mirror of
https://github.com/hedgedoc/hedgedoc.git
synced 2024-11-27 12:08:02 -05:00
c59b94a37b
We can load the xss functions directly from the library instead of loading them through the expose loader of webpack, this should simplify the setup and maybe even improve speed a bit. Signed-off-by: Sheogorath <sheogorath@shivering-isles.com>
355 lines
12 KiB
JavaScript
355 lines
12 KiB
JavaScript
/* eslint-env browser, jquery */
|
|
|
|
import { preventXSS, escapeAttrValue } from './render'
|
|
import { md } from './extra'
|
|
|
|
/**
|
|
* The reveal.js markdown plugin. Handles parsing of
|
|
* markdown inside of presentations as well as loading
|
|
* of external markdown documents.
|
|
*/
|
|
(function (root, factory) {
|
|
if (typeof exports === 'object') {
|
|
module.exports = factory()
|
|
} else {
|
|
// Browser globals (root is window)
|
|
root.RevealMarkdown = factory()
|
|
root.RevealMarkdown.initialize()
|
|
}
|
|
}(this, function () {
|
|
var DEFAULT_SLIDE_SEPARATOR = '^\r?\n---\r?\n$'
|
|
var DEFAULT_NOTES_SEPARATOR = '^note:'
|
|
var DEFAULT_ELEMENT_ATTRIBUTES_SEPARATOR = '\\.element\\s*?(.+?)$'
|
|
var DEFAULT_SLIDE_ATTRIBUTES_SEPARATOR = '\\.slide:\\s*?(\\S.+?)$'
|
|
|
|
var SCRIPT_END_PLACEHOLDER = '__SCRIPT_END__'
|
|
|
|
/**
|
|
* Retrieves the markdown contents of a slide section
|
|
* element. Normalizes leading tabs/whitespace.
|
|
*/
|
|
function getMarkdownFromSlide (section) {
|
|
var template = section.querySelector('script')
|
|
|
|
// strip leading whitespace so it isn't evaluated as code
|
|
var text = (template || section).textContent
|
|
|
|
// restore script end tags
|
|
text = text.replace(new RegExp(SCRIPT_END_PLACEHOLDER, 'g'), '</script>')
|
|
|
|
var leadingWs = text.match(/^\n?(\s*)/)[1].length
|
|
var leadingTabs = text.match(/^\n?(\t*)/)[1].length
|
|
|
|
if (leadingTabs > 0) {
|
|
text = text.replace(new RegExp('\\n?\\t{' + leadingTabs + '}', 'g'), '\n')
|
|
} else if (leadingWs > 1) {
|
|
text = text.replace(new RegExp('\\n? {' + leadingWs + '}', 'g'), '\n')
|
|
}
|
|
|
|
return text
|
|
}
|
|
|
|
/**
|
|
* Given a markdown slide section element, this will
|
|
* return all arguments that aren't related to markdown
|
|
* parsing. Used to forward any other user-defined arguments
|
|
* to the output markdown slide.
|
|
*/
|
|
function getForwardedAttributes (section) {
|
|
var attributes = section.attributes
|
|
var result = []
|
|
|
|
for (var i = 0, len = attributes.length; i < len; i++) {
|
|
var name = attributes[i].name
|
|
var value = attributes[i].value
|
|
|
|
// disregard attributes that are used for markdown loading/parsing
|
|
if (/data-(markdown|separator|vertical|notes)/gi.test(name)) continue
|
|
|
|
if (value) {
|
|
result.push(name + '="' + value + '"')
|
|
} else {
|
|
result.push(name)
|
|
}
|
|
}
|
|
|
|
return result.join(' ')
|
|
}
|
|
|
|
/**
|
|
* Inspects the given options and fills out default
|
|
* values for what's not defined.
|
|
*/
|
|
function getSlidifyOptions (options) {
|
|
options = options || {}
|
|
options.separator = options.separator || DEFAULT_SLIDE_SEPARATOR
|
|
options.notesSeparator = options.notesSeparator || DEFAULT_NOTES_SEPARATOR
|
|
options.attributes = options.attributes || ''
|
|
|
|
return options
|
|
}
|
|
|
|
/**
|
|
* Helper function for constructing a markdown slide.
|
|
*/
|
|
function createMarkdownSlide (content, options) {
|
|
options = getSlidifyOptions(options)
|
|
|
|
var notesMatch = content.split(new RegExp(options.notesSeparator, 'mgi'))
|
|
|
|
if (notesMatch.length === 2) {
|
|
content = notesMatch[0] + '<aside class="notes" data-markdown>' + notesMatch[1].trim() + '</aside>'
|
|
}
|
|
|
|
// prevent script end tags in the content from interfering
|
|
// with parsing
|
|
content = content.replace(/<\/script>/g, SCRIPT_END_PLACEHOLDER)
|
|
|
|
return '<script type="text/template">' + content + '</script>'
|
|
}
|
|
|
|
/**
|
|
* Parses a data string into multiple slides based
|
|
* on the passed in separator arguments.
|
|
*/
|
|
function slidify (markdown, options) {
|
|
options = getSlidifyOptions(options)
|
|
|
|
var separatorRegex = new RegExp(options.separator + (options.verticalSeparator ? '|' + options.verticalSeparator : ''), 'mg')
|
|
var horizontalSeparatorRegex = new RegExp(options.separator)
|
|
|
|
var matches
|
|
var lastIndex = 0
|
|
var isHorizontal
|
|
var wasHorizontal = true
|
|
var content
|
|
var sectionStack = []
|
|
|
|
// iterate until all blocks between separators are stacked up
|
|
while ((matches = separatorRegex.exec(markdown)) !== null) {
|
|
// determine direction (horizontal by default)
|
|
isHorizontal = horizontalSeparatorRegex.test(matches[0])
|
|
|
|
if (!isHorizontal && wasHorizontal) {
|
|
// create vertical stack
|
|
sectionStack.push([])
|
|
}
|
|
|
|
// pluck slide content from markdown input
|
|
content = markdown.substring(lastIndex, matches.index)
|
|
|
|
if (isHorizontal && wasHorizontal) {
|
|
// add to horizontal stack
|
|
sectionStack.push(content)
|
|
} else {
|
|
// add to vertical stack
|
|
sectionStack[sectionStack.length - 1].push(content)
|
|
}
|
|
|
|
lastIndex = separatorRegex.lastIndex
|
|
wasHorizontal = isHorizontal
|
|
}
|
|
|
|
// add the remaining slide
|
|
(wasHorizontal ? sectionStack : sectionStack[sectionStack.length - 1]).push(markdown.substring(lastIndex))
|
|
|
|
var markdownSections = ''
|
|
|
|
// flatten the hierarchical stack, and insert <section data-markdown> tags
|
|
for (var i = 0, len = sectionStack.length; i < len; i++) {
|
|
// vertical
|
|
if (sectionStack[i] instanceof Array) {
|
|
markdownSections += '<section ' + options.attributes + '>'
|
|
|
|
sectionStack[i].forEach(function (child) {
|
|
markdownSections += '<section data-markdown>' + createMarkdownSlide(child, options) + '</section>'
|
|
})
|
|
|
|
markdownSections += '</section>'
|
|
} else {
|
|
markdownSections += '<section ' + options.attributes + ' data-markdown>' + createMarkdownSlide(sectionStack[i], options) + '</section>'
|
|
}
|
|
}
|
|
|
|
return markdownSections
|
|
}
|
|
|
|
/**
|
|
* Parses any current data-markdown slides, splits
|
|
* multi-slide markdown into separate sections and
|
|
* handles loading of external markdown.
|
|
*/
|
|
function processSlides () {
|
|
var sections = document.querySelectorAll('[data-markdown]')
|
|
var section
|
|
|
|
for (var i = 0, len = sections.length; i < len; i++) {
|
|
section = sections[i]
|
|
|
|
if (section.getAttribute('data-markdown').length) {
|
|
var xhr = new XMLHttpRequest()
|
|
var url = section.getAttribute('data-markdown')
|
|
|
|
var datacharset = section.getAttribute('data-charset')
|
|
|
|
// see https://developer.mozilla.org/en-US/docs/Web/API/element.getAttribute#Notes
|
|
if (datacharset !== null && datacharset !== '') {
|
|
xhr.overrideMimeType('text/html; charset=' + datacharset)
|
|
}
|
|
|
|
xhr.onreadystatechange = function () {
|
|
if (xhr.readyState === 4) {
|
|
// file protocol yields status code 0 (useful for local debug, mobile applications etc.)
|
|
if ((xhr.status >= 200 && xhr.status < 300) || xhr.status === 0) {
|
|
section.outerHTML = slidify(xhr.responseText, {
|
|
separator: section.getAttribute('data-separator'),
|
|
verticalSeparator: section.getAttribute('data-separator-vertical'),
|
|
notesSeparator: section.getAttribute('data-separator-notes'),
|
|
attributes: getForwardedAttributes(section)
|
|
})
|
|
} else {
|
|
section.outerHTML = '<section data-state="alert">' +
|
|
'ERROR: The attempt to fetch ' + url + ' failed with HTTP status ' + xhr.status + '.' +
|
|
'Check your browser\'s JavaScript console for more details.' +
|
|
'<p>Remember that you need to serve the presentation HTML from a HTTP server.</p>' +
|
|
'</section>'
|
|
}
|
|
}
|
|
}
|
|
|
|
xhr.open('GET', url, false)
|
|
|
|
try {
|
|
xhr.send()
|
|
} catch (e) {
|
|
alert('Failed to get the Markdown file ' + url + '. Make sure that the presentation and the file are served by a HTTP server and the file can be found there. ' + e)
|
|
}
|
|
} else if (section.getAttribute('data-separator') || section.getAttribute('data-separator-vertical') || section.getAttribute('data-separator-notes')) {
|
|
section.outerHTML = slidify(getMarkdownFromSlide(section), {
|
|
separator: section.getAttribute('data-separator'),
|
|
verticalSeparator: section.getAttribute('data-separator-vertical'),
|
|
notesSeparator: section.getAttribute('data-separator-notes'),
|
|
attributes: getForwardedAttributes(section)
|
|
})
|
|
} else {
|
|
section.innerHTML = createMarkdownSlide(getMarkdownFromSlide(section))
|
|
}
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Check if a node value has the attributes pattern.
|
|
* If yes, extract it and add that value as one or several attributes
|
|
* the the terget element.
|
|
*
|
|
* You need Cache Killer on Chrome to see the effect on any FOM transformation
|
|
* directly on refresh (F5)
|
|
* http://stackoverflow.com/questions/5690269/disabling-chrome-cache-for-website-development/7000899#answer-11786277
|
|
*/
|
|
function addAttributeInElement (node, elementTarget, separator) {
|
|
var mardownClassesInElementsRegex = new RegExp(separator, 'mg')
|
|
var mardownClassRegex = new RegExp('([^"= ]+?)="([^"=]+?)"', 'mg')
|
|
var nodeValue = node.nodeValue
|
|
var matches
|
|
var matchesClass
|
|
if ((matches = mardownClassesInElementsRegex.exec(nodeValue))) {
|
|
var classes = matches[1]
|
|
nodeValue = nodeValue.substring(0, matches.index) + nodeValue.substring(mardownClassesInElementsRegex.lastIndex)
|
|
node.nodeValue = nodeValue
|
|
while ((matchesClass = mardownClassRegex.exec(classes))) {
|
|
var name = matchesClass[1]
|
|
var value = matchesClass[2]
|
|
if (name.substr(0, 5) === 'data-' || window.whiteListAttr.indexOf(name) !== -1) { elementTarget.setAttribute(name, escapeAttrValue(value)) }
|
|
}
|
|
return true
|
|
}
|
|
return false
|
|
}
|
|
|
|
/**
|
|
* Add attributes to the parent element of a text node,
|
|
* or the element of an attribute node.
|
|
*/
|
|
function addAttributes (section, element, previousElement, separatorElementAttributes, separatorSectionAttributes) {
|
|
if (element != null && element.childNodes !== undefined && element.childNodes.length > 0) {
|
|
var previousParentElement = element
|
|
for (var i = 0; i < element.childNodes.length; i++) {
|
|
var childElement = element.childNodes[i]
|
|
if (i > 0) {
|
|
let j = i - 1
|
|
while (j >= 0) {
|
|
var aPreviousChildElement = element.childNodes[j]
|
|
if (typeof aPreviousChildElement.setAttribute === 'function' && aPreviousChildElement.tagName !== 'BR') {
|
|
previousParentElement = aPreviousChildElement
|
|
break
|
|
}
|
|
j = j - 1
|
|
}
|
|
}
|
|
var parentSection = section
|
|
if (childElement.nodeName === 'section') {
|
|
parentSection = childElement
|
|
previousParentElement = childElement
|
|
}
|
|
if (typeof childElement.setAttribute === 'function' || childElement.nodeType === Node.COMMENT_NODE) {
|
|
addAttributes(parentSection, childElement, previousParentElement, separatorElementAttributes, separatorSectionAttributes)
|
|
}
|
|
}
|
|
}
|
|
|
|
if (element.nodeType === Node.COMMENT_NODE) {
|
|
if (addAttributeInElement(element, previousElement, separatorElementAttributes) === false) {
|
|
addAttributeInElement(element, section, separatorSectionAttributes)
|
|
}
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Converts any current data-markdown slides in the
|
|
* DOM to HTML.
|
|
*/
|
|
function convertSlides () {
|
|
var sections = document.querySelectorAll('[data-markdown]')
|
|
|
|
for (var i = 0, len = sections.length; i < len; i++) {
|
|
var section = sections[i]
|
|
|
|
// Only parse the same slide once
|
|
if (!section.getAttribute('data-markdown-parsed')) {
|
|
section.setAttribute('data-markdown-parsed', true)
|
|
|
|
var notes = section.querySelector('aside.notes')
|
|
var markdown = getMarkdownFromSlide(section)
|
|
markdown = markdown.replace(/</g, '<').replace(/>/g, '>')
|
|
var rendered = md.render(markdown)
|
|
rendered = preventXSS(rendered)
|
|
var result = window.postProcess(rendered)
|
|
section.innerHTML = result[0].outerHTML
|
|
addAttributes(section, section, null, section.getAttribute('data-element-attributes') ||
|
|
section.parentNode.getAttribute('data-element-attributes') ||
|
|
DEFAULT_ELEMENT_ATTRIBUTES_SEPARATOR,
|
|
section.getAttribute('data-attributes') ||
|
|
section.parentNode.getAttribute('data-attributes') ||
|
|
DEFAULT_SLIDE_ATTRIBUTES_SEPARATOR)
|
|
|
|
// If there were notes, we need to re-add them after
|
|
// having overwritten the section's HTML
|
|
if (notes) {
|
|
section.appendChild(notes)
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// API
|
|
return {
|
|
initialize: function () {
|
|
processSlides()
|
|
convertSlides()
|
|
},
|
|
// TODO: Do these belong in the API?
|
|
processSlides: processSlides,
|
|
convertSlides: convertSlides,
|
|
slidify: slidify
|
|
}
|
|
}))
|