v/vlib/strconv/f32_f64_to_string_test.v

173 lines
3.0 KiB
V
Raw Normal View History

import strconv
2020-02-25 11:12:37 +01:00
/**********************************************************************
*
* Float to string Test
*
**********************************************************************/
import math
union Ufloat32 {
mut:
f f32 = f32(0)
b u32
}
union Ufloat64 {
mut:
f f64 = f64(0)
b u64
}
fn f64_from_bits1(b u64) f64 {
mut x := Ufloat64{}
x.b = b
//C.printf("bin: %016llx\n",x.f)
return x.f
}
fn f32_from_bits1(b u32) f32 {
mut x := Ufloat32{}
x.b = b
//C.printf("bin: %08x\n",x.f)
return x.f
}
fn test_float_to_str() {
test_cases_f32 := [
f32_from_bits1(0x0000_0000), // +0
f32_from_bits1(0x8000_0000), // -0
f32_from_bits1(0xFFC0_0001), // sNan
f32_from_bits1(0xFF80_0001), // qNan
f32_from_bits1(0x7F80_0000), // +inf
f32_from_bits1(0xFF80_0000), // -inf
1,
-1,
10,
-10,
0.3,
-0.3,
1000000,
123456.7,
123e35,
-123.45,
1e23,
f32_from_bits1(0x0080_0000), // smallest float32
math.max_f32,
383260575764816448,
]
2020-02-25 11:12:37 +01:00
exp_result_f32 := [
"0e+00",
"-0e+00",
"nan",
"nan",
"+inf",
"-inf",
"1.e+00",
"-1.e+00",
"1.e+01",
"-1.e+01",
"3.e-01",
"-3.e-01",
"1.e+06",
"1.234567e+05",
"1.23e+37",
"-1.2345e+02",
"1.e+23",
"1.1754944e-38", // aprox from 1.1754943508 × 1038,
"3.4028235e+38",
"3.8326058e+17",
]
2020-02-25 11:12:37 +01:00
test_cases_f64 := [
f64_from_bits1(0x0000_0000_0000_0000), // +0
f64_from_bits1(0x8000_0000_0000_0000), // -0
f64_from_bits1(0x7FF0_0000_0000_0001), // sNan
f64_from_bits1(0x7FF8_0000_0000_0001), // qNan
f64_from_bits1(0x7FF0_0000_0000_0000), // +inf
f64_from_bits1(0xFFF0_0000_0000_0000), // -inf
1,
-1,
10,
-10,
0.3,
-0.3,
1000000,
123456.7,
123e45,
-123.45,
1e23,
f64_from_bits1(0x0010_0000_0000_0000), // smallest float64
math.max_f32,
383260575764816448,
383260575764816448,
// C failing cases
123e300,
123e-300,
5.e-324,
-5.e-324,
]
2020-02-25 11:12:37 +01:00
exp_result_f64 := [
"0e+00",
"-0e+00",
"nan",
"nan",
"+inf",
"-inf",
"1.e+00",
"-1.e+00",
"1.e+01",
"-1.e+01",
"3.e-01",
"-3.e-01",
"1.e+06",
"1.234567e+05",
"1.23e+47",
"-1.2345e+02",
"1.e+23",
"2.2250738585072014e-308",
"3.4028234663852886e+38",
"3.8326057576481645e+17",
"3.8326057576481645e+17",
2020-02-25 11:12:37 +01:00
"1.23e+302", // this test is failed from C sprintf!!
"1.23e-298",
"5.e-324",
"-5.e-324",
]
2020-02-25 11:12:37 +01:00
// test f32
for c,x in test_cases_f32 {
println(x)
s := strconv.f32_to_str(x,8)
2020-02-25 11:12:37 +01:00
s1 := exp_result_f32[c]
//println("$s1 $s")
2020-02-25 11:12:37 +01:00
assert s == s1
}
// test f64
for c,x in test_cases_f64 {
s := strconv.f64_to_str(x,17)
2020-02-25 11:12:37 +01:00
s1 := exp_result_f64[c]
//println("$s1 $s")
assert s == s1
}
// test long format
for exp := 1 ; exp < 120 ; exp++ {
a := strconv.f64_to_str_l(("1e"+exp.str()).f64())
2020-02-25 11:12:37 +01:00
//println(a)
assert a.len == exp + 1
b := strconv.f64_to_str_l(("1e-"+exp.str()).f64())
2020-02-25 11:12:37 +01:00
//println(b)
assert b.len == exp + 2
}
// test rounding str conversion
//println( ftoa.f64_to_str(0.3456789123456, 4) )
//assert ftoa.f64_to_str(0.3456789123456, 4)=="3.4568e-01"
//assert ftoa.f32_to_str(0.345678, 3)=="3.457e-01"
2020-02-25 11:12:37 +01:00
}