From 5105c3651a8f1b153a9ce7c1ac327f6933283be3 Mon Sep 17 00:00:00 2001
From: Teknium <127238744+teknium1@users.noreply.github.com>
Date: Sun, 14 Jun 2026 03:25:49 -0700
Subject: [PATCH] perf(api-server): normalize chat content linearly (#46079)

---
 gateway/platforms/api_server.py            | 12 +++++++++---
 tests/gateway/test_api_server_normalize.py | 17 +++++++++++++++++
 2 files changed, 26 insertions(+), 3 deletions(-)

diff --git a/gateway/platforms/api_server.py b/gateway/platforms/api_server.py
index 1599eda9e6..da86952a09 100644
--- a/gateway/platforms/api_server.py
+++ b/gateway/platforms/api_server.py
@@ -156,18 +156,23 @@ def _normalize_chat_content(
 
     if isinstance(content, list):
         parts: List[str] = []
+        total_len = 0
         items = content[:MAX_CONTENT_LIST_SIZE] if len(content) > MAX_CONTENT_LIST_SIZE else content
         for item in items:
             if isinstance(item, str):
                 if item:
-                    parts.append(item[:MAX_NORMALIZED_TEXT_LENGTH])
+                    part = item[:MAX_NORMALIZED_TEXT_LENGTH]
+                    parts.append(part)
+                    total_len += len(part)
             elif isinstance(item, dict):
                 item_type = str(item.get("type") or "").strip().lower()
                 if item_type in {"text", "input_text", "output_text"}:
                     text = item.get("text", "")
                     if text:
                         try:
-                            parts.append(str(text)[:MAX_NORMALIZED_TEXT_LENGTH])
+                            part = str(text)[:MAX_NORMALIZED_TEXT_LENGTH]
+                            parts.append(part)
+                            total_len += len(part)
                         except Exception:
                             pass
                 # Silently skip image_url / other non-text parts
@@ -175,8 +180,9 @@ def _normalize_chat_content(
                 nested = _normalize_chat_content(item, _max_depth=_max_depth, _depth=_depth + 1)
                 if nested:
                     parts.append(nested)
+                    total_len += len(nested)
             # Check accumulated size
-            if sum(len(p) for p in parts) >= MAX_NORMALIZED_TEXT_LENGTH:
+            if total_len >= MAX_NORMALIZED_TEXT_LENGTH:
                 break
         result = "\n".join(parts)
         return result[:MAX_NORMALIZED_TEXT_LENGTH] if len(result) > MAX_NORMALIZED_TEXT_LENGTH else result
diff --git a/tests/gateway/test_api_server_normalize.py b/tests/gateway/test_api_server_normalize.py
index 2dd2c70f72..1f943ced01 100644
--- a/tests/gateway/test_api_server_normalize.py
+++ b/tests/gateway/test_api_server_normalize.py
@@ -1,5 +1,6 @@
 """Tests for _normalize_chat_content in the API server adapter."""
 
+from gateway.platforms import api_server
 from gateway.platforms.api_server import _normalize_chat_content
 
 
@@ -85,3 +86,19 @@ class TestNormalizeChatContent:
 
     def test_empty_list_returns_empty(self):
         assert _normalize_chat_content([]) == ""
+
+    def test_many_small_parts_normalize_without_quadratic_rescan(self, monkeypatch):
+        """Large content arrays should normalize in linear time."""
+        content = [{"type": "text", "text": "x"} for _ in range(1000)]
+        sum_calls = 0
+
+        def counting_sum(values):
+            nonlocal sum_calls
+            sum_calls += 1
+            return sum(values)
+
+        monkeypatch.setattr(api_server, "sum", counting_sum, raising=False)
+        result = _normalize_chat_content(content)
+
+        assert result.count("x") == 1000
+        assert sum_calls == 0