Improve emphasis algorithm

This fixes quadratic complexity in `**<...>**a**<...>**`
pathological case.

CHANGELOG.md

@@ -6,6 +6,14 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
 
+## [12.3.0] - WIP
+### Changed
+- `StateInline.delimiters[].jump` is removed.
+
+### Fixed
+- Fixed quadratic complexity in pathological `***<10k stars>***a***<10k stars>***` case.
+
+
 ## [12.2.0] - 2021-08-02
 ### Added
 - Ordered lists: add order value to token info.
@@ -575,6 +583,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Renamed presets folder (configs -> presets).
 
 
+[12.3.0]: https://github.com/markdown-it/markdown-it/compare/12.2.0...12.3.0
 [12.2.0]: https://github.com/markdown-it/markdown-it/compare/12.1.0...12.2.0
 [12.1.0]: https://github.com/markdown-it/markdown-it/compare/12.0.6...12.1.0
 [12.0.6]: https://github.com/markdown-it/markdown-it/compare/12.0.5...12.0.6

lib/rules_inline/balance_pairs.js

@@ -9,9 +9,28 @@ function processDelimiters(state, delimiters) {
       openersBottom = {},
       max = delimiters.length;
 
+  if (!max) return;
+
+  // headerIdx is the first delimiter of the current (where closer is) delimiter run
+  var headerIdx = 0;
+  var lastTokenIdx = -2; // needs any value lower than -1
+  var jumps = [];
+
   for (closerIdx = 0; closerIdx < max; closerIdx++) {
     closer = delimiters[closerIdx];
 
+    jumps.push(0);
+
+    // markers belong to same delimiter run if:
+    //  - they have adjacent tokens
+    //  - AND markers are the same
+    //
+    if (delimiters[headerIdx].marker !== closer.marker || lastTokenIdx !== closer.token - 1) {
+      headerIdx = closerIdx;
+    }
+
+    lastTokenIdx = closer.token;
+
     // Length is only used for emphasis-specific "rule of 3",
     // if it's not defined (in strikethrough or 3rd party plugins),
     // we can default it to 0 to disable those checks.
@@ -30,14 +49,11 @@ function processDelimiters(state, delimiters) {
 
     minOpenerIdx = openersBottom[closer.marker][(closer.open ? 3 : 0) + (closer.length % 3)];
 
-    openerIdx = closerIdx - closer.jump - 1;
-
-    // avoid crash if `closer.jump` is pointing outside of the array, see #742
-    if (openerIdx < -1) openerIdx = -1;
+    openerIdx = headerIdx - jumps[headerIdx] - 1;
 
     newMinOpenerIdx = openerIdx;
 
-    for (; openerIdx > minOpenerIdx; openerIdx -= opener.jump + 1) {
+    for (; openerIdx > minOpenerIdx; openerIdx -= jumps[openerIdx] + 1) {
       opener = delimiters[openerIdx];
 
       if (opener.marker !== closer.marker) continue;
@@ -67,15 +83,19 @@ function processDelimiters(state, delimiters) {
           // sure algorithm has linear complexity (see *_*_*_*_*_... case).
           //
           lastJump = openerIdx > 0 && !delimiters[openerIdx - 1].open ?
-            delimiters[openerIdx - 1].jump + 1 :
+            jumps[openerIdx - 1] + 1 :
             0;
 
-          closer.jump  = closerIdx - openerIdx + lastJump;
+          jumps[closerIdx] = closerIdx - openerIdx + lastJump;
+          jumps[openerIdx] = lastJump;
+
           closer.open  = false;
           opener.end   = closerIdx;
-          opener.jump  = lastJump;
           opener.close = false;
           newMinOpenerIdx = -1;
+          // treat next token as start of run,
+          // it optimizes skips in **<...>**a**<...>** pathological case
+          lastTokenIdx = -2;
           break;
         }
       }

lib/rules_inline/emphasis.js

@@ -29,15 +29,6 @@ module.exports.tokenize = function emphasis(state, silent) {
       //
       length: scanned.length,
 
-      // An amount of characters before this one that's equivalent to
-      // current one. In plain English: if this delimiter does not open
-      // an emphasis, neither do previous `jump` characters.
-      //
-      // Used to skip sequences like "*****" in one step, for 1st asterisk
-      // value will be 0, for 2nd it's 1 and so on.
-      //
-      jump:   i,
-
       // A position of the token this delimiter corresponds to.
       //
       token:  state.tokens.length - 1,
@@ -91,9 +82,11 @@ function postProcess(state, delimiters) {
     //
     isStrong = i > 0 &&
                delimiters[i - 1].end === startDelim.end + 1 &&
+               // check that first two markers match and adjacent
+               delimiters[i - 1].marker === startDelim.marker &&
                delimiters[i - 1].token === startDelim.token - 1 &&
-               delimiters[startDelim.end + 1].token === endDelim.token + 1 &&
-               delimiters[i - 1].marker === startDelim.marker;
+               // check that last two markers are adjacent (we can safely assume they match)
+               delimiters[startDelim.end + 1].token === endDelim.token + 1;
 
     ch = String.fromCharCode(startDelim.marker);
 

lib/rules_inline/strikethrough.js

@@ -33,7 +33,6 @@ module.exports.tokenize = function strikethrough(state, silent) {
     state.delimiters.push({
       marker: marker,
       length: 0,     // disable "rule of 3" length checks meant for emphasis
-      jump:   i / 2, // for `~~` 1 marker = 2 characters
       token:  state.tokens.length - 1,
       end:    -1,
       open:   scanned.can_open,

test/pathological.js

@@ -57,6 +57,10 @@ describe('Pathological sequences speed', () => {
       );
     });
 
+    it('nested inlines', async () => {
+      await test_pattern('*'.repeat(60000) + 'a' + '*'.repeat(60000));
+    });
+
     it('nested strong emph', async () => {
       await test_pattern('*a **a '.repeat(5000) + 'b' + ' a** a*'.repeat(5000));
     });

test/pathological.json

@@ -1 +1 @@
-{ "md5": "10730e0b6dec7355412bb632e9eb1d98" }
+{ "md5": "c417101e12950cc61ff0a6d2cebb80e0" }