⚡ Bolt: Cache compiled context lexicon regexes

.jules/bolt.md

@@ -0,0 +1,3 @@
+## 2024-05-19 - Regex Compilation Caching
+**Learning:** Repetitive regex compilation, particularly when driven by a language parameter across multi-layer logic like `_compiled_context_lexicon`, forms a severe performance bottleneck if not cached.
+**Action:** Always memoize functions that return deterministic regular expressions derived from lexicons using `@functools.lru_cache`, provided their arguments are hashable and the returned object is immutable.

openmed/openmed/clinical/context.py

@@ -35,6 +35,7 @@
 
 from __future__ import annotations
 
+import functools
 import re
 from collections.abc import Iterable, Iterator, Mapping, Sequence
 from dataclasses import dataclass, replace
@@ -154,6 +155,9 @@ class _CompiledContextLexicon:
     backward_context_cues: frozenset[str]
 
 
+# ⚡ Bolt: Cache compiled context lexicon regexes to avoid expensive
+# compilation during high-throughput execution across span assertions.
+@functools.lru_cache(maxsize=16)
 def _compiled_context_lexicon(language: str | None = None) -> _CompiledContextLexicon:
     lexicon = get_clinical_cue_lexicon(language)
     token_boundaries = lexicon.token_boundaries