⚡ Bolt: Cache lexicon regex patterns to reduce compilation overhead

Implemented an explicit dictionary cache for compiled regex patterns in `KokoroEngine.apply_lexicon`. This avoids recompiling regex patterns on every call for every text segment, which significantly improves performance and prevents evicting Python's internal regex cache when processing large documents with extensive lexicons.

Author: CoffeeMethod

.jules/bolt.md

@@ -0,0 +1,3 @@
+## 2025-02-12 - Explicit Regex Caching in Tight Loops
+**Learning:** Python has an internal cache for compiled regex patterns (usually up to 512 entries). However, dynamically compiling a large number of varying regex patterns in tight loops can quickly evict entries, leading to silent performance degradation as patterns are constantly recompiled. In `kokoro_engine.py`, `apply_lexicon` was compiling regex patterns on every single lexicon entry for every text segment processed, introducing measurable overhead during large document generation.
+**Action:** When performing regex-based text replacements with dynamic or user-provided mappings (like lexicons), explicitly cache the `re.compile()` objects in a class attribute or dictionary. This avoids recompilation overhead when processing large or split text blocks where the same lexicon mappings are used repeatedly.

kokoro_engine.py

@@ -77,6 +77,8 @@ def __init__(self):
         if not os.path.exists(CACHE_DIR):
             os.makedirs(CACHE_DIR)
 
+        self._lexicon_cache = {}  # Cache for compiled regex patterns
+
         # Callbacks
         self.on_progress = None # func(percentage, time_elapsed, eta, detail_text)
         self.on_status = None   # func(msg, is_error)
@@ -93,8 +95,10 @@ def apply_lexicon(self, text, lexicon):
         for src, dest in lexicon.items():
             if not src: continue
             try:
-                # Escape the search term to treat it as literal text
-                pattern = re.compile(re.escape(src), re.IGNORECASE)
+                if src not in self._lexicon_cache:
+                    # Escape the search term to treat it as literal text
+                    self._lexicon_cache[src] = re.compile(re.escape(src), re.IGNORECASE)
+                pattern = self._lexicon_cache[src]
                 text = pattern.sub(dest, text)
             except Exception as e:
                 print(f"Lexicon error for '{src}': {e}")