M-Press Flavoured Markdown specification

Fast, readable documentation

M-Press turns plain Markdown into complete documentation.

Read the authoring guide and build the site.

## Fast, readable documentation

M-Press turns _plain Markdown_ into complete documentation.

[Read the authoring guide](/authoring/) and build the site.

• Write the guide.
• Run the checks.
• Publish the site.
  • Keep the source readable.
  • Keep the output accessible.

- [x] Write the guide.
- [x] Run the checks.
- [ ] Publish the site.
  - Keep the source readable.
  - Keep the output accessible.

Metadata

Page metadata uses an optional delimited block. The opening --- must be the first line of the document and must contain no other characters. No blank line, space, or comment may precede it. An optional UTF-8 byte order mark is not document content. A closing --- must occupy its own line.

The metadata body uses M-Press metadata syntax, not YAML. Each field occupies one line and uses key = value syntax.

---
schema = 1
title = "Install Wails"
description = "Install the tools required to build a Wails application."
slug = "quick-start/installation"
order = 20
draft = false
tags = ["installation", "beginner"]
---

A metadata key is an ASCII identifier. It may contain letters, digits, _, -, and ., and must begin with a letter or _. A value must be a JSON string, number, Boolean, null, array, or object on the same logical line.

When a metadata block is present, schema is required, must be its first field, and must be a positive integer. A parser must reject an unsupported schema before parsing document content. The following bare keys are permanently reserved by schema 1:

schema, title, translationKey, description, draft, layout, slug, order, date, template, banner, hero, tags, author, authors, image, imageMode, imageFit, imageBackground, imageWidth, showTags, and headingSize.

These names are case-sensitive. The schema 1 reserved set is frozen. A later schema must not assign MPress meaning to another unprefixed key because an existing document may already use that key as custom metadata.

All keys beginning with mpress. are reserved for present and future MPress features. Custom metadata may use any other unreserved key. Namespaced custom keys such as wails.section, seo.image, and analytics.campaign are recommended. Unknown custom keys are retained exactly and made available to the application. They must never be silently reinterpreted by a later schema.

Duplicate keys are an error. A first-line --- always opens metadata, so a missing closing delimiter is an error. A --- line anywhere else is plain text.

Blocks

A block begins at the start of a logical line. Except inside a list or quote, leading indentation does not change the block type. A block construct must be separated from surrounding paragraphs by a blank line.

A parser must decide a block type from the current line. It must not reinterpret earlier lines after reading later content.

Paragraphs

One or more ordinary lines form a paragraph. A blank line ends the paragraph. The @end directive for the current container can also end it. Another block marker without a preceding blank line remains paragraph text.

MPress builds complete documentation from plain text.
This line starts on a new rendered line.

This sentence is \
continued without a rendered line break.

This is a new paragraph.

A single line ending inside a paragraph creates a hard line break. A backslash immediately before the line ending consumes that line ending. The text on the next physical line continues in the same inline flow. Any space before the backslash remains content, so prose can wrap without joining adjacent words.

Text cannot accidentally become another block after a paragraph has started. For example, # text on the second line of a paragraph remains text.

Headings

A heading starts with one to six # characters followed by one ASCII space. It must occupy one line.

# Install MPress

## Verify the installation

Closing # characters have no special meaning. Setext headings are not part of M-Press Flavoured Markdown.

Rules

The reserved directive @hr creates a thematic break.

Before the break.

@hr

After the break.

Lists

An unordered item begins with - . An ordered item begins with one or more ASCII digits followed by . . List indentation must be a multiple of two spaces. Tabs are not permitted for structural indentation.

- Install Go.
- Install MPress.
  - Build the documentation.
  - Run the checks.

1. Create the project.
2. Preview the site.

The first item determines whether a list is ordered. Every item at that depth must use the same marker type. Ordered item numbers are retained as source metadata. A canonical serializer starts at the first number and increments subsequent numbers.

Continuation lines must align two spaces beyond their item marker. There are no lazy continuation lines. A nested block or list must also begin two spaces beyond its parent item.

- First paragraph in the item.

  Second paragraph in the same item.

  @note type="tip"
  A component in the item.
  @end

A task item places [ ], [x], or [X] immediately after its list marker.

- [ ] Write the guide.
- [x] Run the checks.

Quotes

A quote line begins with > or consists only of >. Every line in the quote must carry the prefix. Repeating the prefix creates a nested quote.

> Documentation is part of the product.
>
> > Clear examples make adoption easier.

The parser removes one prefix at a time and applies the block grammar to the remaining text.

Code blocks

A code block uses a fence of three or more backticks. The opening fence may be followed by a language identifier and an attribute list. The closing fence must contain at least as many backticks as the opening fence and no other non-whitespace characters.

````go {title="Build the site" lineNumbers=true}
site, err := mpress.Build(config)
@end
```
````

The body is opaque. Every byte between the opening and closing fences is code. Directives, inline roles, links, and formatting are not recognised there. An @, an exact @end line, and a shorter backtick fence are all literal content. An author must use a longer opening and closing fence when the body contains a line that would otherwise close the block. No @ escaping is permitted or required inside a code block.

The attributes language, title, filename, highlight, and lineNumbers have the same meanings as their MPress Markdown component equivalents.

A fence whose language is d2 renders a static SVG diagram. Its body is passed unchanged to the bundled D2 compiler using the Dagre layout engine. File imports are disabled, and invalid diagrams fail the build. See D2 diagrams for an example.

Raw HTML

Raw HTML is explicit.

@rawHTML
<details><summary>More</summary>Content</details>

<div class="mpress-preview-source"><div class="mpress-preview-source-label">Markdown</div><pre><code class="language-markdown"><button class="mpress-image-expand" type="button" aria-haspopup="dialog" aria-label="Expand image: &amp;#34;A"><span class="mpress-image-expand-content"><img src="&amp;#34;/images/mpress-hello-output.png&amp;#34;" alt="&amp;#34;A" loading="lazy"></span></button>&#10;&#10;## Metadata&#10;&#10;Page metadata uses an optional delimited block. The opening `---` must be the&#10;first line of the document and must contain no other characters. No blank line,&#10;space, or comment may precede it. An optional UTF-8 byte order mark is not&#10;document content. A closing `---` must occupy its own line.&#10;&#10;The metadata body uses M-Press metadata syntax, not YAML. Each field occupies&#10;one line and uses `key = value` syntax.&#10;&#10;```text&#10;---&#10;schema = 1&#10;title = &#34;Install Wails&#34;&#10;description = &#34;Install the tools required to build a Wails application.&#34;&#10;slug = &#34;quick-start/installation&#34;&#10;order = 20&#10;draft = false&#10;tags = [&#34;installation&#34;, &#34;beginner&#34;]&#10;---&#10;```&#10;&#10;A metadata key is an ASCII identifier. It may contain letters, digits, `_`,&#10;`-`, and `.`, and must begin with a letter or `_`. A value must be a JSON&#10;string, number, Boolean, null, array, or object on the same logical line.&#10;&#10;When a metadata block is present, `schema` is required, must be its first field,&#10;and must be a positive integer. A parser must reject an unsupported schema&#10;before parsing document content. The following bare keys are permanently&#10;reserved by schema 1:&#10;&#10;`schema`, `title`, `translationKey`, `description`, `draft`, `layout`, `slug`,&#10;`order`, `date`, `template`, `banner`, `hero`, `tags`, `author`, `authors`,&#10;`image`, `imageMode`, `imageFit`, `imageBackground`, `imageWidth`, `showTags`,&#10;and `headingSize`.&#10;&#10;These names are case-sensitive. The schema 1 reserved set is frozen. A later&#10;schema must not assign MPress meaning to another unprefixed key because an&#10;existing document may already use that key as custom metadata.&#10;&#10;All keys beginning with `mpress.` are reserved for present and future MPress&#10;features. Custom metadata may use any other unreserved key. Namespaced custom&#10;keys such as `wails.section`, `seo.image`, and `analytics.campaign` are&#10;recommended. Unknown custom keys are retained exactly and made available to&#10;the application. They must never be silently reinterpreted by a later schema.&#10;&#10;Duplicate keys are an error. A first-line `---` always opens metadata, so a&#10;missing closing delimiter is an error. A `---` line anywhere else is plain&#10;text.&#10;&#10;## Blocks&#10;&#10;A block begins at the start of a logical line. Except inside a list or quote,&#10;leading indentation does not change the block type. A block construct must be&#10;separated from surrounding paragraphs by a blank line.&#10;&#10;A parser must decide a block type from the current line. It must not reinterpret&#10;earlier lines after reading later content.&#10;&#10;### Paragraphs&#10;&#10;One or more ordinary lines form a paragraph. A blank line ends the paragraph.&#10;The `@end` directive for the current container can also end it. Another block&#10;marker without a preceding blank line remains paragraph text.&#10;&#10;```text&#10;MPress builds complete documentation from plain text.&#10;This line starts on a new rendered line.&#10;&#10;This sentence is \&#10;continued without a rendered line break.&#10;&#10;This is a new paragraph.&#10;```&#10;&#10;A single line ending inside a paragraph creates a hard line break. A backslash&#10;immediately before the line ending consumes that line ending. The text on the&#10;next physical line continues in the same inline flow. Any space before the&#10;backslash remains content, so prose can wrap without joining adjacent words.&#10;&#10;Text cannot accidentally become another block after a paragraph has started.&#10;For example, `# text` on the second line of a paragraph remains text.&#10;&#10;### Headings&#10;&#10;A heading starts with one to six `#` characters followed by one ASCII space.&#10;It must occupy one line.&#10;&#10;```text&#10;# Install MPress&#10;&#10;## Verify the installation&#10;```&#10;&#10;Closing `#` characters have no special meaning. Setext headings are not part of&#10;M-Press Flavoured Markdown.&#10;&#10;### Rules&#10;&#10;The reserved directive `@hr` creates a thematic break.&#10;&#10;```text&#10;Before the break.&#10;&#10;@hr&#10;&#10;After the break.&#10;```&#10;&#10;### Lists&#10;&#10;An unordered item begins with `- `. An ordered item begins with one or more&#10;ASCII digits followed by `. `. List indentation must be a multiple of two&#10;spaces. Tabs are not permitted for structural indentation.&#10;&#10;```text&#10;- Install Go.&#10;- Install MPress.&#10;  - Build the documentation.&#10;  - Run the checks.&#10;&#10;1. Create the project.&#10;2. Preview the site.&#10;```&#10;&#10;The first item determines whether a list is ordered. Every item at that depth&#10;must use the same marker type. Ordered item numbers are retained as source&#10;metadata. A canonical serializer starts at the first number and increments&#10;subsequent numbers.&#10;&#10;Continuation lines must align two spaces beyond their item marker. There are no&#10;lazy continuation lines. A nested block or list must also begin two spaces&#10;beyond its parent item.&#10;&#10;```text&#10;- First paragraph in the item.&#10;&#10;  Second paragraph in the same item.&#10;&#10;  @note type=&#34;tip&#34;&#10;  A component in the item.&#10;  @end&#10;```&#10;&#10;A task item places `[ ]`, `[x]`, or `[X]` immediately after its list marker.&#10;&#10;```text&#10;- [ ] Write the guide.&#10;- [x] Run the checks.&#10;```&#10;&#10;### Quotes&#10;&#10;A quote line begins with `&gt; ` or consists only of `&gt;`. Every line in the quote&#10;must carry the prefix. Repeating the prefix creates a nested quote.&#10;&#10;```text&#10;&gt; Documentation is part of the product.&#10;&gt;&#10;&gt; &gt; Clear examples make adoption easier.&#10;```&#10;&#10;The parser removes one prefix at a time and applies the block grammar to the&#10;remaining text.&#10;&#10;### Code blocks&#10;&#10;A code block uses a fence of three or more backticks. The opening fence may be&#10;followed by a language identifier and an attribute list. The closing fence must&#10;contain at least as many backticks as the opening fence and no other&#10;non-whitespace characters.&#10;&#10;`````text&#10;````go {title=&#34;Build the site&#34; lineNumbers=true}&#10;site, err := mpress.Build(config)&#10;@end&#10;```&#10;````&#10;`````&#10;&#10;The body is opaque. Every byte between the opening and closing fences is code.&#10;Directives, inline roles, links, and formatting are not recognised there. An&#10;`@`, an exact `@end` line, and a shorter backtick fence are all literal content.&#10;An author must use a longer opening and closing fence when the body contains a&#10;line that would otherwise close the block. No `@` escaping is permitted or&#10;required inside a code block.&#10;&#10;The attributes `language`, `title`, `filename`, `highlight`, and `lineNumbers`&#10;have the same meanings as their MPress Markdown component equivalents.&#10;&#10;A fence whose language is `d2` renders a static SVG diagram. Its body is passed&#10;unchanged to the bundled D2 compiler using the Dagre layout engine. File imports&#10;are disabled, and invalid diagrams fail the build. See [D2 diagrams](/code-components/#d2-diagrams)&#10;for an example.&#10;&#10;### Raw HTML&#10;&#10;Raw HTML is explicit.&#10;&#10;```text&#10;@rawHTML&#10;&lt;details&gt;&lt;summary&gt;More&lt;/summary&gt;Content&lt;/details&gt;</code></pre></div>

</div>
</ot-tabs>

The body is not parsed. It is HTML and the directive has no attributes.

Raw HTML is permitted because raw HTML is part of CommonMark. JSX, imports, exports, expressions, and other MDX constructs are not executable and must not be generated by a Markdown exporter.

Comments

Comments use an opaque block and are retained in the document tree.

@comment
Check this claim before the next release.
@end

Comments are not rendered. A Markdown serializer writes them as HTML comments with unsafe -- sequences escaped.

Inline content

Inline parsing is confined to one paragraph, heading, table cell, or component label. It must not depend on a definition that appears later in the document.

An inline delimiter can be escaped with \. An unmatched delimiter is literal text. Delimiters must be properly nested and must not cross.

Emphasis and strong text

Single _ delimiters create emphasis. Single * delimiters create strong text.

Use _emphasis_ sparingly and make *important text* clear.

A delimiter may open when it is followed by a non-whitespace byte. It may close when it is preceded by a non-whitespace byte. If the matching delimiter is not at the top of the delimiter stack, the byte is literal text.

This rule does not use Unicode categories, flanking rules, or doubled markers.

Code spans

A run of one or more backticks opens a code span. The next run of the same length closes it. A different run length is content. An unclosed span continues to the end of its inline container and produces a warning.

Run `mpress build` or use ``a ` character`` in an example.

Code span content is literal. A canonical serializer chooses a delimiter one character longer than the longest backtick run in the content.

Links and images

An inline link uses [label](destination). An image adds ! before the label.

[Read the guide](/getting-started/)
![Application window](/images/application.png)

The label may contain inline content with properly nested brackets. The destination must occur on the same logical line. It is either a sequence with balanced parentheses and backslash escapes or a JSON-quoted string.

Reference links use [label][identifier]. The second pair is required, so link recognition is local. Identifiers are case-sensitive.

[Download Go][go-download]

@link id="go-download" destination="https://go.dev/dl/"

An undefined reference remains a link node and produces a diagnostic. Shortcut reference syntax such as [identifier] is not part of M-Press Flavoured Markdown.

Automatic links

An absolute http, https, or mailto URI between angle brackets creates an automatic link.

<https://m-press.me>

Plain text that resembles a URL remains text. Automatic linking of bare text is a renderer option, not source syntax.

Emoji shortcodes

A recognised emoji name between colons creates an emoji node.

Build complete :white_check_mark: Ship it :rocket:

An emoji name contains one or more ASCII letters, digits, _, -, or +. Names are case-sensitive and canonical names are lowercase. Schema 1 uses a fixed snapshot of the GitHub-compatible emoji shortcode registry published with the conformance suite. A parser must use that local registry. It must not query a network service or allow a registry update to change the meaning of an existing schema 1 document.

Only a name in the schema registry creates an emoji node. An unknown name, including its colons, remains literal text and does not produce a diagnostic. A backslash before the first colon prevents recognition, so \:rocket: renders as the literal text :rocket:.

Emoji shortcodes are recognised in headings, paragraphs, link labels, table cells, and component prose. They are not recognised in code spans, code fences, raw HTML blocks, comments, metadata values, attribute values, or link destinations. An HTML renderer writes the registry’s native Unicode sequence. It must not require an image service, script, or client-side replacement.

Canonical M-Press Flavoured Markdown writes the canonical shortcode name. A Markdown importer using the mpress-markdown-1 profile recognises the same registry. Canonical MPress Markdown retains the shortcode. Portable CommonMark and GFM export write the native Unicode sequence because emoji shortcodes are not part of those specifications.

Additional inline roles

Less common inline semantics use an explicit role. This avoids assigning common punctuation more than one meaning.

Water is H@sub[2]O.
Press @kbd[Command K].
This is @mark[important] and @delete[obsolete].

The grammar is @name[content]. name is an ASCII identifier. Brackets inside the content must be balanced or escaped. The roles defined by version 0.1 are sub, sup, mark, insert, delete, kbd, var, and cite.

An unknown role remains an inline-role node. It must not disappear.

Metadata references

The reserved inline role metadata inserts a metadata value as text at the reference position.

This page is called @metadata[title].
The previous location was @metadata[wails.redirect].

The content between brackets is one complete, case-sensitive metadata key. Dots are part of the key, so wails.redirect performs an exact lookup of that namespaced key. It does not traverse an object. Reserved and custom metadata use the same lookup rules.

A string inserts its contents without quotation marks. Numbers, Booleans, and null use their canonical JSON spelling. Arrays and objects use compact canonical JSON. Object keys retain metadata source order. The inserted value is text, not markup, and must not be parsed again for roles, formatting, HTML, or components. Typographic substitutions must not change it. An HTML renderer must escape it as ordinary text.

A metadata reference is recognised in every inline-content context, including headings, paragraphs, link labels, table cells, and component prose. It is not recognised inside code fences, raw HTML blocks, comments, metadata values, attribute values, or link destinations.

If the key does not exist, the renderer retains the literal @metadata[key] source and produces an mpd-metadata-reference diagnostic. It must not insert an empty string or silently choose another key.

Escapes and entities

A backslash before ASCII punctuation makes the next byte literal. A backslash before any other byte remains a backslash.

M-Press Flavoured Markdown does not interpret HTML entities. Write the UTF-8 character itself. The Markdown adapter retains entity spelling as provenance and decodes it for the document model.

Components

A component is either a container or a leaf declared by the schema registry. Its opening line begins with @. A container body uses ordinary M-Press Flavoured Markdown, and @end closes it.

@note type="tip" title="No MDX required"
Components contain *ordinary document content*.
@end

Components can nest. The parser uses a stack and the closest @end closes the current container component.

The schema component registry declares each component as a container or a leaf. A leaf component consists only of its opening line and does not use @end.

@image light="/light.png" dark="/dark.png" alt="Architecture"

MPress controls the complete component registry. User-defined components are not part of M-Press Flavoured Markdown. An unknown directive produces an mpd-directive diagnostic and remains literal text. It does not open a container.

Attributes

Attributes follow the component name on the opening line. At least one ASCII space separates the name from its first attribute, and one or more ASCII spaces separate adjacent attributes. Braces are not part of M-Press Flavoured Markdown component syntax.

The opening line uses this grammar:

opening    = "@" name [spaces attributes] line-end
attributes = attribute *(spaces attribute)
attribute  = name | name "=" value
value      = string | number | boolean | null | array | object
spaces     = 1*(ASCII space)

Names use the metadata key grammar. Values use single-line JSON syntax. Bare string values are not permitted. A name without = is shorthand for name=true and is valid only for a declared boolean attribute. Duplicate attributes are an error.

Attribute parsing is confined to the opening line. JSON strings, arrays, and objects can contain spaces. The parser determines their ends from JSON syntax, not by splitting the line at every space.

The schema registry determines whether an opening line is complete or requires a matching @end. The parser does not infer this from source punctuation. A slash inside a quoted value, such as the slashes in a URL, is ordinary value content. A canonical serializer writes one space between the component name and each attribute.

These are canonical opening lines:

@note type="tip" title="Read this first"
@table header search page-size=20
@image src="/images/overview.png" alt="Product overview"
@hr

Literal directive lines

At the start of a logical line, a backslash escapes a directive’s initial @. This is an application of the normal punctuation escape rule, not a separate escape mechanism.

This rule does not apply inside code blocks. Code block bodies are opaque and retain every @ without an escape.

\@note is displayed as @note.

Video

Video is a leaf component backed by the browser’s native video element. The common case requires only a source filename:

@video src="setup.mp4"

MPress adds native controls, inline playback, metadata preloading, a responsive layout, a 1280 by 720 intrinsic size, the inferred MIME type, and a fallback link. Authors do not repeat those defaults.

The same src attribute accepts an array when a video has multiple encodings. The browser tries them in source order. base prefixes every relative media path, including sources, the poster, captions, and transcript:

@video base="/media/setup" src=["setup.webm","setup.mp4"] poster="poster.webp" captions=["captions.en.vtt","captions.fr.vtt"] title="Setup walkthrough" transcript="transcript.html"

The optional attributes are deliberately limited:

• base removes a repeated directory from relative media paths.
• poster sets the image shown before playback.
• captions accepts one WebVTT filename or an array. The first track is the default track.
• lang supplies the caption language when it cannot be inferred. It defaults to en.
• title supplies the accessible name and visible caption.
• transcript adds a transcript link.
• width and height replace the intrinsic 1280 by 720 dimensions. CSS keeps the player responsive.

Caption filenames can include a BCP 47 language tag immediately before the extension, such as captions.en.vtt or captions.pt-BR.vtt. MPress infers that tag before it uses lang. Video MIME types are inferred from .mp4, .m4v, .webm, .ogv, .ogg, and .mov filenames. Relative paths and HTTP or HTTPS URLs are accepted. Unsafe URL schemes are rejected.

MPress does not add autoplay, custom playback controls, or a JavaScript player API. A site that needs an unusual player policy can use @rawHTML; those cases do not enlarge the portable video component.

Tables

A table is an explicit block. Each row begins and ends with |. A backslash escapes a literal pipe.

@table header=true
| Name | Purpose |
| Markdown | Portable authoring |
| M-Press Flavoured Markdown | Deterministic authoring |
@end

Every row must contain the same number of cells. Leading and trailing ASCII spaces inside a cell are formatting trivia and are not cell content. Cells use the inline grammar.

Column alignment uses an optional align array with left, centre, right, or default values.

@table header=true align=["left", "right"]
| Stage | Time |
| Parse | 83 ms |
@end

Tables can provide reader controls. Each control is optional and disabled by default:

• search adds one case-insensitive search field for all cells.
• filter adds an exact-value filter menu to each column header.
• sort makes each header sortable in ascending or descending order.
• paginate divides matching rows into pages. page-size sets the number of rows per page and defaults to 10.
• column-separators draws each column boundary through the header and body.

@table header search filter sort paginate column-separators page-size=10
| Package | Platform | Downloads |
| Core | Linux | 1840 |
| Team | macOS | 920 |
| Core | Windows | 1260 |
@end

Search, filters, sorting, and pagination combine. A search or filter change returns to the first page. Sorting is stable, locale-aware, and compares embedded numbers numerically. Generated controls must have accessible names, sortable headers must expose aria-sort, and the visible row count must be an aria-live status. Without script support, every row remains visible and the underlying table remains readable.

Footnotes

An inline footnote reference uses [^identifier]. Its definition is an explicit block and can contain any block content.

The result is reproducible.[^benchmark]

@footnote id="benchmark"
The benchmark uses twenty paired cold builds.
@end

Identifiers are case-sensitive. A duplicate definition is an error. An undefined reference produces a diagnostic without deleting the reference.

Imports and includes

An include reads another M-Press Flavoured Markdown file or fragment. An import invokes a format adapter.

@include src="shared/prerequisites.md"

@import src="README.md" format="markdown" profile="gfm"

@import src="api/openapi.yaml" format="openapi"

src is resolved relative to the containing file. Resolution must remain inside the configured project root unless the project explicitly permits an additional root. Symlinks must be resolved before this check.

Remote imports are disabled by default. A remote adapter must pin immutable content by digest and must not retrieve content during an offline build.

The parser creates an import node without opening the target. Import resolution is a separate build phase. This keeps parsing deterministic and permits caching.

An adapter returns the same document node model as the native parser. Every imported node records its source format, source file, byte range when available, and adapter version.

An include cycle is an error. Implementations must report the complete cycle.

Import modes

The optional mode attribute has these values:

Markdown interchange

Markdown is a required adapter. MDX is not.

Version 0.1 defines these profiles:

• commonmark-0.31.2;
• gfm for tables, task lists, strikethrough, and automatic links; and
• mpress-markdown-1 for the complete MPress Markdown component grammar, footnotes, typographic substitutions, and page frontmatter.

The default profile is mpress-markdown-1.

Import guarantee

The adapter must retain every byte of a Markdown source file. It must never discard an unknown extension or an unsupported construct.

Known constructs become typed document nodes. Raw HTML becomes a raw HTML node. MDX-like syntax becomes an opaque Markdown node and a diagnostic. It is never executed.

If a construct is valid Markdown but has no native M-Press Flavoured Markdown equivalent, the adapter creates an opaque-markdown node containing its exact source bytes. This is the lossless fallback, not an error-recovery shortcut.

Every imported node records:

• the original byte range;
• leading and trailing trivia;
• the original delimiter spelling;
• the source line endings;
• the Markdown profile; and
• whether the node or any child has changed.

Export guarantee

A Markdown exporter has three modes.

The exporter must generate Markdown, not MDX. It must not emit imports, exports, JSX elements, JavaScript expressions, or framework component calls.

MPress components export with the MPress @component ... @end Markdown extension. A generic Markdown reader therefore sees plain text and the component body rather than executable syntax.

Canonical MPress Markdown retains @metadata[key]. Portable Markdown resolves the reference and writes its plain-text value. Preserve mode retains the original source bytes.

Markdown mapping

Persisted provenance

Source files remain Markdown, so provenance refers directly to the .md file. The parser records the original bytes, their digest, the selected Markdown profile, and node ranges. Canonical formatting does not require a second source format or a sidecar file.

An exporter must verify the source digest before applying preserved ranges. If the digest does not match, it must use canonical mode or stop with an error. It must not claim byte preservation.

Document model

The normative model is an ordered tree of nodes. Every node contains:

• kind;
• source with file, start byte, end byte, start line, and start column;
• ordered attributes;
• ordered children where the kind permits them;
• source provenance when imported; and
• diagnostics attached to that node.

Text content refers to immutable source slices where possible. A parser is not required to allocate one string for every text node.

The core block kinds are document, metadata, paragraph, heading, rule, list, item, quote, code, raw-html, comment, table, table-row, table-cell, footnote, component, import, include, and opaque-markdown.

The core inline kinds are text, soft-break, hard-break, emphasis, strong, code-span, link, image, automatic-link, emoji, footnote-reference, metadata-reference, and inline-role.

Parser requirements

A conforming native parser must:

1. scan input from start to finish without backtracking;
2. use bounded lookahead within the current line;
3. recognise blocks before parsing their inline content;
4. protect opaque code fences before resolving components;
5. resolve components with a stack;
6. resolve inline delimiters with a stack;
7. parse references without consulting later definitions;
8. keep import resolution outside the parser;
9. recover at a blank line or the current container’s @end; and
10. emit source ranges even when a node contains a diagnostic.

For n input bytes and nesting depth d, native parsing must be O(n) time and O(d + r) auxiliary memory, where r is the number of references retained by the document. The document tree itself is not auxiliary memory.

A conforming parser must not require regular expressions, HTML tokenisation, Unicode category lookup, entity lookup, network access, or component-specific parsing to discover the document structure.

Diagnostics and recovery

A diagnostic contains a stable code, severity, message, source range, and an optional repair suggestion.

Required diagnostic codes include:

An unexpected @end becomes literal text after producing a diagnostic. An unclosed ordinary component closes at the end of its containing block. An unclosed raw HTML block consumes the remaining file and produces a diagnostic. No recovery rule may discard source bytes.

Canonical serialization

Canonical M-Press Flavoured Markdown:

• uses LF line endings;
• omits metadata when the document has no metadata and uses schema 1;
• writes --- on the first line when metadata is present;
• writes schema as the first metadata field;
• writes metadata in source order;
• uses one blank line between block siblings;
• uses two spaces for each list depth;
• uses JSON quoting for attribute values;
• writes attributes in source order;
• uses a code fence one backtick longer than the longest potential closing fence in the body, with a minimum length of three;
• uses lowercase Boolean and null values;
• writes explicit @end lines for container components;
• omits @end for leaf components; and
• ends with one line ending.

Canonical output must parse to a document model equivalent to the input model.

Conformance tests

The public conformance suite must include:

• one example for every normative syntax rule;
• malformed and adversarial inputs for every recovery rule;
• deeply nested components up to the documented limit;
• mixed LF, CRLF, and CR Markdown imports;
• all CommonMark 0.31.2 examples;
• the applicable GitHub Flavoured Markdown examples;
• every MPress Markdown component fixture;
• Markdown import followed by preserve export with byte equality;
• Markdown import followed by canonical export and semantic re-import;
• M-Press Flavoured Markdown export to Markdown followed by semantic re-import;
• fuzz tests asserting linear progress and no panics; and
• corpus benchmarks using the complete Wails v3 documentation.

The initial Go implementation should meet these non-normative performance targets on the Wails v3 corpus:

• at least twice Goldmark’s parse throughput for equivalent native content;
• no more than half Goldmark’s parser allocations;
• source-range extraction during the native parse, with no second parse; and
• unchanged generated HTML compared with equivalent MPress Markdown fixtures.

Performance claims must include the corpus revision, Go version, CPU count, sample count, before and after distributions, and generated-output comparison.

Versioning

The metadata schema number changes when existing valid source can acquire a different document meaning or when the component registry changes. A document without metadata is schema 1. Additive node kinds, roles, adapters, and attributes that do not change directive classification do not require a new schema when an older parser can retain them as unknown nodes.

The first-line metadata delimiter and the schema field grammar are stable across schema versions. This allows a parser to select the schema before it parses the document body. A future schema must use the mpress.* namespace for new engine-owned metadata rather than claiming an existing custom key.

During version 0.x, this document can change after implementation experiments. M-Press must not describe a schema as stable until its grammar, Markdown adapter, and conformance suite have shipped together.