* array.c (rb_ary_sum): Array#sum is implemented.

  Kahan's compensated summation algorithm for precise sum of float
  numbers is moved from ary_inject_op in enum.c.

* enum.c (ary_inject_op): Don't specialize for float numbers.

  [ruby-core:74569] [Feature#12217] proposed by mrkn.



git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@54565 b2dd03c8-39d4-4d8f-98ff-823fe69b080e

5 files changed, 139 insertions, 47 deletions

diff --git a/ChangeLog b/ChangeLog
index 98622ee09f..8c42fa59a4 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,13 @@
+Wed Apr 13 22:51:38 2016  Tanaka Akira  <[email protected]>
+
+	* array.c (rb_ary_sum): Array#sum is implemented.
+	  Kahan's compensated summation algorithm for precise sum of float
+	  numbers is moved from ary_inject_op in enum.c.
+
+	* enum.c (ary_inject_op): Don't specialize for float numbers.
+
+	  [ruby-core:74569] [Feature#12217] proposed by mrkn.
+
 Wed Apr 13 15:56:35 2016  Nobuyoshi Nakada  <[email protected]>
 
 	* numeric.c (flo_ceil): add an optional parameter, digits, as
diff --git a/array.c b/array.c
index 7a01dbc984..85e3177b21 100644
--- a/array.c
+++ b/array.c
@@ -5651,6 +5651,92 @@ rb_ary_dig(int argc, VALUE *argv, VALUE self)
 }
 
 /*
+ * call-seq:
+ *   ary.sum                 -> number
+ *
+ * Returns the sum of elements.
+ * For example, [e1, e2, e3].sum returns 0 + e1 + e2 + e3.
+ *
+ * If <i>ary</i> is empty, it returns 0.
+ *
+ * [].sum                       #=> 0
+ * [1, 2, 3].sum                #=> 6
+ * [3, 5.5].sum                 #=> 8.5
+ *
+ * This method may not respect method redefinition of "+" methods
+ * such as Fixnum#+.
+ *
+ */
+
+VALUE
+rb_ary_sum(VALUE ary)
+{
+    VALUE v, e;
+    long i, n;
+
+    if (RARRAY_LEN(ary) == 0)
+        return LONG2FIX(0);
+
+    v = LONG2FIX(0);
+
+    n = 0;
+    for (i = 0; i < RARRAY_LEN(ary); i++) {
+        e = RARRAY_AREF(ary, i);
+        if (FIXNUM_P(e)) {
+            n += FIX2LONG(e); /* should not overflow long type */
+            if (!FIXABLE(n)) {
+                v = rb_big_plus(LONG2NUM(n), v);
+                n = 0;
+            }
+        }
+        else if (RB_TYPE_P(e, T_BIGNUM))
+            v = rb_big_plus(e, v);
+        else
+            goto not_integer;
+    }
+    if (n != 0)
+        v = rb_fix_plus(LONG2FIX(n), v);
+    return v;
+
+  not_integer:
+    if (n != 0)
+        v = rb_fix_plus(LONG2FIX(n), v);
+
+    if (RB_FLOAT_TYPE_P(e)) {
+        /* Kahan's compensated summation algorithm */
+        double f, c;
+        f = NUM2DBL(v);
+        c = 0.0;
+        for (; i < RARRAY_LEN(ary); i++) {
+            double x, y, t;
+            e = RARRAY_AREF(ary, i);
+            if (RB_FLOAT_TYPE_P(e))
+                x = RFLOAT_VALUE(e);
+            else if (FIXNUM_P(e))
+                x = FIX2LONG(e);
+            else if (RB_TYPE_P(e, T_BIGNUM))
+                x = rb_big2dbl(e);
+            else
+                goto not_float;
+
+            y = x - c;
+            t = f + y;
+            c = (t - f) - y;
+            f = t;
+        }
+        return DBL2NUM(f);
+
+      not_float:
+        v = DBL2NUM(f);
+    }
+
+    for (; i < RARRAY_LEN(ary); i++) {
+        v = rb_funcall(v, idPLUS, 1, RARRAY_AREF(ary, i));
+    }
+    return v;
+}
+
+/*
  *  Arrays are ordered, integer-indexed collections of any object.
  *
  *  Array indexing starts at 0, as in C or Java.  A negative index is assumed
@@ -6005,6 +6091,7 @@ Init_Array(void)
     rb_define_method(rb_cArray, "bsearch_index", rb_ary_bsearch_index, 0);
     rb_define_method(rb_cArray, "any?", rb_ary_any_p, 0);
     rb_define_method(rb_cArray, "dig", rb_ary_dig, -1);
+    rb_define_method(rb_cArray, "sum", rb_ary_sum, 0);
 
     id_cmp = rb_intern("<=>");
     id_random = rb_intern("random");
diff --git a/enum.c b/enum.c
index 23e4f5a5e4..9387ec3b6a 100644
--- a/enum.c
+++ b/enum.c
@@ -634,7 +634,6 @@ ary_inject_op(VALUE ary, VALUE init, VALUE op)
     ID id;
     VALUE v, e;
     long i, n;
-    double f, c;
 
     if (RARRAY_LEN(ary) == 0)
         return init == Qundef ? Qnil : init;
@@ -656,7 +655,7 @@ ary_inject_op(VALUE ary, VALUE init, VALUE op)
              rb_method_basic_definition_p(rb_cFixnum, idPLUS) &&
              rb_method_basic_definition_p(rb_cBignum, idPLUS)) {
             n = 0;
-            while (1) {
+            for (; i < RARRAY_LEN(ary); i++) {
                 e = RARRAY_AREF(ary, i);
                 if (FIXNUM_P(e)) {
                     n += FIX2LONG(e); /* should not overflow long type */
@@ -668,49 +667,18 @@ ary_inject_op(VALUE ary, VALUE init, VALUE op)
                 else if (RB_TYPE_P(e, T_BIGNUM))
                     v = rb_big_plus(e, v);
                 else
-                    break;
-                i++;
-                if (RARRAY_LEN(ary) <= i)
-                    return n == 0 ? v : rb_fix_plus(LONG2FIX(n), v);
+                    goto not_integer;
             }
-            if (n != 0) {
+            if (n != 0)
                 v = rb_fix_plus(LONG2FIX(n), v);
-            }
-            if (RB_FLOAT_TYPE_P(e) &&
-                rb_method_basic_definition_p(rb_cFloat, idPLUS)) {
-                f = NUM2DBL(v);
-                goto sum_float;
-            }
-        }
-        else if (RB_FLOAT_TYPE_P(v) &&
-                 rb_method_basic_definition_p(rb_cFloat, idPLUS)) {
-            f = RFLOAT_VALUE(v);
-          sum_float:
-            c = 0.0;
-            while (1) {
-                double x, y, t;
-                e = RARRAY_AREF(ary, i);
-                if (RB_FLOAT_TYPE_P(e))
-                    x = RFLOAT_VALUE(e);
-                else if (FIXNUM_P(e))
-                    x = FIX2LONG(e);
-                else if (RB_TYPE_P(e, T_BIGNUM))
-                    x = rb_big2dbl(e);
-                else
-                    break;
-
-                y = x - c;
-                t = f + y;
-                c = (t - f) - y;
-                f = t;
+            return v;
 
-                i++;
-                if (RARRAY_LEN(ary) <= i)
-                    return DBL2NUM(f);
-            }
+          not_integer:
+            if (n != 0)
+                v = rb_fix_plus(LONG2FIX(n), v);
         }
     }
-    for (; i<RARRAY_LEN(ary); i++) {
+    for (; i < RARRAY_LEN(ary); i++) {
         v = rb_funcall(v, id, 1, RARRAY_AREF(ary, i));
     }
     return v;
diff --git a/test/ruby/test_array.rb b/test/ruby/test_array.rb
index fa158185f8..1881b86694 100644
--- a/test/ruby/test_array.rb
+++ b/test/ruby/test_array.rb
@@ -1,6 +1,7 @@
 # coding: US-ASCII
 # frozen_string_literal: false
 require 'test/unit'
+require "rbconfig/sizeof"
 
 class TestArray < Test::Unit::TestCase
   def setup
@@ -2710,6 +2711,39 @@ class TestArray < Test::Unit::TestCase
     assert_raise(TypeError) {h.dig(1, 0)}
   end
 
+  FIXNUM_MIN = -(1 << (8 * RbConfig::SIZEOF['long'] - 2))
+  FIXNUM_MAX = (1 << (8 * RbConfig::SIZEOF['long'] - 2)) - 1
+
+  def assert_float_equal(e, v, msg=nil)
+    assert_equal(Float, v.class, msg)
+    assert_equal(e, v, msg)
+  end
+
+  def test_sum
+    assert_equal(0, [].sum)
+    assert_equal(3, [3].sum)
+    assert_equal(8, [3, 5].sum)
+    assert_equal(15, [3, 5, 7].sum)
+    assert_float_equal(15.0, [3, 5, 7.0].sum)
+    assert_equal(2*FIXNUM_MAX, Array.new(2, FIXNUM_MAX).sum)
+    assert_equal(2*(FIXNUM_MAX+1), Array.new(2, FIXNUM_MAX+1).sum)
+    assert_equal(10*FIXNUM_MAX, Array.new(10, FIXNUM_MAX).sum)
+    assert_equal(0, ([FIXNUM_MAX, 1, -FIXNUM_MAX, -1]*10).sum)
+    assert_equal(FIXNUM_MAX*10, ([FIXNUM_MAX+1, -1]*10).sum)
+    assert_equal(2*FIXNUM_MIN, Array.new(2, FIXNUM_MIN).sum)
+    assert_equal((FIXNUM_MAX+1).to_f, [FIXNUM_MAX, 1, 0.0].sum)
+    assert_float_equal(8.0, [3.0, 5].sum)
+    assert_float_equal((FIXNUM_MAX+1).to_f, [0.0, FIXNUM_MAX+1].sum)
+    assert_equal(2.0+3.0i, [2.0, 3.0i].sum)
+
+    large_number = 100000000
+    small_number = 1e-9
+    until (large_number + small_number) == large_number
+      small_number /= 10
+    end
+    assert_equal(large_number+(small_number*10), [large_number, *[small_number]*10].sum)
+  end
+
   private
   def need_continuation
     unless respond_to?(:callcc, true)
diff --git a/test/ruby/test_enum.rb b/test/ruby/test_enum.rb
index ba973e2d48..97730f919f 100644
--- a/test/ruby/test_enum.rb
+++ b/test/ruby/test_enum.rb
@@ -217,13 +217,6 @@ class TestEnumerable < Test::Unit::TestCase
     assert_float_equal(10.0, [3.0, 5].inject(2.0, :+))
     assert_float_equal((FIXNUM_MAX+1).to_f, [0.0, FIXNUM_MAX+1].inject(:+))
     assert_equal(2.0+3.0i, [2.0, 3.0i].inject(:+))
-
-    large_number = 100000000
-    small_number = 1e-9
-    until (large_number + small_number) == large_number
-      small_number /= 10
-    end
-    assert_equal(large_number+(small_number*10), [large_number, *[small_number]*10].inject(:+))
   end
 
   def test_inject_array_plus_redefined