Merged the int/long unification branch, by very crude means (sorry Thomas!).
I banged on the code (beyond what's in that branch) to make fewer tests fail;
the only tests that fail now are:
test_descr -- can't pickle ints?!
test_pickletools -- ???
test_socket -- See python.org/sf/1619659
test_sqlite -- ???
I'll deal with those later.
diff --git a/INTBENCH b/INTBENCH
new file mode 100644
index 0000000..495a824
--- /dev/null
+++ b/INTBENCH
@@ -0,0 +1,50 @@
+Measurements of _testcapi.profile_int, on an 800MHz G3, OSX 10.4.7
+best of three runs
+
+r51476 (original p3yk)
+Test 1: 1.601978s
+Test 2: 1.696623s
+Test 3: 1.900683s
+Test 4: 5.307155s
+Test 5: 2.546707s
+Test 6: 1.670252s
+Test 7: 1.910734s
+
+r51506 (int type dropped)
+Test 1: 4.134757s
+Test 2: 4.398235s
+Test 3: 4.611636s
+Test 4: 10.665429s
+
+r51509 (small int cache)
+Test 1: 3.457184s
+Test 2: 4.514800s
+Test 3: 4.999010s
+Test 4: 10.818277s
+
+r51452 (special-casing medium int allocation)
+Test 1: 3.258219s
+Test 2: 4.255007s
+Test 3: 4.547923s
+Test 4: 10.615123s
+Test 5: 5.255545s
+Test 6: 3.775941s
+Test 7: 4.001805s
+
+r51562 (special-case one-digit operations)
+Test 1: 3.527860s
+Test 2: 3.975953s
+Test 3: 4.226751s
+Test 4: 10.605721s
+Test 5: 5.233576s
+Test 6: 2.161525s
+Test 7: 3.421624s
+
+r51573 (speed up PyLong_FromLong)
+Test 1: 3.149116s
+Test 2: 3.948204s
+Test 3: 4.012080s
+Test 4: 8.589921s
+Test 5: 4.481723s
+Test 6: 2.259849s
+Test 7: 3.453212s