⚡ Bolt: 핫 패스에서 dict.fromkeys() 제너레이터 오버헤드 최적화

.jules/bolt.md

@@ -77,3 +77,7 @@
 ## 2024-11-20 - Optimize multiple tuple generation from a single collection
 **Learning:** `build_rule_metadata` derives exactly two collections, `owasp` and `cwe`, from the same references. Replacing its two generator traversals with one explicit loop reduces element visits from about 2N to N. Both versions remain O(N), so this is a constant-factor optimization rather than an asymptotic complexity improvement.
 **Action:** Combine repeated traversal when fixed derived collections share one source, while preserving ordering and classification semantics. Benchmark the production hot path before claiming a material wall-clock improvement.
+
+## 2024-07-21 - Optimize dict.fromkeys() overhead in hot paths
+**Learning:** Using `dict.fromkeys(generator_expression)` in hot paths creates unnecessary overhead. The generator comprehension requires frame allocation and iteration overhead, which can be significantly slower than a direct loop.
+**Action:** In Python hot paths, prefer an explicit `for` loop that updates a local dictionary (e.g., `seen = {}`; `seen[item] = None`) over `dict.fromkeys(generator)`. This avoids generator object instantiation overhead and frame allocation, yielding measurable performance improvements.

appguardrail_core/rules.py

@@ -106,12 +106,10 @@ def as_dict(self) -> dict[str, Any]:
 
 def extract_public_references(message: str) -> tuple[str, ...]:
     """Extract OWASP, CWE, and CVE references already embedded in rule copy."""
-    return tuple(
-        dict.fromkeys(
-            " ".join(match.group(1).split())
-            for match in REFERENCE_RE.finditer(message or "")
-        )
-    )
+    seen = {}
+    for match in REFERENCE_RE.finditer(message or ""):
+        seen[" ".join(match.group(1).split())] = None
+    return tuple(seen)
 
 
 def _category_for_references(references: tuple[str, ...], fallback: str) -> str:
@@ -174,8 +172,9 @@ def validate_rule_metadata(metadata: RuleMetadata | dict[str, Any]) -> list[str]
 
 
 def _merge_references(*groups: tuple[str, ...]) -> tuple[str, ...]:
-    return tuple(
-        dict.fromkeys(
-            reference for group in groups for reference in group if reference
-        )
-    )
+    seen = {}
+    for group in groups:
+        for reference in group:
+            if reference:
+                seen[reference] = None
+    return tuple(seen)