Module ast 

AST node definitions.

Node taxonomy (heading / paragraph / list / html / emphasis / …), sibling order and spans follow CommonMark’s grammar; a printer’s dispatch is written per this unit.

Beyond that the shapes deliberately diverge from mdast:

• No cooked values: consumers slice the original source through Spans. Inline node spans are original-source offsets, never rebuilt-buffer ones.
• Style facts a printer would otherwise re-derive by peeking at the source (markers, fence chars, reference kinds, break kinds, …) are first-class fields.
• Formatter policy (sentence splitting, CJK classification, alignment judgment) stays out; the AST records lexical facts only.

Blank lines are not stored on nodes: ParserReturn::blanks lists them once, and sibling spans locate the ones between two blocks (List::tight is computed from it).

Structs

Autolink

<https://example.com> / <user@example.com>.

AutolinkLiteral

GFM autolink literal (www.example.com, bare https://… / email).

Blockquote

CodeBlock

CodeSpan

ContainerDirective

A container directive:

Definition

A link reference definition: [label]: dest "title".

Destination

A link/image destination, excluding the title.

Emphasis

FootnoteDefinition

FootnoteReference

GFM [^label].

HardBreak

Heading

HtmlBlock

A raw HTML block. Contents are never parsed or formatted.

HtmlInline

A raw inline HTML tag or comment. The tag is a verbatim token; text around it is markdown.

Image

![alt](dest "title") and the reference forms.

Link

[text](dest "title") or [text][label] / [text][] / [text].

Liquid

A {% … %} / {{ … }} template tag, block-level or inline. Never interpreted: the whole construct, delimiters included, is kept verbatim.

List

ListItem

MathBlock

$$ … $$ display math, content is kept raw. As with code fences, the fence length is not stored (printers normalize to $$; the $ run at span.start gives it if ever needed).

MathSpan

$ … $ inline math. The whole construct, $ sequences included, is kept raw; content is never trimmed or decoded.

MdxEsm

An MDX import/export statement. The parser only establishes the boundary (via the host’s JS parser callback); the content stays raw for embedded formatting.

MdxExpression

An MDX {expr}. Boundary only; the expression stays raw.

MdxJsxFlow

An MDX JSX element at flow level. Tags stay raw for embedded formatting; children are markdown blocks.

MdxJsxText

An MDX JSX element in text. Children are inline markdown.

Paragraph

Root

The document.

SoftBreak

Strikethrough

GFM ~~x~~ (~x~ too with gfm_strikethrough_single_tilde).

Strong

Table

TableCell

TableRow

TaskCheckbox

[ ], [x] or [X] (the raw span keeps the case).

Text

ThematicBreak

WikiLink

[[target]], alias-less. Always single-line; the target is span + 2 .. span - 2, kept raw.

Enums

Block

Flow (block-level) content.

CodeBlockKind

HardBreakKind

HeadingKind

Inline

Phrasing (inline) content.

LinkKind

Titles and labels are logical-line pieces for the same reason as on Definition.

ListMarker

ReferenceKind

Reference-link form. Only Full is reformatted; the others are printed verbatim.

TableAlign