fix quadratic complexity in reference parser (#1004)

lib/helpers/parse_link_destination.mjs

@@ -10,7 +10,6 @@ export default function parseLinkDestination (str, start, max) {
   const result = {
     ok: false,
     pos: 0,
-    lines: 0,
     str: ''
   }
 

lib/helpers/parse_link_title.mjs

@@ -3,50 +3,64 @@
 
 import { unescapeAll } from '../common/utils.mjs'
 
-export default function parseLinkTitle (str, start, max) {
-  let code, marker
-  let lines = 0
+// Parse link title within `str` in [start, max] range,
+// or continue previous parsing if `prev_state` is defined (equal to result of last execution).
+//
+export default function parseLinkTitle (str, start, max, prev_state) {
+  let code
   let pos = start
 
-  const result = {
+  const state = {
+    // if `true`, this is a valid link title
     ok: false,
+    // if `true`, this link can be continued on the next line
+    can_continue: false,
+    // if `ok`, it's the position of the first character after the closing marker
     pos: 0,
-    lines: 0,
-    str: ''
+    // if `ok`, it's the unescaped title
+    str: '',
+    // expected closing marker character code
+    marker: 0
   }
 
-  if (pos >= max) { return result }
+  if (prev_state) {
+    // this is a continuation of a previous parseLinkTitle call on the next line,
+    // used in reference links only
+    state.str = prev_state.str
+    state.marker = prev_state.marker
+  } else {
+    if (pos >= max) { return state }
 
-  marker = str.charCodeAt(pos)
+    let marker = str.charCodeAt(pos)
+    if (marker !== 0x22 /* " */ && marker !== 0x27 /* ' */ && marker !== 0x28 /* ( */) { return state }
 
-  if (marker !== 0x22 /* " */ && marker !== 0x27 /* ' */ && marker !== 0x28 /* ( */) { return result }
+    start++
+    pos++
 
-  pos++
+    // if opening marker is "(", switch it to closing marker ")"
+    if (marker === 0x28) { marker = 0x29 }
 
-  // if opening marker is "(", switch it to closing marker ")"
-  if (marker === 0x28) { marker = 0x29 }
+    state.marker = marker
+  }
 
   while (pos < max) {
     code = str.charCodeAt(pos)
-    if (code === marker) {
-      result.pos = pos + 1
-      result.lines = lines
-      result.str = unescapeAll(str.slice(start + 1, pos))
-      result.ok = true
-      return result
-    } else if (code === 0x28 /* ( */ && marker === 0x29 /* ) */) {
-      return result
-    } else if (code === 0x0A) {
-      lines++
+    if (code === state.marker) {
+      state.pos = pos + 1
+      state.str += unescapeAll(str.slice(start, pos))
+      state.ok = true
+      return state
+    } else if (code === 0x28 /* ( */ && state.marker === 0x29 /* ) */) {
+      return state
     } else if (code === 0x5C /* \ */ && pos + 1 < max) {
       pos++
-      if (str.charCodeAt(pos) === 0x0A) {
-        lines++
-      }
     }
 
     pos++
   }
 
-  return result
+  // no closing marker found, but this link title may continue on the next line (for references)
+  state.can_continue = true
+  state.str += unescapeAll(str.slice(start, pos))
+  return state
 }

lib/rules_block/reference.mjs

@@ -1,8 +1,6 @@
 import { isSpace, normalizeReference } from '../common/utils.mjs'
 
 export default function reference (state, startLine, _endLine, silent) {
-  let lines = 0
-
   let pos = state.bMarks[startLine] + state.tShift[startLine]
   let max = state.eMarks[startLine]
   let nextLine = startLine + 1
@@ -12,45 +10,53 @@ export default function reference (state, startLine, _endLine, silent) {
 
   if (state.src.charCodeAt(pos) !== 0x5B/* [ */) { return false }
 
-  // Simple check to quickly interrupt scan on [link](url) at the start of line.
-  // Can be useful on practice: https://github.com/markdown-it/markdown-it/issues/54
-  while (++pos < max) {
-    if (state.src.charCodeAt(pos) === 0x5D /* ] */ &&
-        state.src.charCodeAt(pos - 1) !== 0x5C/* \ */) {
-      if (pos + 1 === max) { return false }
-      if (state.src.charCodeAt(pos + 1) !== 0x3A/* : */) { return false }
-      break
-    }
-  }
-
-  const endLine = state.lineMax
+  function getNextLine (nextLine) {
+    const endLine = state.lineMax
 
-  // jump line-by-line until empty one or EOF
-  const terminatorRules = state.md.block.ruler.getRules('reference')
+    if (nextLine >= endLine || state.isEmpty(nextLine)) {
+      // empty line or end of input
+      return null
+    }
 
-  const oldParentType = state.parentType
-  state.parentType = 'reference'
+    let isContinuation = false
 
-  for (; nextLine < endLine && !state.isEmpty(nextLine); nextLine++) {
     // this would be a code block normally, but after paragraph
     // it's considered a lazy continuation regardless of what's there
-    if (state.sCount[nextLine] - state.blkIndent > 3) { continue }
+    if (state.sCount[nextLine] - state.blkIndent > 3) { isContinuation = true }
 
     // quirk for blockquotes, this line should already be checked by that rule
-    if (state.sCount[nextLine] < 0) { continue }
-
-    // Some tags can terminate paragraph without empty line.
-    let terminate = false
-    for (let i = 0, l = terminatorRules.length; i < l; i++) {
-      if (terminatorRules[i](state, nextLine, endLine, true)) {
-        terminate = true
-        break
+    if (state.sCount[nextLine] < 0) { isContinuation = true }
+
+    if (!isContinuation) {
+      const terminatorRules = state.md.block.ruler.getRules('reference')
+      const oldParentType = state.parentType
+      state.parentType = 'reference'
+
+      // Some tags can terminate paragraph without empty line.
+      let terminate = false
+      for (let i = 0, l = terminatorRules.length; i < l; i++) {
+        if (terminatorRules[i](state, nextLine, endLine, true)) {
+          terminate = true
+          break
+        }
+      }
+
+      state.parentType = oldParentType
+      if (terminate) {
+        // terminated by another block
+        return null
       }
     }
-    if (terminate) { break }
+
+    const pos = state.bMarks[nextLine] + state.tShift[nextLine]
+    const max = state.eMarks[nextLine]
+
+    // max + 1 explicitly includes the newline
+    return state.src.slice(pos, max + 1)
   }
 
-  const str = state.getLines(startLine, nextLine, state.blkIndent, false).trim()
+  let str = state.src.slice(pos, max + 1)
+
   max = str.length
   let labelEnd = -1
 
@@ -62,11 +68,21 @@ export default function reference (state, startLine, _endLine, silent) {
       labelEnd = pos
       break
     } else if (ch === 0x0A /* \n */) {
-      lines++
+      const lineContent = getNextLine(nextLine)
+      if (lineContent !== null) {
+        str += lineContent
+        max = str.length
+        nextLine++
+      }
     } else if (ch === 0x5C /* \ */) {
       pos++
       if (pos < max && str.charCodeAt(pos) === 0x0A) {
-        lines++
+        const lineContent = getNextLine(nextLine)
+        if (lineContent !== null) {
+          str += lineContent
+          max = str.length
+          nextLine++
+        }
       }
     }
   }
@@ -78,7 +94,12 @@ export default function reference (state, startLine, _endLine, silent) {
   for (pos = labelEnd + 2; pos < max; pos++) {
     const ch = str.charCodeAt(pos)
     if (ch === 0x0A) {
-      lines++
+      const lineContent = getNextLine(nextLine)
+      if (lineContent !== null) {
+        str += lineContent
+        max = str.length
+        nextLine++
+      }
     } else if (isSpace(ch)) {
       /* eslint no-empty:0 */
     } else {
@@ -95,11 +116,10 @@ export default function reference (state, startLine, _endLine, silent) {
   if (!state.md.validateLink(href)) { return false }
 
   pos = destRes.pos
-  lines += destRes.lines
 
   // save cursor state, we could require to rollback later
   const destEndPos = pos
-  const destEndLineNo = lines
+  const destEndLineNo = nextLine
 
   // [label]:   destination   'title'
   //                       ^^^ skipping those spaces
@@ -107,7 +127,12 @@ export default function reference (state, startLine, _endLine, silent) {
   for (; pos < max; pos++) {
     const ch = str.charCodeAt(pos)
     if (ch === 0x0A) {
-      lines++
+      const lineContent = getNextLine(nextLine)
+      if (lineContent !== null) {
+        str += lineContent
+        max = str.length
+        nextLine++
+      }
     } else if (isSpace(ch)) {
       /* eslint no-empty:0 */
     } else {
@@ -117,17 +142,25 @@ export default function reference (state, startLine, _endLine, silent) {
 
   // [label]:   destination   'title'
   //                          ^^^^^^^ parse this
-  const titleRes = state.md.helpers.parseLinkTitle(str, pos, max)
+  let titleRes = state.md.helpers.parseLinkTitle(str, pos, max)
+  while (titleRes.can_continue) {
+    const lineContent = getNextLine(nextLine)
+    if (lineContent === null) break
+    str += lineContent
+    pos = max
+    max = str.length
+    nextLine++
+    titleRes = state.md.helpers.parseLinkTitle(str, pos, max, titleRes)
+  }
   let title
 
   if (pos < max && start !== pos && titleRes.ok) {
     title = titleRes.str
     pos = titleRes.pos
-    lines += titleRes.lines
   } else {
     title = ''
     pos = destEndPos
-    lines = destEndLineNo
+    nextLine = destEndLineNo
   }
 
   // skip trailing spaces until the rest of the line
@@ -143,7 +176,7 @@ export default function reference (state, startLine, _endLine, silent) {
       // but it could still be a valid reference if we roll back
       title = ''
       pos = destEndPos
-      lines = destEndLineNo
+      nextLine = destEndLineNo
       while (pos < max) {
         const ch = str.charCodeAt(pos)
         if (!isSpace(ch)) { break }
@@ -174,8 +207,6 @@ export default function reference (state, startLine, _endLine, silent) {
     state.env.references[label] = { title, href }
   }
 
-  state.parentType = oldParentType
-
-  state.line = startLine + lines + 1
+  state.line = nextLine
   return true
 }

test/fixtures/markdown-it/commonmark_extras.txt

@@ -49,6 +49,40 @@ Reference labels: support ligatures (equivalent according to unicode case foldin
 <p><a href="/url">fffifl</a></p>
 .
 
+Reference can be interrupted by other rules
+.
+[foo]: /url 'title
+ - - -
+'
+
+[foo]
+.
+<p>[foo]: /url 'title</p>
+<hr>
+<p>’</p>
+<p>[foo]</p>
+.
+
+Escape character in link reference title doesn't escape newlines
+.
+[foo]: /url "
+hello
+\
+\
+\
+world
+"
+
+[foo]
+.
+<p><a href="/url" title="
+hello
+\
+\
+\
+world
+">foo</a></p>
+.
 
 Issue #35. `<` should work as punctuation
 .