package goja import ( "math" "math/bits" "sync" ) func (r *Runtime) math_abs(call FunctionCall) Value { return floatToValue(math.Abs(call.Argument(0).ToFloat())) } func (r *Runtime) math_acos(call FunctionCall) Value { return floatToValue(math.Acos(call.Argument(0).ToFloat())) } func (r *Runtime) math_acosh(call FunctionCall) Value { return floatToValue(math.Acosh(call.Argument(0).ToFloat())) } func (r *Runtime) math_asin(call FunctionCall) Value { return floatToValue(math.Asin(call.Argument(0).ToFloat())) } func (r *Runtime) math_asinh(call FunctionCall) Value { return floatToValue(math.Asinh(call.Argument(0).ToFloat())) } func (r *Runtime) math_atan(call FunctionCall) Value { return floatToValue(math.Atan(call.Argument(0).ToFloat())) } func (r *Runtime) math_atanh(call FunctionCall) Value { return floatToValue(math.Atanh(call.Argument(0).ToFloat())) } func (r *Runtime) math_atan2(call FunctionCall) Value { y := call.Argument(0).ToFloat() x := call.Argument(1).ToFloat() return floatToValue(math.Atan2(y, x)) } func (r *Runtime) math_cbrt(call FunctionCall) Value { return floatToValue(math.Cbrt(call.Argument(0).ToFloat())) } func (r *Runtime) math_ceil(call FunctionCall) Value { return floatToValue(math.Ceil(call.Argument(0).ToFloat())) } func (r *Runtime) math_clz32(call FunctionCall) Value { return intToValue(int64(bits.LeadingZeros32(toUint32(call.Argument(0))))) } func (r *Runtime) math_cos(call FunctionCall) Value { return floatToValue(math.Cos(call.Argument(0).ToFloat())) } func (r *Runtime) math_cosh(call FunctionCall) Value { return floatToValue(math.Cosh(call.Argument(0).ToFloat())) } func (r *Runtime) math_exp(call FunctionCall) Value { return floatToValue(math.Exp(call.Argument(0).ToFloat())) } func (r *Runtime) math_expm1(call FunctionCall) Value { return floatToValue(math.Expm1(call.Argument(0).ToFloat())) } func (r *Runtime) math_floor(call FunctionCall) Value { return floatToValue(math.Floor(call.Argument(0).ToFloat())) } func (r *Runtime) math_fround(call FunctionCall) Value { return floatToValue(float64(float32(call.Argument(0).ToFloat()))) } func (r *Runtime) math_hypot(call FunctionCall) Value { var max float64 var hasNaN bool absValues := make([]float64, 0, len(call.Arguments)) for _, v := range call.Arguments { arg := nilSafe(v).ToFloat() if math.IsNaN(arg) { hasNaN = true } else { abs := math.Abs(arg) if abs > max { max = abs } absValues = append(absValues, abs) } } if math.IsInf(max, 1) { return _positiveInf } if hasNaN { return _NaN } if max == 0 { return _positiveZero } // Kahan summation to avoid rounding errors. // Normalize the numbers to the largest one to avoid overflow. var sum, compensation float64 for _, n := range absValues { n /= max summand := n*n - compensation preliminary := sum + summand compensation = (preliminary - sum) - summand sum = preliminary } return floatToValue(math.Sqrt(sum) * max) } func (r *Runtime) math_imul(call FunctionCall) Value { x := toUint32(call.Argument(0)) y := toUint32(call.Argument(1)) return intToValue(int64(int32(x * y))) } func (r *Runtime) math_log(call FunctionCall) Value { return floatToValue(math.Log(call.Argument(0).ToFloat())) } func (r *Runtime) math_log1p(call FunctionCall) Value { return floatToValue(math.Log1p(call.Argument(0).ToFloat())) } func (r *Runtime) math_log10(call FunctionCall) Value { return floatToValue(math.Log10(call.Argument(0).ToFloat())) } func (r *Runtime) math_log2(call FunctionCall) Value { return floatToValue(math.Log2(call.Argument(0).ToFloat())) } func (r *Runtime) math_max(call FunctionCall) Value { result := math.Inf(-1) args := call.Arguments for i, arg := range args { n := nilSafe(arg).ToFloat() if math.IsNaN(n) { args = args[i+1:] goto NaNLoop } result = math.Max(result, n) } return floatToValue(result) NaNLoop: // All arguments still need to be coerced to number according to the specs. for _, arg := range args { nilSafe(arg).ToFloat() } return _NaN } func (r *Runtime) math_min(call FunctionCall) Value { result := math.Inf(1) args := call.Arguments for i, arg := range args { n := nilSafe(arg).ToFloat() if math.IsNaN(n) { args = args[i+1:] goto NaNLoop } result = math.Min(result, n) } return floatToValue(result) NaNLoop: // All arguments still need to be coerced to number according to the specs. for _, arg := range args { nilSafe(arg).ToFloat() } return _NaN } func pow(x, y Value) Value { if x, ok := x.(valueInt); ok { if y, ok := y.(valueInt); ok && y >= 0 { if y == 0 { return intToValue(1) } if x == 0 { return intToValue(0) } ip := ipow(int64(x), int64(y)) if ip != 0 { return intToValue(ip) } } } xf := x.ToFloat() yf := y.ToFloat() if math.Abs(xf) == 1 && math.IsInf(yf, 0) { return _NaN } if xf == 1 && math.IsNaN(yf) { return _NaN } return floatToValue(math.Pow(xf, yf)) } func (r *Runtime) math_pow(call FunctionCall) Value { return pow(call.Argument(0), call.Argument(1)) } func (r *Runtime) math_random(call FunctionCall) Value { return floatToValue(r.rand()) } func (r *Runtime) math_round(call FunctionCall) Value { f := call.Argument(0).ToFloat() if math.IsNaN(f) { return _NaN } if f == 0 && math.Signbit(f) { return _negativeZero } t := math.Trunc(f) if f >= 0 { if f-t >= 0.5 { return floatToValue(t + 1) } } else { if t-f > 0.5 { return floatToValue(t - 1) } } return floatToValue(t) } func (r *Runtime) math_sign(call FunctionCall) Value { arg := call.Argument(0) num := arg.ToFloat() if math.IsNaN(num) || num == 0 { // this will match -0 too return arg } if num > 0 { return intToValue(1) } return intToValue(-1) } func (r *Runtime) math_sin(call FunctionCall) Value { return floatToValue(math.Sin(call.Argument(0).ToFloat())) } func (r *Runtime) math_sinh(call FunctionCall) Value { return floatToValue(math.Sinh(call.Argument(0).ToFloat())) } func (r *Runtime) math_sqrt(call FunctionCall) Value { return floatToValue(math.Sqrt(call.Argument(0).ToFloat())) } func (r *Runtime) math_tan(call FunctionCall) Value { return floatToValue(math.Tan(call.Argument(0).ToFloat())) } func (r *Runtime) math_tanh(call FunctionCall) Value { return floatToValue(math.Tanh(call.Argument(0).ToFloat())) } func (r *Runtime) math_trunc(call FunctionCall) Value { arg := call.Argument(0) if i, ok := arg.(valueInt); ok { return i } return floatToValue(math.Trunc(arg.ToFloat())) } func createMathTemplate() *objectTemplate { t := newObjectTemplate() t.protoFactory = func(r *Runtime) *Object { return r.global.ObjectPrototype } t.putStr("E", func(r *Runtime) Value { return valueProp(valueFloat(math.E), false, false, false) }) t.putStr("LN10", func(r *Runtime) Value { return valueProp(valueFloat(math.Ln10), false, false, false) }) t.putStr("LN2", func(r *Runtime) Value { return valueProp(valueFloat(math.Ln2), false, false, false) }) t.putStr("LOG10E", func(r *Runtime) Value { return valueProp(valueFloat(math.Log10E), false, false, false) }) t.putStr("LOG2E", func(r *Runtime) Value { return valueProp(valueFloat(math.Log2E), false, false, false) }) t.putStr("PI", func(r *Runtime) Value { return valueProp(valueFloat(math.Pi), false, false, false) }) t.putStr("SQRT1_2", func(r *Runtime) Value { return valueProp(valueFloat(sqrt1_2), false, false, false) }) t.putStr("SQRT2", func(r *Runtime) Value { return valueProp(valueFloat(math.Sqrt2), false, false, false) }) t.putSym(SymToStringTag, func(r *Runtime) Value { return valueProp(asciiString(classMath), false, false, true) }) t.putStr("abs", func(r *Runtime) Value { return r.methodProp(r.math_abs, "abs", 1) }) t.putStr("acos", func(r *Runtime) Value { return r.methodProp(r.math_acos, "acos", 1) }) t.putStr("acosh", func(r *Runtime) Value { return r.methodProp(r.math_acosh, "acosh", 1) }) t.putStr("asin", func(r *Runtime) Value { return r.methodProp(r.math_asin, "asin", 1) }) t.putStr("asinh", func(r *Runtime) Value { return r.methodProp(r.math_asinh, "asinh", 1) }) t.putStr("atan", func(r *Runtime) Value { return r.methodProp(r.math_atan, "atan", 1) }) t.putStr("atanh", func(r *Runtime) Value { return r.methodProp(r.math_atanh, "atanh", 1) }) t.putStr("atan2", func(r *Runtime) Value { return r.methodProp(r.math_atan2, "atan2", 2) }) t.putStr("cbrt", func(r *Runtime) Value { return r.methodProp(r.math_cbrt, "cbrt", 1) }) t.putStr("ceil", func(r *Runtime) Value { return r.methodProp(r.math_ceil, "ceil", 1) }) t.putStr("clz32", func(r *Runtime) Value { return r.methodProp(r.math_clz32, "clz32", 1) }) t.putStr("cos", func(r *Runtime) Value { return r.methodProp(r.math_cos, "cos", 1) }) t.putStr("cosh", func(r *Runtime) Value { return r.methodProp(r.math_cosh, "cosh", 1) }) t.putStr("exp", func(r *Runtime) Value { return r.methodProp(r.math_exp, "exp", 1) }) t.putStr("expm1", func(r *Runtime) Value { return r.methodProp(r.math_expm1, "expm1", 1) }) t.putStr("floor", func(r *Runtime) Value { return r.methodProp(r.math_floor, "floor", 1) }) t.putStr("fround", func(r *Runtime) Value { return r.methodProp(r.math_fround, "fround", 1) }) t.putStr("hypot", func(r *Runtime) Value { return r.methodProp(r.math_hypot, "hypot", 2) }) t.putStr("imul", func(r *Runtime) Value { return r.methodProp(r.math_imul, "imul", 2) }) t.putStr("log", func(r *Runtime) Value { return r.methodProp(r.math_log, "log", 1) }) t.putStr("log1p", func(r *Runtime) Value { return r.methodProp(r.math_log1p, "log1p", 1) }) t.putStr("log10", func(r *Runtime) Value { return r.methodProp(r.math_log10, "log10", 1) }) t.putStr("log2", func(r *Runtime) Value { return r.methodProp(r.math_log2, "log2", 1) }) t.putStr("max", func(r *Runtime) Value { return r.methodProp(r.math_max, "max", 2) }) t.putStr("min", func(r *Runtime) Value { return r.methodProp(r.math_min, "min", 2) }) t.putStr("pow", func(r *Runtime) Value { return r.methodProp(r.math_pow, "pow", 2) }) t.putStr("random", func(r *Runtime) Value { return r.methodProp(r.math_random, "random", 0) }) t.putStr("round", func(r *Runtime) Value { return r.methodProp(r.math_round, "round", 1) }) t.putStr("sign", func(r *Runtime) Value { return r.methodProp(r.math_sign, "sign", 1) }) t.putStr("sin", func(r *Runtime) Value { return r.methodProp(r.math_sin, "sin", 1) }) t.putStr("sinh", func(r *Runtime) Value { return r.methodProp(r.math_sinh, "sinh", 1) }) t.putStr("sqrt", func(r *Runtime) Value { return r.methodProp(r.math_sqrt, "sqrt", 1) }) t.putStr("tan", func(r *Runtime) Value { return r.methodProp(r.math_tan, "tan", 1) }) t.putStr("tanh", func(r *Runtime) Value { return r.methodProp(r.math_tanh, "tanh", 1) }) t.putStr("trunc", func(r *Runtime) Value { return r.methodProp(r.math_trunc, "trunc", 1) }) return t } var mathTemplate *objectTemplate var mathTemplateOnce sync.Once func getMathTemplate() *objectTemplate { mathTemplateOnce.Do(func() { mathTemplate = createMathTemplate() }) return mathTemplate } func (r *Runtime) getMath() *Object { ret := r.global.Math if ret == nil { ret = &Object{runtime: r} r.global.Math = ret r.newTemplatedObject(getMathTemplate(), ret) } return ret }