/**
* 解决两个数相加精度丢失问题
* @param a
* @param b
* @returns {Number}
*/
function floatAdd(a, b) {
var c, d, e;
if(undefined==a||null==a||""==a||isNaN(a)){a=0;}
if(undefined==b||null==b||""==b||isNaN(b)){b=0;}
try {
c = a.toString().split(".")[1].length;
} catch (f) {
c = 0;
}
try {
d = b.toString().split(".")[1].length;
} catch (f) {
d = 0;
}
e = Math.pow(10, Math.max(c, d));
return (floatMul(a, e) + floatMul(b, e)) / e;
}
/**
* 解决两个数相减精度丢失问题
* @param a
* @param b
* @returns {Number}
*/
function floatSub(a, b) {
var c, d, e;
if(undefined==a||null==a||""==a||isNaN(a)){a=0;}
if(undefined==b||null==b||""==b||isNaN(b)){b=0;}
try {
c = a.toString().split(".")[1].length;
} catch (f) {
c = 0;
}
try {
d = b.toString().split(".")[1].length;
} catch (f) {
d = 0;
}
e = Math.pow(10, Math.max(c, d));
return (floatMul(a, e) - floatMul(b, e)) / e;
}
/**
* 解决两个数相乘精度丢失问题
* @param a
* @param b
* @returns {Number}
*/
function floatMul(a, b) {
var c = 0,
d = a.toString(),
e = b.toString();
try {
c += d.split(".")[1].length;
} catch (f) {}
try {
c += e.split(".")[1].length;
} catch (f) {}
return Number(d.replace(".", "")) * Number(e.replace(".", "")) / Math.pow(10, c);
}
/**
* 解决两个数相除精度丢失问题
* @param a
* @param b
* @returns
*/
function floatDiv(a, b) {
var c, d, e = 0,
f = 0;
try {
e = a.toString().split(".")[1].length;
} catch (g) {}
try {
f = b.toString().split(".")[1].length;
} catch (g) {}
return c = Number(a.toString().replace(".", "")), d = Number(b.toString().replace(".", "")), floatMul(c / d, Math.pow(10, f - e));
}
export {
floatAdd,
floatSub,
floatDiv,
floatMul
}