IndexedDB: Replace O(n^2) algorithm computing multientry index keys

third_party/WebKit/Source/modules/indexeddb/IDBObjectStore.cpp

Index: third_party/WebKit/Source/modules/indexeddb/IDBObjectStore.cpp
diff --git a/third_party/WebKit/Source/modules/indexeddb/IDBObjectStore.cpp b/third_party/WebKit/Source/modules/indexeddb/IDBObjectStore.cpp
index 8ca3177513f2d504be24887d9d42b37c5816514c..d84113941ad493cb317b8a7e137df954ef7dda93 100644
--- a/third_party/WebKit/Source/modules/indexeddb/IDBObjectStore.cpp
+++ b/third_party/WebKit/Source/modules/indexeddb/IDBObjectStore.cpp
@@ -330,9 +330,7 @@ static void generateIndexKeysForValue(v8::Isolate* isolate,
     DCHECK(indexMetadata.multiEntry);
     DCHECK_EQ(indexKey->getType(), IDBKey::ArrayType);
     indexKey = IDBKey::createMultiEntryArray(indexKey->array());
-
-    for (size_t i = 0; i < indexKey->array().size(); ++i)
-      indexKeys->push_back(indexKey->array()[i]);
+    indexKeys->appendVector(indexKey->array());

[pwnall, 2017/01/18 02:35:19]

Would it make sense to have IDBKey::createMultiEntryArray return a KeyArray, and
pass the return type directly to appendVector()?

codesearch says that IDBKey::createMultiEntryArray isn't called elsewhere, and
removing the packing/unpacking could let you get rid of some code.
Performance-wise, I'd expect/hope that RVO would make it free to return the
vector directly (no unnecessary copies).

[jsbell, 2017/01/18 18:18:56]

Done; also made it non-static to simplify further.

[pwnall, 2017/01/18 19:04:15]

Cool, nice find!

   }
 }