|
| 1 | +title: $:/changenotes/5.4.0/#9676 |
| 2 | +description: Fix critical freelinks bugs: first character loss and false positive matches in v5.4.0 |
| 3 | +release: 5.4.0 |
| 4 | +tags: $:/tags/ChangeNote |
| 5 | +change-type: bugfix |
| 6 | +change-category: plugin |
| 7 | +github-links: https://github.com/TiddlyWiki/TiddlyWiki5/pull/9084 https://github.com/TiddlyWiki/TiddlyWiki5/pull/9397 https://github.com/TiddlyWiki/TiddlyWiki5/pull/9676 |
| 8 | +github-contributors: s793016 |
| 9 | + |
| 10 | +Fixes and optimizations to the Freelinks plugin's Aho-Corasick implementation following #9397. |
| 11 | + |
| 12 | +Fixes: |
| 13 | +* Failure Links Non-Functional (Critical): The failure link map used a plain object `{}` with trie nodes as keys. Since all JavaScript objects coerce to the same string `[object Object]`, every node resolved to the same map entry. Failure links were silently broken for all overlapping patterns. Fixed by replacing with `WeakMap`. |
| 14 | +* Cache Rebuilt on Every UI Interaction (Performance): Any `$:/state/...` update (e.g. clicking tabs) would trigger a full Aho-Corasick rebuild, causing severe lag on large wikis. The `refresh` logic now ignores system tiddlers, with an explicit allowlist for plugin config tiddlers. |
| 15 | +* Short Match Blocking Longer Match: A shorter title appearing earlier (e.g. "The New") could prevent a longer overlapping title (e.g. "New York City") from matching. Replaced left-to-right greedy selection with global length-first sorting and interval occupation tracking. |
| 16 | +* Unicode Index Desync in ignoreCase Mode: Calling `toLowerCase()` on the full text before searching could change string length (e.g. Turkish "İ" expands), causing `substring()` to split Emoji surrogate pairs and produce garbage output. Case conversion is now done per-character during search. |
| 17 | +* Removed Vestigial Regex Escaping: `escapeRegExp()` was called during trie construction but Aho-Corasick operates on literal character transitions, not regex. Removed. |
| 18 | + |
| 19 | +Impact: |
| 20 | +* Overlapping titles now match correctly for the first time. |
| 21 | +* No cache rebuilds during normal UI interactions on large wikis. |
| 22 | +* Correct longest-match behavior for titles sharing substrings. |
| 23 | +* Safe Emoji and complex Unicode handling in case-insensitive mode. |
| 24 | + |
| 25 | + |
| 26 | +#9397 |
| 27 | +This note addresses two major bugs introduced in the Freelinks plugin with the v5.4.0 release: |
| 28 | + |
| 29 | +Fixes: |
| 30 | +* First Character Loss: The first character of a matched word would incorrectly disappear (e.g., "The" became "he"). This was fixed by correctly timing the filtering of the current tiddler's title during match validation, ensuring proper substring handling. |
| 31 | +* False Positive Matches: Unrelated words (like "it is" or "Choose") would incorrectly link to a tiddler title. This was resolved by fixing wrong output merging in the Aho-Corasick failure-link handling, eliminating spurious matches from intermediate nodes, and adding cycle detection. |
| 32 | + |
| 33 | +Impact: |
| 34 | +* Significantly improved correctness and reliability of automatic linking for all users, especially in multilingual and large wikis. |
| 35 | + |
| 36 | + |
| 37 | +#9084 |
| 38 | +This change introduces a fully optimized override of the core text widget, integrating an enhanced Aho-Corasick algorithm for automatic linkification of tiddler titles within text (freelinks). The new implementation prioritizes performance for large wikis and correct support for non-Latin scripts such as Chinese. |
| 39 | + |
| 40 | +Highlights: |
| 41 | +- Full switch from regex-based matching to a custom, robust Aho-Corasick engine dedicated to rapid, multi-pattern title detection—drastically decreasing linkification time (tested: 1–5s reduced to 100–500ms on ~12,000 tiddlers). |
| 42 | +- Handles extremely large title sets gracefully, including a chunked insertion process and use of a persistent cache (`$:/config/Freelinks/PersistAhoCorasickCache`) to further accelerate subsequent linking operations in large/active wikis. |
| 43 | +- Improvements for CJK and non-Latin text: supports linking using long or full-width symbol titles such as '：' (U+FF1A) with no split or mismatch. |
| 44 | +- Smart prioritization: longer titles are automatically matched before shorter, more ambiguous ones, preventing partial/incorrect linking. |
| 45 | +- Actively skips self-linking in the current tiddler and prevents overlapping matches for clean, deterministic linkification. |
| 46 | +- End users with large or multilingual wikis see massive performance boost and 100% accurate linking for complex, full-width, or multi-language titles. |
| 47 | +- New options for persistent match cache and word boundary checking (`$:/config/Freelinks/WordBoundary`), both can be tuned based on wiki size and content language needs. |
| 48 | +- Safe for gradual rollout: legacy behavior is preserved if the new freelinks override is not enabled. |
0 commit comments