nodePoints.
  startIndex: number
   * End index of interval in nodePoints.
  endIndex: number
   * Array of NodePoint which contains all the contents of this line.
  nodePoints: ReadonlyArray<NodePoint>
   * The index of first non-blank character in the rest of the current line
  firstNonWhitespaceIndex: number
   * The precede space count, one tab equals four space.
   * @see https://github.github.com/gfm/#tabs
  countOfPrecedeSpaces: number

The life cycle methods at this stage is subdivided into the following methods (see match-block for the type definition details):

  • eatOpener: (Required) Try to match a new block node.

  • eatAndInterruptPreviousSibling: (optional) try to interrupt the previous sibling node and match a new block node.

  • eatContinuationText: (Optional) Try to match the continuation text of current block node, that is, consume the current PhrasingContentLine with the current block node. There may be many kinds of results at this stage, which are distinguished according to the value of status in the returned result:

    • notMatched: Not matched.

    • closing: Matched and this is the last line of the current block node. That is, the current block node is in a saturated state and is closing.

    • opening: Matched, and not closing yet.

    • failedAndRollback: The match fails, and the content of the previous lines are to be rolled back. For convenience, it is assumed that the rollback operation does not affect the previously satisfied nested structure.

    • closingAndRollback: Matching failed, but only the last line needs to be rollback, the current node is still a valid one and will be closed soon.

  • eatLazyContinuationText: (Optional) Try to match Laziness Continuation Text. Actually only the @yozora/tokenizer-paragraph needs to implement this method, see https://github.github.com/gfm/#phase-1-block-structure step3 for details.

  • onClose: (Optional) Called when the current node is closed, used to perform

  • some cleanup operations.

  • extractPhrasingContentLines: (Optional) Convert a Block Token generated by the current tokenizer to PhrasingContentLines[]. This method is only needed when the matching node of this type may be rolled back.

  • buildBlockToken: (Optional) Convert PhrasingContentLines[] into a Block Token. This method is only needed when the matching node of this type may be rolled back

parse-block phase

The life cycle methods at this stage is subdivided into the following methods (see parse-block for the complete type definition):

  • parseBlock: Convert a Block Token into Yast Node

Inline Tokenizer

The parsing step of the inline parser is divided into two life cycles

  • match-inline: Match the inline contents and get an InlineToken
  • parse-inline: Parse an InlineToken into a YAST node

match-inline phase

After a block node is closed, we can start matching inline nodes, so when we match inline nodes, we get a continuous text without the concept of "line". But inline nodes have priority. For example, link has a higher priority than emphasis (see https://github.github.com/gfm/#example-529). In order to enable unperceptual coordination between tokenizers, when designing the life cycle function of the inline tokenizer in the match-inline phase, put priority-related logic in @yozora/core-parser In processing, each tokenizer only provides four types of separators: opener, both, closer, full. Then the processor in @yozora/core-parser completes the coordination work.

The lifecycle methods at this stage is subdivided into the following methods (see match-inline for the complete type definition):

  • findDelimiter: (Required) Find a delimiter
  • isDelimiterPair: (Optional) Check whether the given two delimiters can match
  • processDelimiterPair: (Optional) Process the matched two delimiters. Such as @yozora/tokenizer-emphasis
  • processSingleDelimiter: (Optional) Process a single delimiter. Such as @yozora/tokenizer-text

parser-inline phase

The lifecycle methods at this stage is subdivided into the following methods (see [parse-inline][lifecycle-pase-inline] for the complete type definition):

  • processToken: (Required) Convert an Inline Token to a YAST node.





npm i @yozora/core-tokenizer

DownloadsWeekly Downloads






Unpacked Size

88.6 kB

Total Files


Last publish


  • lemonclown