QHash: remove optimization for QHash<int, T>

QHash employs an optimization for int/uints, squashing the hash
value and the key value inside an union. This obviously works
iff qHash(int k) = k. If the hash value gets salted, the hash
table is corrupted.

This patch removes that optimization by means of a #if 0,
so if further research finds out that we want those 4 bytes back
it's pretty simple to revert.

Change-Id: If273f0bf2ff007f4f2f7c46d2aab364a3b455cf1
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>

diff --git a/src/corelib/tools/qhash.h b/src/corelib/tools/qhash.h
index 1aa4fa7..e318872 100644 (file)
--- a/src/corelib/tools/qhash.h
+++ b/src/corelib/tools/qhash.h
@@ -218,7 +218,19 @@ struct QHashNode
     inline bool same_key(uint h0, const Key &key0) { return h0 == h && key0 == key; }
 };
 
-
+#if 0
+// ###
+// The introduction of the QHash random seed breaks this optimization, as it
+// relies on qHash(int i) = i. If the hash value is salted, then the hash
+// table becomes corrupted.
+//
+// A bit more research about whether it makes sense or not to salt integer
+// keys (and in general keys whose hash value is easy to invert)
+// is needed, or about how keep this optimization and the seed (f.i. by
+// specializing QHash for integer values, and re-apply the seed during lookup).
+//
+// Be aware that such changes can easily be binary incompatible, and therefore
+// cannot be made during the Qt 5 lifetime.
 #define Q_HASH_DECLARE_INT_NODES(key_type) \
     template <class T> \
     struct QHashDummyNode<key_type, T> { \
@@ -246,6 +258,7 @@ Q_HASH_DECLARE_INT_NODES(ushort);
 Q_HASH_DECLARE_INT_NODES(int);
 Q_HASH_DECLARE_INT_NODES(uint);
 #undef Q_HASH_DECLARE_INT_NODES
+#endif // #if 0
 
 template <class Key, class T>
 class QHash