[Bug #22092] Improve Array#sum when the initial value is a Float

Author: nobu

array.c

@@ -8247,7 +8247,11 @@ rb_ary_sum(int argc, VALUE *argv, VALUE ary)
     n = 0;
     r = Qundef;
 
-    if (!FIXNUM_P(v) && !RB_BIGNUM_TYPE_P(v) && !RB_TYPE_P(v, T_RATIONAL)) {
+    bool init_is_float = RB_FLOAT_TYPE_P(v);
+    if (init_is_float) {
+        v = LONG2FIX(0);
+    }
+    else if (!RB_INTEGER_TYPE_P(v) && !RB_TYPE_P(v, T_RATIONAL)) {
         i = 0;
         goto init_is_a_value;
     }
@@ -8275,12 +8279,13 @@ rb_ary_sum(int argc, VALUE *argv, VALUE ary)
             goto not_exact;
     }
     v = finish_exact_sum(n, r, v, argc!=0);
+    if (init_is_float) v = rb_float_plus(argv[0], v);
     return v;
 
   not_exact:
     v = finish_exact_sum(n, r, v, i!=0);
 
-    if (RB_FLOAT_TYPE_P(e)) {
+    if (init_is_float ? (--i, e = argv[0], true) : RB_FLOAT_TYPE_P(e)) {
         /*
          * Kahan-Babuska balancing compensated summation algorithm
          * See https://link.springer.com/article/10.1007/s00607-005-0139-x

test/ruby/test_array.rb

@@ -3524,6 +3524,7 @@ def test_sum
     assert_float_equal(3.5, [3].sum(0.5))
     assert_float_equal(8.5, [3.5, 5].sum)
     assert_float_equal(10.5, [2, 8.5].sum)
+    assert_float_equal(1_000 * 0.1, Array.new(1_000, 0.1).sum(0.0))
     assert_float_equal((FIXNUM_MAX+1).to_f, [FIXNUM_MAX, 1, 0.0].sum)
     assert_float_equal((FIXNUM_MAX+1).to_f, [0.0, FIXNUM_MAX+1].sum)