#128 Add tinyPow to workaround V8 Math.pow change

decimal.js

@@ -317,7 +317,7 @@
     external = false;
 
     // Initial estimate.
-    s = x.s * Math.pow(x.s * x, 1 / 3);
+    s = x.s * mathpow(x.s * x, 1 / 3);
 
      // Math.cbrt underflow/overflow?
      // Pass x to Math.pow as integer, then adjust the exponent of the result.
@@ -327,7 +327,7 @@
 
       // Adjust n exponent so it is a multiple of 3 away from x exponent.
       if (s = (e - n.length + 1) % 3) n += (s == 1 || s == -2 ? '0' : '00');
-      s = Math.pow(n, 1 / 3);
+      s = mathpow(n, 1 / 3);
 
       // Rarely, e may be one less than the result exponent value.
       e = mathfloor((e + 1) / 3) - (e % 3 == (e < 0 ? -1 : 2));
@@ -546,7 +546,7 @@
     // TODO? Estimation reused from cosine() and may not be optimal here.
     if (len < 32) {
       k = Math.ceil(len / 3);
-      n = Math.pow(4, -k).toString();
+      n = (1 / tinyPow(4, k)).toString();
     } else {
       k = 16;
       n = '2.3283064365386962890625e-10';
@@ -626,8 +626,7 @@
       k = 1.4 * Math.sqrt(len);
       k = k > 16 ? 16 : k | 0;
 
-      x = x.times(Math.pow(5, -k));
-
+      x = x.times(1 / tinyPow(5, k));
       x = taylorSeries(Ctor, 2, x, x, true);
 
       // Reverse argument reduction
@@ -2638,7 +2637,7 @@
     // Estimate the optimum number of times to use the argument reduction.
     if (len < 32) {
       k = Math.ceil(len / 3);
-      y = Math.pow(4, -k).toString();
+      y = (1 / tinyPow(4, k)).toString();
     } else {
       k = 16;
       y = '2.3283064365386962890625e-10';
@@ -3649,7 +3648,7 @@
     if (isFloat) x = divide(x, divisor, len * 4);
 
     // Multiply by the binary exponent part if present.
-    if (p) x = x.times(Math.abs(p) < 54 ? Math.pow(2, p) : Decimal.pow(2, p));
+    if (p) x = x.times(Math.abs(p) < 54 ? mathpow(2, p) : Decimal.pow(2, p));
     external = true;
 
     return x;
@@ -3675,8 +3674,7 @@
     k = 1.4 * Math.sqrt(len);
     k = k > 16 ? 16 : k | 0;
 
-    // Max k before Math.pow precision loss is 22
-    x = x.times(Math.pow(5, -k));
+    x = x.times(1 / tinyPow(5, k));
     x = taylorSeries(Ctor, 2, x, x);
 
     // Reverse argument reduction
@@ -3729,6 +3727,14 @@
   }
 
 
+  // Exponent e must be positive and non-zero.
+  function tinyPow(b, e) {
+    var n = b;
+    while (--e) n *= b;
+    return n;
+  }
+
+
   // Return the absolute value of `x` reduced to less than or equal to half pi.
   function toLessThanHalfPi(Ctor, x) {
     var t,

decimal.mjs

@@ -313,7 +313,7 @@ P.cubeRoot = P.cbrt = function () {
   external = false;
 
   // Initial estimate.
-  s = x.s * Math.pow(x.s * x, 1 / 3);
+  s = x.s * mathpow(x.s * x, 1 / 3);
 
    // Math.cbrt underflow/overflow?
    // Pass x to Math.pow as integer, then adjust the exponent of the result.
@@ -323,7 +323,7 @@ P.cubeRoot = P.cbrt = function () {
 
     // Adjust n exponent so it is a multiple of 3 away from x exponent.
     if (s = (e - n.length + 1) % 3) n += (s == 1 || s == -2 ? '0' : '00');
-    s = Math.pow(n, 1 / 3);
+    s = mathpow(n, 1 / 3);
 
     // Rarely, e may be one less than the result exponent value.
     e = mathfloor((e + 1) / 3) - (e % 3 == (e < 0 ? -1 : 2));
@@ -542,7 +542,7 @@ P.hyperbolicCosine = P.cosh = function () {
   // TODO? Estimation reused from cosine() and may not be optimal here.
   if (len < 32) {
     k = Math.ceil(len / 3);
-    n = Math.pow(4, -k).toString();
+    n = (1 / tinyPow(4, k)).toString();
   } else {
     k = 16;
     n = '2.3283064365386962890625e-10';
@@ -622,8 +622,7 @@ P.hyperbolicSine = P.sinh = function () {
     k = 1.4 * Math.sqrt(len);
     k = k > 16 ? 16 : k | 0;
 
-    x = x.times(Math.pow(5, -k));
-
+    x = x.times(1 / tinyPow(5, k));
     x = taylorSeries(Ctor, 2, x, x, true);
 
     // Reverse argument reduction
@@ -2634,7 +2633,7 @@ function cosine(Ctor, x) {
   // Estimate the optimum number of times to use the argument reduction.
   if (len < 32) {
     k = Math.ceil(len / 3);
-    y = Math.pow(4, -k).toString();
+    y = (1 / tinyPow(4, k)).toString();
   } else {
     k = 16;
     y = '2.3283064365386962890625e-10';
@@ -3645,7 +3644,7 @@ function parseOther(x, str) {
   if (isFloat) x = divide(x, divisor, len * 4);
 
   // Multiply by the binary exponent part if present.
-  if (p) x = x.times(Math.abs(p) < 54 ? Math.pow(2, p) : Decimal.pow(2, p));
+  if (p) x = x.times(Math.abs(p) < 54 ? mathpow(2, p) : Decimal.pow(2, p));
   external = true;
 
   return x;
@@ -3671,8 +3670,7 @@ function sine(Ctor, x) {
   k = 1.4 * Math.sqrt(len);
   k = k > 16 ? 16 : k | 0;
 
-  // Max k before Math.pow precision loss is 22
-  x = x.times(Math.pow(5, -k));
+  x = x.times(1 / tinyPow(5, k));
   x = taylorSeries(Ctor, 2, x, x);
 
   // Reverse argument reduction
@@ -3725,6 +3723,14 @@ function taylorSeries(Ctor, n, x, y, isHyperbolic) {
 }
 
 
+// Exponent e must be positive and non-zero.
+function tinyPow(b, e) {
+  var n = b;
+  while (--e) n *= b;
+  return n;
+}                                             
+
+
 // Return the absolute value of `x` reduced to less than or equal to half pi.
 function toLessThanHalfPi(Ctor, x) {
   var t,