--- calc-round --- #test(calc.round(calc.e, digits: 2), 2.72) #test(calc.round(calc.pi, digits: 2), 3.14) #test(type(calc.round(3.1415, digits: 2)), float) #test(type(calc.round(5, digits: 2)), int) #test(type(calc.round(decimal("3.1415"), digits: 2)), decimal) --- calc-round-large-inputs --- #test(calc.round(31114, digits: 4000000000), 31114) #test(calc.round(9223372036854775807, digits: 12), 9223372036854775807) #test(calc.round(238959235.129590203, digits: 4000000000), 238959235.129590203) #test(calc.round(1.7976931348623157e+308, digits: 12), 1.7976931348623157e+308) #test(calc.round(decimal("238959235.129590203"), digits: 4000000000), decimal("238959235.129590203")) #test(calc.round(decimal("79228162514264337593543950335"), digits: 12), decimal("79228162514264337593543950335")) --- calc-round-negative-digits --- // Error: 29-31 number must be at least zero #calc.round(243.32, digits: -2) --- calc-abs --- // Test the `abs` function. #test(calc.abs(-3), 3) #test(calc.abs(3), 3) #test(calc.abs(-0.0), 0.0) #test(calc.abs(0.0), -0.0) #test(calc.abs(-3.14), 3.14) #test(calc.abs(50%), 50%) #test(calc.abs(-25%), 25%) #test(calc.abs(decimal("4932.493249324932")), decimal("4932.493249324932")) #test(calc.abs(decimal("-12402.593295932041")), decimal("12402.593295932041")) --- calc-abs-bad-type --- // Error: 11-22 expected integer, float, length, angle, ratio, fraction, or decimal, found string #calc.abs("no number") --- calc-even-and-odd --- // Test the `even` and `odd` functions. #test(calc.even(2), true) #test(calc.odd(2), false) #test(calc.odd(-1), true) #test(calc.even(-11), false) --- calc-rem --- // Test the `rem` function. #test(calc.rem(1, 1), 0) #test(calc.rem(5, 3), 2) #test(calc.rem(5, -3), 2) #test(calc.rem(22.5, 10), 2.5) #test(calc.rem(9, 4.5), 0) #test(calc.rem(decimal("5"), -3), decimal("2")) #test(calc.rem(decimal("22.5"), decimal("10")), decimal("2.5")) #test(calc.rem(9, decimal("4.5")), decimal("0")) #test(calc.rem(decimal("7"), decimal("3")), decimal("1")) #test(calc.rem(decimal("7"), decimal("-3")), decimal("1")) #test(calc.rem(decimal("-7"), decimal("3")), decimal("-1")) #test(calc.rem(decimal("-7"), decimal("-3")), decimal("-1")) --- calc-rem-divisor-zero-1 --- // Error: 14-15 divisor must not be zero #calc.rem(5, 0) --- calc-rem-divisor-zero-2 --- // Error: 16-19 divisor must not be zero #calc.rem(3.0, 0.0) --- calc-rem-divisor-zero-3 --- // Error: 27-39 divisor must not be zero #calc.rem(decimal("4.0"), decimal("0")) --- calc-div-euclid --- // Test the `div-euclid` function. #test(calc.div-euclid(7, 3), 2) #test(calc.div-euclid(7, -3), -2) #test(calc.div-euclid(-7, 3), -3) #test(calc.div-euclid(-7, -3), 3) #test(calc.div-euclid(2.5, 2), 1) #test(calc.div-euclid(decimal("7"), decimal("3")), decimal("2")) #test(calc.div-euclid(decimal("7"), decimal("-3")), decimal("-2")) #test(calc.div-euclid(decimal("-7"), decimal("3")), decimal("-3")) #test(calc.div-euclid(decimal("-7"), decimal("-3")), decimal("3")) #test(calc.div-euclid(decimal("2.5"), decimal("2")), decimal("1")) --- calc-div-euclid-divisor-zero-1 --- // Error: 21-22 divisor must not be zero #calc.div-euclid(5, 0) --- calc-div-euclid-divisor-zero-2 --- // Error: 23-26 divisor must not be zero #calc.div-euclid(3.0, 0.0) --- calc-div-euclid-divisor-zero-3 --- // Error: 35-50 divisor must not be zero #calc.div-euclid(decimal("3.00"), decimal("0.00")) --- calc-rem-euclid --- // Test the `rem-euclid` function. #test(calc.rem-euclid(7, 3), 1) #test(calc.rem-euclid(7, -3), 1) #test(calc.rem-euclid(-7, 3), 2) #test(calc.rem-euclid(-7, -3), 2) #test(calc.rem-euclid(2.5, 2), 0.5) #test(calc.rem-euclid(decimal("7"), decimal("3")), decimal("1")) #test(calc.rem-euclid(decimal("7"), decimal("-3")), decimal("1")) #test(calc.rem-euclid(decimal("-7"), decimal("3")), decimal("2")) #test(calc.rem-euclid(decimal("-7"), decimal("-3")), decimal("2")) #test(calc.rem-euclid(decimal("2.5"), decimal("2")), decimal("0.5")) --- calc-rem-euclid-divisor-zero-1 --- // Error: 21-22 divisor must not be zero #calc.rem-euclid(5, 0) --- calc-rem-euclid-divisor-zero-2 --- // Error: 23-26 divisor must not be zero #calc.rem-euclid(3.0, 0.0) --- calc-rem-euclid-divisor-zero-3 --- // Error: 35-50 divisor must not be zero #calc.rem-euclid(decimal("3.00"), decimal("0.00")) --- calc-quo --- // Test the `quo` function. #test(calc.quo(1, 1), 1) #test(calc.quo(5, 3), 1) #test(calc.quo(5, -3), -1) #test(calc.quo(22.5, 10), 2.0) #test(calc.quo(9, 4.5), 2.0) #test(calc.quo(decimal("22.5"), 10), decimal("2")) #test(calc.quo(decimal("9"), decimal("4.5")), decimal("2")) --- calc-quo-divisor-zero-1 --- // Error: 14-15 divisor must not be zero #calc.quo(5, 0) --- calc-quo-divisor-zero-2 --- // Error: 16-19 divisor must not be zero #calc.quo(3.0, 0.0) --- calc-quo-divisor-zero-3 --- // Error: 27-41 divisor must not be zero #calc.quo(decimal("4.0"), decimal("0.0")) --- calc-min-and-max --- // Test the `min` and `max` functions. #test(calc.min(2, -4), -4) #test(calc.min(3.5, 1e2, -0.1, 3), -0.1) #test(calc.min(decimal("3.5"), 4, decimal("-3213.99999")), decimal("-3213.99999")) #test(calc.max(-3, 11), 11) #test(calc.max(decimal("3"), 45), 45) #test(calc.min("hi"), "hi") --- calc-pow-log-exp-ln --- // Test the `pow`, `log`, `exp`, and `ln` functions. #test(calc.pow(10, 0), 1) #test(calc.pow(2, 4), 16) #test(calc.pow(decimal("0.5"), 18), decimal("0.000003814697265625")) #test(calc.exp(2), calc.pow(calc.e, 2)) #test(calc.ln(10), calc.log(10, base: calc.e)) --- calc-bit-logical --- // Test the `bit-not`, `bit-and`, `bit-or` and `bit-xor` functions. #test(64.bit-not(), -65) #test(0.bit-not(), -1) #test((-56).bit-not(), 55) #test(128.bit-and(192), 128) #test(192.bit-and(224), 192) #test((-1).bit-and(325532), 325532) #test(0.bit-and(-53), 0) #test(0.bit-or(-1), -1) #test(5.bit-or(3), 7) #test((-50).bit-or(3), -49) #test(64.bit-or(32), 96) #test((-1).bit-xor(1), -2) #test(64.bit-xor(96), 32) #test((-1).bit-xor(-7), 6) #test(0.bit-xor(492), 492) --- calc-bit-shift --- // Test the `bit-lshift` and `bit-rshift` functions. #test(32.bit-lshift(2), 128) #test(694.bit-lshift(0), 694) #test(128.bit-rshift(2), 32) #test(128.bit-rshift(12345), 0) #test((-7).bit-rshift(2), -2) #test((-7).bit-rshift(12345), -1) #test(128.bit-rshift(2, logical: true), 32) #test((-7).bit-rshift(61, logical: true), 7) #test(128.bit-rshift(12345, logical: true), 0) #test((-7).bit-rshift(12345, logical: true), 0) --- calc-bit-shift-too-large --- // Error: 2-18 the result is too large #1.bit-lshift(64) --- calc-bit-lshift-negative --- // Error: 15-17 number must be at least zero #1.bit-lshift(-1) --- calc-bit-rshift-negative --- // Error: 15-17 number must be at least zero #1.bit-rshift(-1) --- calc-pow-zero-to-power-of-zero --- // Error: 2-16 zero to the power of zero is undefined #calc.pow(0, 0) --- calc-pow-exponent-too-large --- // Error: 14-31 exponent is too large #calc.pow(2, 10000000000000000) --- calc-pow-too-large --- // Error: 2-25 the result is too large #calc.pow(2, 2147483647) --- calc-pow-too-large-decimal --- // Error: 2-56 the result is too large #calc.pow(decimal("2222222222222222222222222222"), 100) --- calc-pow-bad-exponent --- // Error: 14-36 exponent may not be infinite, subnormal, or NaN #calc.pow(2, calc.pow(2.0, 10000.0)) --- calc-pow-not-real --- // Error: 2-19 the result is not a real number #calc.pow(-1, 0.5) --- calc-sqrt-not-real --- // Error: 12-14 cannot take square root of negative number #calc.sqrt(-1) --- calc-root --- #test(calc.root(12.0, 1), 12.0) #test(calc.root(9.0, 2), 3.0) #test(calc.root(27.0, 3), 3.0) #test(calc.root(-27.0, 3), -3.0) // 100^(-1/2) = (100^(1/2))^-1 = 1/sqrt(100) #test(calc.root(100.0, -2), 0.1) --- calc-root-zeroth --- // Error: 17-18 cannot take the 0th root of a number #calc.root(1.0, 0) --- calc-root-negative-even --- // Error: 24-25 negative numbers do not have a real nth root when n is even #test(calc.root(-27.0, 4), -3.0) --- calc-log-negative --- // Error: 11-13 value must be strictly positive #calc.log(-1) --- calc-log-bad-base --- // Error: 20-21 base may not be zero, NaN, infinite, or subnormal #calc.log(1, base: 0) --- calc-log-not-real --- // Error: 2-24 the result is not a real number #calc.log(10, base: -1) --- calc-fact --- // Test the `fact` function. #test(calc.fact(0), 1) #test(calc.fact(5), 120) --- calc-fact-too-large --- // Error: 2-15 the result is too large #calc.fact(21) --- calc-perm --- // Test the `perm` function. #test(calc.perm(0, 0), 1) #test(calc.perm(5, 3), 60) #test(calc.perm(5, 5), 120) #test(calc.perm(5, 6), 0) --- calc-perm-too-large --- // Error: 2-19 the result is too large #calc.perm(21, 21) --- calc-binom --- // Test the `binom` function. #test(calc.binom(0, 0), 1) #test(calc.binom(5, 3), 10) #test(calc.binom(5, 5), 1) #test(calc.binom(5, 6), 0) #test(calc.binom(6, 2), 15) --- calc-gcd --- // Test the `gcd` function. #test(calc.gcd(112, 77), 7) #test(calc.gcd(12, 96), 12) #test(calc.gcd(13, 9), 1) #test(calc.gcd(13, -9), 1) #test(calc.gcd(272557, 272557), 272557) #test(calc.gcd(0, 0), 0) #test(calc.gcd(7, 0), 7) --- calc-lcm --- // Test the `lcm` function. #test(calc.lcm(112, 77), 1232) #test(calc.lcm(12, 96), 96) #test(calc.lcm(13, 9), 117) #test(calc.lcm(13, -9), 117) #test(calc.lcm(272557, 272557), 272557) #test(calc.lcm(0, 0), 0) #test(calc.lcm(8, 0), 0) --- calc-lcm-too-large --- // Error: 2-41 the result is too large #calc.lcm(15486487489457, 4874879896543) --- calc-rounding-larger-than-max-int --- #test(calc.round(decimal("9223372036854775809.5")), decimal("9223372036854775810")) #test(calc.round(9223372036854775809.5), 9223372036854775810.0) #test(calc.floor(decimal("9223372036854775809.5")), decimal("9223372036854775809")) #test(calc.floor(9223372036854775809.5), 9223372036854775809.0) #test(calc.ceil(decimal("9223372036854775809.5")), decimal("9223372036854775810")) #test(calc.ceil(9223372036854775809.5), 9223372036854775810.0) #test(calc.trunc(decimal("9223372036854775809.5")), decimal("9223372036854775809")) #test(calc.trunc(9223372036854775809.5), 9223372036854775809.0) #test(calc.quo(decimal("9223372036854775809.5"), 1), decimal("9223372036854775809")) #test(calc.quo(9223372036854775809.5, 1), 9223372036854775809.0) --- calc-rounding-smaller-than-min-int --- #test(calc.round(decimal("-9223372036854775809.5")), decimal("-9223372036854775810")) #test(calc.round(-9223372036854775809.5), -9223372036854775810.0) #test(calc.floor(decimal("-9223372036854775809.5")), decimal("-9223372036854775810")) #test(calc.floor(-9223372036854775809.5), -9223372036854775810.0) #test(calc.ceil(decimal("-9223372036854775809.5")), decimal("-9223372036854775809")) #test(calc.ceil(-9223372036854775809.5), -9223372036854775809.0) #test(calc.trunc(decimal("-9223372036854775809.5")), decimal("-9223372036854775809")) #test(calc.trunc(-9223372036854775809.5), -9223372036854775809.0) #test(calc.quo(decimal("-9223372036854775809.5"), 1), decimal("-9223372036854775810")) #test(calc.quo(-9223372036854775809.5, 1), -9223372036854775810.0) --- calc-min-nothing --- // Error: 2-12 expected at least one value #calc.min() --- calc-min-uncomparable --- // Error: 14-18 cannot compare string and integer #calc.min(1, "hi") --- calc-max-uncomparable --- // Error: 16-19 cannot compare 1pt with 1em #calc.max(1em, 1pt) --- calc-clamp-decimal-float --- // Error: 2-37 cannot apply this operation to a decimal and a float // Hint: 2-37 if loss of precision is acceptable, explicitly cast the decimal to a float with `float(value)` #calc.clamp(decimal("10"), 5.5, 6.6)