package cmarkit

  1. Overview
  2. Docs

Rendering CommonMark to CommonMark.

Generates CommonMark. If your document was parsed with layout:true, it preserves most of the source layout on output. This won't be perfect, make sure you understand the details before reporting issues.

See an example.

Warning. Rendering outputs are unstable. They may be tweaked even between minor versions of the library.

Rendering

val of_doc : Cmarkit.Doc.t -> string

of_doc d is a CommonMark document for d. See renderer for more details.

Renderer

val renderer : unit -> Cmarkit_renderer.t

renderer () is the default CommonMark renderer. This renders the strict CommonMark abstract syntax tree and the supported Cmarkit extensions.

The inline, block and document renderers always return true. Unknown block and inline values are rendered by an HTML comment (as permitted by the CommonMark specification).

See this example to extend or selectively override the renderer.

Render functions

Only useful if you extend the renderer.

Newlines and indentation

val newline : Cmarkit_renderer.context -> unit

newline c starts a new line, except on the first call on c which is a nop.

type indent = [
  1. | `I of int
    (*

    Identation by given amount.

    *)
  2. | `L of int * string * int * Uchar.t option
    (*

    Indent before, list marker, indent after, list item task extension

    *)
  3. | `Q of int
    (*

    Identation followed by a block quote marker and a space

    *)
  4. | `Fn of int * Cmarkit.Label.t
    (*

    Indent before, label (footnote extension)

    *)
]

The type for specifying block indentation.

val push_indent : Cmarkit_renderer.context -> indent -> unit

push_indent c i pushes i on the current indentation of c. This does not render anything.

val pop_indent : Cmarkit_renderer.context -> unit

pop_indent c pops the last indentation pushed on c. This does not render anything.

val indent : Cmarkit_renderer.context -> unit

indent i c outputs current indentation on c. Note that `L and `Fn get replaced by an `I indent on subsequent lines, that is the list or foonote marker is output only once.

Backslash escaping

module Char_set : Set.S with type elt = char

Sets of US-ASCII characters.

val escaped_string : ?esc_ctrl:bool -> Cmarkit_renderer.context -> Char_set.t -> string -> unit

escaped_string ?esc_ctrl c cs s renders s on c with characters in cs backslash escaped. If esc_ctrl is true (default) ASCII control characters are escaped to decimal escapes.

val buffer_add_escaped_string : ?esc_ctrl:bool -> Buffer.t -> Char_set.t -> string -> unit

buffer_add_escaped_string b cs s is escaped_string but appends to a buffer value.

val escaped_text : Cmarkit_renderer.context -> string -> unit

escaped_text c s renders s on c trying to be smart about escaping Commonmark structural symbols for Cmarkit.Inline.Text inlines. We assume text can be anywhere in a sequence of inlines and in particular that it can start a line. This function also takes into account the existence of the extensions.

As such we escape:

  • These block markers: -+_= only if present at s.[0].
  • Only the first of runs of them: #`
  • Only the first of a run longer than 1: ~ (strikethrough extension).
  • & if followed by an US-ASCII letter or #.
  • ! if it is the last character of s.
  • . or ) only if preceeded by a single 1 and zero or more 0 to the start of text.
  • Everywhere, *_\ <>[], ASCII control characters, $ (inline math extension), | (table extension)
val buffer_add_escaped_text : Buffer.t -> string -> unit

buffer_add_escaped_text b s is escaped_text but appends to a buffer value.

Source layout preservation

The abstract syntax tree has a few block cases and data fields to represent the source document layout. This allows to update CommonMark documents without normalizing them too much when they are parsed with layout:true.

To keep things reasonably simple a few things are not attempted like:

  1. Preserving entities and character references.
  2. Preserving the exact line by line indentation layout of container blocks.
  3. Preserving lazy continuation lines.
  4. Keeping track of used newlines except for the first one.
  5. Preserving layout source location information when it can be reconstructed from the document data source location.

In general we try to keep the following desirable properties for the abstract syntax tree definition:

  1. Layout information should not interfere with document data or be affected by it. Otherwise data updates also needs to update the layout data, which is error prone and unconvenient.
  2. Abstract syntax trees resulting from the source document, from renders of the source document parsed with or without layout:tree should all render to the same HTML.

In practice CommonMark being not context free point 1. is not always achieved. In particular in Cmarkit.Inline.Code_span the number of delimiting backticks depends on the code content (Cmarkit.Inline.Code_span.of_string, computes that for you).

The renderer performs almost no checks on the layout data. You should be careful if you fill these yourself since you could generate CommonMark that will be misinterpreted. Layout data of pristine nodes coming out of Cmarkit.Doc.of_string, created with the Cmarkit.Inline and Cmarkit.Block constructors should not need your attention (respect their input constraints though).

Classifying renderings

We say that a CommonMark render:

  • is correct, if the result renders the same HTML as the source document. This can be checked with the cmarkit tool included in the distribution:

    cmarkit commonmark --html-diff mydoc.md

    If a difference shows up, the rendering is said to be incorrect.

  • round trips, if the result is byte-for-byte equal to the source document. This can be checked with the cmarkit tool included in the distribution:

    cmarkit commonmark --diff mydoc.md

    If a difference shows up, the rendering does not round trip but it may still be correct.

Known correct differences

In general lack of round trip is due to:

  • Loss of layout on input (see above).
  • Eager escaping of CommonMark delimiters (the escape strategy is here).
  • Churn around blank lines which can be part of blocks without adhering to their structural convention.

Please do not report issues for differences that are due to the following:

  1. Source US-ASCII control characters in textual data render as decimal character references in the output.
  2. Source entity and character references are lost during parsing and thus replaced by their definition in the output.
  3. Source tab stop advances may be replaced by spaces in the output.
  4. Source escaped characters may end up unescaped in the output.
  5. Source unescaped characters may end up escaped in the output.
  6. Source lazy continuation lines are made part of blocks in the output.
  7. Source indented blank lines following indented code blocks lose four spaces of indentation (as per specification these are not part of the block).
  8. Source unindented blank lines in indented code blocks are indented in the output.
  9. Source fenced code block indentation is retained from the opening fence and used for the following lines in the output.
  10. Source block quote indentation is retained from the first line and used for the following lines in the output. The optional space following the quotation mark '>' is made mandatory.
  11. Source list item indentation is regularized, in particular blank lines will indent.
  12. Source list item that start with an empty line get a space after their marker.
  13. The newline used in the output is the one found in the rendered Cmarkit.Doc.t value.

Simple and implemented round trip improvements to the renderer are welcome.

Known incorrect renderings

Please do not report issues incorrect renderings that are due to the following (and unlikely to be fixed):

  1. Use of entities and character references around structural CommonMark symbols can make things go wrong. These get resolved after inline parsing because they can't be used to stand for structural CommonMark symbols, however once they have been resolved they can interact with parsing. Here is an example:

    *emph&nbsp;*

    It parses as emphasis. But if we render it to CommonMark non-breaking space renders as is and we get:

    *emph *

    which no longer parses as emphasis.

    Note in this particular case it is possible to do something about it by being smarter about the context when escaping. However there's a trade-off between renderer complexity and the (conjectured) paucity of these cases.

Otherwise, if you spot an incorrect rendering please report a minimal reproduction case.

Simple and implemented round trip improvements to the renderer are welcome.

OCaml

Innovation. Community. Security.