Fossil

The primary purpose of a footnote is to provide a reader with supplementary information in an unobtrusive way, so that the reader's attention is not distracted from the main line of a narrative. Thus a natural length of a footnote is usually a couple of sentences, maybe a paragraph. If a footnote exceeds a paragraph then it may be beneficial to rearrange a composition and incorporate that footnote as a subsection of the current (or some other) document and point there using a regular hyperlink.

There may be (at least) two ways to think of footnotes:

1. From the viewpoint of a document's author (i.e. Markdown syntax):

   Within this point of view a footnote is defined either in the place where the corresponding numeric marker is inserted or at some other location in the document (e.g. after a paragraph or at the end of a document's source) and referenced by the corresponding label.
   The former will be called as inline while the later as referenced (or labeled).

2. From the viewpoint of a person who reads a rendered document:

   Within this point of view either a footnote is explicitly associated with a specific text or it is the task of the reader to deduce an exact phrase to which that footnote applies. The former will be called span-bounded (or fragment-bounded) and the later as free-standing.

Hence there are four types of footnotes (suggestions for better naming are welcomed), and this branch implements all of them. Footnotes rendering is supported in all places where Markdown is supported (but see issues and limitations).

The syntax

Proposed syntax is documented along with the other markdown rules.¹ It was desired to have a syntax that naturally extends conventional Markdown and which is consistent for all four types of footnotes.²

Free-standing labeled footnotes

It seems that by "footnotes in Markdown" Internet typically means referenced free-standing case. The syntax for that is more or less widespread and settled as:

  Lorem ipsum dolor sit amet, consectetur adipisicing elit, sed doeiusmod
  tempor incididunt[^label] ut labore et dolore magna aliqua.

  [^label]: Ut enimad minim veniam, quis nostrud exercitation ullamco
     laboris nisi utaliquip ex ea commodo consequat.

The advantage of referenced footnotes is that a single footnote may be referenced multiple times and the place(s) of use may appear before the footnote's definition.

Free-standing inline footnotes

These may be more convenient for simple cases and as noted in the forum might be easier in maintenance.

For the time being, free markdown processors that support inline footnotes are unknown. The following variants were considered:

  1.  Lorem ipsum dolor sit amet(^Ut enimad minim veniam)
   
  2.  Lorem ipsum dolor sit amet[^Ut enimad minim veniam]
   
  3.  Lorem ipsum dolor sit amet^[Ut enimad minim veniam]
   
  4.  Lorem ipsum dolor sit amet^(Ut enimad minim veniam)

The first variant was chosen since it looks a bit more natural in the source form.

Span-bounded footnotes

These seems to be a pretty rare thing.
For referenced footnotes the following syntax variants were considered:

  1.  Lorem ipsum [dolor sit amet][^label]
   
  2.  Lorem ipsum [dolor sit amet](^label)
  
  3.  Lorem ipsum [dolor sit amet]^[label]
  
  4.  Lorem ipsum [dolor sit amet]^(label)

and for inline footnotes the following:

  1.  Lorem ipsum [dolor sit amet](^Ut enimad minim veniam)
  
  2.  Lorem ipsum [dolor sit amet][^Ut enimad minim veniam]
  
  3.  Lorem ipsum [dolor sit amet]^[Ut enimad minim veniam]
  
  4.  Lorem ipsum [dolor sit amet]^(Ut enimad minim veniam)

In the examples above both the scope of the footnote and its content are pretty short. That might become the typical case but the parser is capable of a longer snippets that span several lines (albeit not paragraphs).

In all cases the scope is specified inside square brackets because it is consistent with the syntax for regular links. The first variant in each category was selected as it seems consistent with the syntax for regular hyperlinks and simplifies implementation.

The corresponding text fragment of a span-bounded footnote is highlighted when a user follows footnote's back-reference or when that span is hovered over.

Styling with user-provided classes

If a footnote's text starts with a token of the special form then this token is used to derive a set of CSS classes that are added to that footnote and its references. This enables users to style elements of a particular footnote provided that the administrator provisioned and documented some special CSS classes in a custom skin. Default skin does not provide any of such special classes.

The token must start with a dot and must end with a colon; in between of these it must be a dot-separated list of words; each word may contain only ASCII alphanumeric characters and hyphens.

Linting

The numbers for misreferences, unreferenced footnotes and joined footnotes (that have several definitions with the same label) are counted.

If any of these counter is non-zero then TH1 variable $footnotes_issues_counters is set to the space separated list of corresponding integers. This simplifies reporting about issues with footnotes from within a header of a page (if such warning is provisioned in a custom skin).

Also --lint-footnotes option is added to the test-markdown-render command³. If this flag is given and footnotes in the input have issues, then the above-mentioned counters are printed to stderr and non-zero exit code is set.

Known issues and limitations

1. There is an issue for webpages where <base href="..."> is inconsistent with the actual REQUEST_URI of a page. As of this writing preview tabs of the /wikiedit and /fileedit pages are affected. There was an attempt to fix base href but as of this writing it's not merged onto mainline. An attempt to fix it by including RQUEST_URI into the generated hyperlinks does not seem to help for these AJAXified pages.
   There is a hope to solve this issue before version 2.19 of Fossil.

2. A footnote's text is parsed and rendered in "inline mode". This is what is needed for the usual case, but it means that block-level markup might not work inside of a footnote. If the "block-level" markup is desired then it can be done via HTML tags in the text of a footnote.

3. Source text of an inline footnote may not contain blank lines. This limitation comes from the current architecture of the parser where paragraphs are identified before the links (and footnotes alike).
   That does not seem like a big problem in the light of the second point and the provisioned purpose of footnotes (described in the beginning).

Notes about implementation

A static integer counter is incremented upon each rendering of a Markdown document within a single process execution. This enables to generate unique IDs for footnotes and back-references within a single webpage. The counter is incremented for each invocation of markdown_to_html() in the hope for better durability of hyperlinks that point to a footnote (or a footnote's marker) within a particular post of the forum.

Counters for issues are global variables (g.ftntsIssues). The other options were considered as over-complication.

Security considerations

This implementation adds three places (above the ordinary Markdown processing) where the user input is used to construct HTML.

1. REQUEST_URI is escaped via newly added escape_quotes() function before getting into the values of href attributes.

2. Labels and source code of unreferenced footnotes are processed in the same way as other code blocks in Markdown source (via html_escape() function).

3. User provided classes are constrained to ASCII alphanumeric characters and hyphens.

Thus it is believed that this feature does not introduce vulnerability for HTML injection.

Performance considerations

Footnotes processing always terminates. The amount of CPU required for the processing of typical footnotes should be on par with the rest of the Markdown processor. The worst theoretical case constitutes of a long chain of recursively nested inline footnotes. For this case the theoretical complexity is O(n²). Thus to keep CPU consumption in bounds the maximal depth of nesting that is considered is limited to 5.

See also:

Wikipedia
GitHub
StackExchange
https://daringfireball.net/2005/07/footnotes
https://www.markdownguide.org/extended-syntax/
https://support.typora.io/Markdown-Reference/#footnotes
https://michelf.ca/projects/php-markdown/extra/#footnotes
https://rephrase.net/box/word/footnotes/syntax/

Footnotes

1. ^{^} If this version of Fossil supports⁴ footnotes then the actual syntax should be documented at /md_rules.
2. ^{^} Some examples (albeit unusual) may be seen at /doc/markdown-footnotes/test/markdown-test3.md
3. ^{^} A link to the corresponding check-in should also test extraction of backlinks from within footnotes.
4. ^{^} These four footnotes should test the functionality of footnotes on Wiki pages.