⚡ Bolt: Memoize clinical context lexicon regex compilation

.jules/bolt.md

@@ -0,0 +1,3 @@
+## 2024-05-14 - Clinical Regex Compilation Overhead
+**Learning:** `_compiled_context_lexicon` in `openmed.clinical.context` builds an array of complex regular expressions across large clinical cue lexicons (historical, negation, uncertainty, etc). Calling it repeatedly inside functions that evaluate every single span (like `scan_context_cues`, `_scope_bounds`) caused severe performance bottlenecks because it recompiled the regexes on every call.
+**Action:** Always memoize deterministic, heavily-invoked regex compilations or lexicon generations using `@functools.lru_cache` in text processing pipelines (especially in `openmed.clinical`) to avoid repeated, expensive string evaluations.

openmed/openmed/clinical/context.py

@@ -37,6 +37,7 @@
 
 import re
 from collections.abc import Iterable, Iterator, Mapping, Sequence
+from functools import lru_cache
 from dataclasses import dataclass, replace
 from datetime import date
 from typing import Any, Literal
@@ -154,6 +155,10 @@ class _CompiledContextLexicon:
     backward_context_cues: frozenset[str]
 
 
+
+# Cache regex compilations for clinical cues to prevent severe performance
+# bottlenecks during repeated span evaluations.
+@lru_cache(maxsize=32)
 def _compiled_context_lexicon(language: str | None = None) -> _CompiledContextLexicon:
     lexicon = get_clinical_cue_lexicon(language)
     token_boundaries = lexicon.token_boundaries