require_relative "testbase" require_relative "../ruby/envutil" require 'bigdecimal/math' require 'thread' class TestBigDecimal < Test::Unit::TestCase include TestBigDecimalBase ROUNDING_MODE_MAP = [ [ BigDecimal::ROUND_UP, :up], [ BigDecimal::ROUND_DOWN, :down], [ BigDecimal::ROUND_DOWN, :truncate], [ BigDecimal::ROUND_HALF_UP, :half_up], [ BigDecimal::ROUND_HALF_UP, :default], [ BigDecimal::ROUND_HALF_DOWN, :half_down], [ BigDecimal::ROUND_HALF_EVEN, :half_even], [ BigDecimal::ROUND_HALF_EVEN, :banker], [ BigDecimal::ROUND_CEILING, :ceiling], [ BigDecimal::ROUND_CEILING, :ceil], [ BigDecimal::ROUND_FLOOR, :floor], ] def assert_nan(x) assert(x.nan?, "Expected #{x.inspect} to be NaN") end def assert_positive_infinite(x) assert(x.infinite?, "Expected #{x.inspect} to be positive infinite") assert_operator(x, :>, 0) end def assert_negative_infinite(x) assert(x.infinite?, "Expected #{x.inspect} to be negative infinite") assert_operator(x, :<, 0) end def assert_positive_zero(x) assert_equal(BigDecimal::SIGN_POSITIVE_ZERO, x.sign, "Expected #{x.inspect} to be positive zero") end def assert_negative_zero(x) assert_equal(BigDecimal::SIGN_NEGATIVE_ZERO, x.sign, "Expected #{x.inspect} to be negative zero") end def test_not_equal assert_not_equal BigDecimal("1"), BigDecimal.allocate end def test_global_new assert_equal(1, BigDecimal("1")) assert_equal(1, BigDecimal("1", 1)) assert_raise(ArgumentError) { BigDecimal("1", -1) } assert_raise(ArgumentError) { BigDecimal(4.2) } begin BigDecimal(4.2) rescue ArgumentError => error assert_match(/Float/, error.message) end assert_raise(ArgumentError) { BigDecimal(42.quo(7)) } begin BigDecimal(42.quo(7)) rescue ArgumentError => error assert_match(/Rational/, error.message) end end def test_global_new_with_integer assert_equal(BigDecimal("1"), BigDecimal(1)) assert_equal(BigDecimal("-1"), BigDecimal(-1)) assert_equal(BigDecimal((2**100).to_s), BigDecimal(2**100)) assert_equal(BigDecimal((-2**100).to_s), BigDecimal(-2**100)) end def test_global_new_with_rational assert_equal(BigDecimal("0.333333333333333333333"), BigDecimal(1.quo(3), 21)) assert_equal(BigDecimal("-0.333333333333333333333"), BigDecimal(-1.quo(3), 21)) assert_raise(ArgumentError) { BigDecimal(1.quo(3)) } end def test_global_new_with_float assert_equal(BigDecimal("0.1235"), BigDecimal(0.1234567, 4)) assert_equal(BigDecimal("-0.1235"), BigDecimal(-0.1234567, 4)) assert_raise(ArgumentError) { BigDecimal(0.1) } assert_raise(ArgumentError) { BigDecimal(0.1, Float::DIG + 2) } assert_nothing_raised { BigDecimal(0.1, Float::DIG + 1) } bug9214 = '[ruby-core:58858]' assert_equal(BigDecimal(-0.0, Float::DIG).sign, -1, bug9214) end def test_global_new_with_big_decimal assert_equal(BigDecimal(1), BigDecimal(BigDecimal(1))) assert_equal(BigDecimal('+0'), BigDecimal(BigDecimal('+0'))) assert_equal(BigDecimal('-0'), BigDecimal(BigDecimal('-0'))) BigDecimal.save_exception_mode do BigDecimal.mode(BigDecimal::EXCEPTION_OVERFLOW, false) BigDecimal.mode(BigDecimal::EXCEPTION_NaN, false) assert_positive_infinite(BigDecimal(BigDecimal('Infinity'))) assert_negative_infinite(BigDecimal(BigDecimal('-Infinity'))) assert_nan(BigDecimal(BigDecimal('NaN'))) end end def test_global_new_with_tainted_string Thread.new { $SAFE = 1 BigDecimal('1'.taint) }.join end def test_new assert_equal(1, BigDecimal.new("1")) assert_equal(1, BigDecimal.new("1", 1)) assert_equal(1, BigDecimal.new(" 1 ")) assert_equal(111, BigDecimal.new("1_1_1_")) assert_equal(0, BigDecimal.new("_1_1_1")) assert_equal(10**(-1), BigDecimal.new("1E-1"), '#4825') BigDecimal.mode(BigDecimal::EXCEPTION_OVERFLOW, false) BigDecimal.mode(BigDecimal::EXCEPTION_NaN, false) assert_equal( 1, BigDecimal.new("Infinity").infinite?) assert_equal(-1, BigDecimal.new("-Infinity").infinite?) assert_equal(true, BigDecimal.new("NaN").nan?) assert_equal( 1, BigDecimal.new("1E1111111111111111111").infinite?) end def test_new_with_integer assert_equal(BigDecimal("1"), BigDecimal.new(1)) assert_equal(BigDecimal("-1"), BigDecimal.new(-1)) assert_equal(BigDecimal((2**100).to_s), BigDecimal.new(2**100)) assert_equal(BigDecimal((-2**100).to_s), BigDecimal.new(-2**100)) end def test_new_with_rational assert_equal(BigDecimal("0.333333333333333333333"), BigDecimal.new(1.quo(3), 21)) assert_equal(BigDecimal("-0.333333333333333333333"), BigDecimal.new(-1.quo(3), 21)) assert_raise(ArgumentError) { BigDecimal.new(1.quo(3)) } end def test_new_with_float assert_equal(BigDecimal("0.1235"), BigDecimal(0.1234567, 4)) assert_equal(BigDecimal("-0.1235"), BigDecimal(-0.1234567, 4)) assert_raise(ArgumentError) { BigDecimal.new(0.1) } assert_raise(ArgumentError) { BigDecimal.new(0.1, Float::DIG + 2) } assert_nothing_raised { BigDecimal.new(0.1, Float::DIG + 1) } end def test_new_with_big_decimal assert_equal(BigDecimal(1), BigDecimal.new(BigDecimal(1))) assert_equal(BigDecimal('+0'), BigDecimal.new(BigDecimal('+0'))) assert_equal(BigDecimal('-0'), BigDecimal.new(BigDecimal('-0'))) BigDecimal.save_exception_mode do BigDecimal.mode(BigDecimal::EXCEPTION_OVERFLOW, false) BigDecimal.mode(BigDecimal::EXCEPTION_NaN, false) assert_positive_infinite(BigDecimal.new(BigDecimal('Infinity'))) assert_negative_infinite(BigDecimal.new(BigDecimal('-Infinity'))) assert_nan(BigDecimal(BigDecimal.new('NaN'))) end end def test_new_with_tainted_string Thread.new { $SAFE = 1 BigDecimal.new('1'.taint) }.join end def _test_mode(type) BigDecimal.mode(type, true) assert_raise(FloatDomainError) { yield } BigDecimal.mode(type, false) assert_nothing_raised { yield } end def test_mode assert_raise(ArgumentError) { BigDecimal.mode(BigDecimal::EXCEPTION_ALL, 1) } assert_raise(ArgumentError) { BigDecimal.mode(BigDecimal::ROUND_MODE, 256) } assert_raise(ArgumentError) { BigDecimal.mode(BigDecimal::ROUND_MODE, :xyzzy) } assert_raise(TypeError) { BigDecimal.mode(0xf000, true) } begin saved_mode = BigDecimal.mode(BigDecimal::ROUND_MODE) [ BigDecimal::ROUND_UP, BigDecimal::ROUND_DOWN, BigDecimal::ROUND_HALF_UP, BigDecimal::ROUND_HALF_DOWN, BigDecimal::ROUND_CEILING, BigDecimal::ROUND_FLOOR, BigDecimal::ROUND_HALF_EVEN, ].each do |mode| BigDecimal.mode(BigDecimal::ROUND_MODE, mode) assert_equal(mode, BigDecimal.mode(BigDecimal::ROUND_MODE)) end ensure BigDecimal.mode(BigDecimal::ROUND_MODE, saved_mode) end BigDecimal.save_rounding_mode do ROUNDING_MODE_MAP.each do |const, sym| BigDecimal.mode(BigDecimal::ROUND_MODE, sym) assert_equal(const, BigDecimal.mode(BigDecimal::ROUND_MODE)) end end end def test_thread_local_mode begin saved_mode = BigDecimal.mode(BigDecimal::ROUND_MODE) BigDecimal.mode(BigDecimal::ROUND_MODE, BigDecimal::ROUND_UP) Thread.start { BigDecimal.mode(BigDecimal::ROUND_MODE, BigDecimal::ROUND_HALF_EVEN) assert_equal(BigDecimal::ROUND_HALF_EVEN, BigDecimal.mode(BigDecimal::ROUND_MODE)) }.join assert_equal(BigDecimal::ROUND_UP, BigDecimal.mode(BigDecimal::ROUND_MODE)) ensure BigDecimal.mode(BigDecimal::ROUND_MODE, saved_mode) end end def test_save_exception_mode BigDecimal.mode(BigDecimal::EXCEPTION_OVERFLOW, false) mode = BigDecimal.mode(BigDecimal::EXCEPTION_OVERFLOW) BigDecimal.save_exception_mode do BigDecimal.mode(BigDecimal::EXCEPTION_OVERFLOW, true) end assert_equal(mode, BigDecimal.mode(BigDecimal::EXCEPTION_OVERFLOW)) BigDecimal.mode(BigDecimal::ROUND_MODE, BigDecimal::ROUND_FLOOR) BigDecimal.save_exception_mode do BigDecimal.mode(BigDecimal::ROUND_MODE, BigDecimal::ROUND_HALF_EVEN) end assert_equal(BigDecimal::ROUND_HALF_EVEN, BigDecimal.mode(BigDecimal::ROUND_MODE)) assert_equal(42, BigDecimal.save_exception_mode { 42 }) end def test_save_rounding_mode BigDecimal.mode(BigDecimal::ROUND_MODE, BigDecimal::ROUND_FLOOR) BigDecimal.save_rounding_mode do BigDecimal.mode(BigDecimal::ROUND_MODE, BigDecimal::ROUND_HALF_EVEN) end assert_equal(BigDecimal::ROUND_FLOOR, BigDecimal.mode(BigDecimal::ROUND_MODE)) assert_equal(42, BigDecimal.save_rounding_mode { 42 }) end def test_save_limit begin old = BigDecimal.limit BigDecimal.limit(100) BigDecimal.save_limit do BigDecimal.limit(200) end assert_equal(100, BigDecimal.limit); ensure BigDecimal.limit(old) end assert_equal(42, BigDecimal.save_limit { 42 }) end def test_exception_nan _test_mode(BigDecimal::EXCEPTION_NaN) { BigDecimal.new("NaN") } end def test_exception_infinity _test_mode(BigDecimal::EXCEPTION_INFINITY) { BigDecimal.new("Infinity") } end def test_exception_underflow _test_mode(BigDecimal::EXCEPTION_UNDERFLOW) do x = BigDecimal.new("0.1") 100.times do x *= x end end end def test_exception_overflow _test_mode(BigDecimal::EXCEPTION_OVERFLOW) do x = BigDecimal.new("10") 100.times do x *= x end end end def test_exception_zerodivide BigDecimal.mode(BigDecimal::EXCEPTION_OVERFLOW, false) _test_mode(BigDecimal::EXCEPTION_ZERODIVIDE) { 1 / BigDecimal.new("0") } _test_mode(BigDecimal::EXCEPTION_ZERODIVIDE) { -1 / BigDecimal.new("0") } end def test_round_up n4 = BigDecimal.new("4") # n4 / 9 = 0.44444... n5 = BigDecimal.new("5") # n5 / 9 = 0.55555... n6 = BigDecimal.new("6") # n6 / 9 = 0.66666... m4, m5, m6 = -n4, -n5, -n6 n2h = BigDecimal.new("2.5") n3h = BigDecimal.new("3.5") m2h, m3h = -n2h, -n3h BigDecimal.mode(BigDecimal::ROUND_MODE, BigDecimal::ROUND_UP) assert_operator(n4, :<, n4 / 9 * 9) assert_operator(n5, :<, n5 / 9 * 9) assert_operator(n6, :<, n6 / 9 * 9) assert_operator(m4, :>, m4 / 9 * 9) assert_operator(m5, :>, m5 / 9 * 9) assert_operator(m6, :>, m6 / 9 * 9) assert_equal(3, n2h.round) assert_equal(4, n3h.round) assert_equal(-3, m2h.round) assert_equal(-4, m3h.round) BigDecimal.mode(BigDecimal::ROUND_MODE, BigDecimal::ROUND_DOWN) assert_operator(n4, :>, n4 / 9 * 9) assert_operator(n5, :>, n5 / 9 * 9) assert_operator(n6, :>, n6 / 9 * 9) assert_operator(m4, :<, m4 / 9 * 9) assert_operator(m5, :<, m5 / 9 * 9) assert_operator(m6, :<, m6 / 9 * 9) assert_equal(2, n2h.round) assert_equal(3, n3h.round) assert_equal(-2, m2h.round) assert_equal(-3, m3h.round) BigDecimal.mode(BigDecimal::ROUND_MODE, BigDecimal::ROUND_HALF_UP) assert_operator(n4, :>, n4 / 9 * 9) assert_operator(n5, :<, n5 / 9 * 9) assert_operator(n6, :<, n6 / 9 * 9) assert_operator(m4, :<, m4 / 9 * 9) assert_operator(m5, :>, m5 / 9 * 9) assert_operator(m6, :>, m6 / 9 * 9) assert_equal(3, n2h.round) assert_equal(4, n3h.round) assert_equal(-3, m2h.round) assert_equal(-4, m3h.round) BigDecimal.mode(BigDecimal::ROUND_MODE, BigDecimal::ROUND_HALF_DOWN) assert_operator(n4, :>, n4 / 9 * 9) assert_operator(n5, :>, n5 / 9 * 9) assert_operator(n6, :<, n6 / 9 * 9) assert_operator(m4, :<, m4 / 9 * 9) assert_operator(m5, :<, m5 / 9 * 9) assert_operator(m6, :>, m6 / 9 * 9) assert_equal(2, n2h.round) assert_equal(3, n3h.round) assert_equal(-2, m2h.round) assert_equal(-3, m3h.round) BigDecimal.mode(BigDecimal::ROUND_MODE, BigDecimal::ROUND_HALF_EVEN) assert_operator(n4, :>, n4 / 9 * 9) assert_operator(n5, :<, n5 / 9 * 9) assert_operator(n6, :<, n6 / 9 * 9) assert_operator(m4, :<, m4 / 9 * 9) assert_operator(m5, :>, m5 / 9 * 9) assert_operator(m6, :>, m6 / 9 * 9) assert_equal(2, n2h.round) assert_equal(4, n3h.round) assert_equal(-2, m2h.round) assert_equal(-4, m3h.round) BigDecimal.mode(BigDecimal::ROUND_MODE, BigDecimal::ROUND_CEILING) assert_operator(n4, :<, n4 / 9 * 9) assert_operator(n5, :<, n5 / 9 * 9) assert_operator(n6, :<, n6 / 9 * 9) assert_operator(m4, :<, m4 / 9 * 9) assert_operator(m5, :<, m5 / 9 * 9) assert_operator(m6, :<, m6 / 9 * 9) assert_equal(3, n2h.round) assert_equal(4, n3h.round) assert_equal(-2, m2h.round) assert_equal(-3, m3h.round) BigDecimal.mode(BigDecimal::ROUND_MODE, BigDecimal::ROUND_FLOOR) assert_operator(n4, :>, n4 / 9 * 9) assert_operator(n5, :>, n5 / 9 * 9) assert_operator(n6, :>, n6 / 9 * 9) assert_operator(m4, :>, m4 / 9 * 9) assert_operator(m5, :>, m5 / 9 * 9) assert_operator(m6, :>, m6 / 9 * 9) assert_equal(2, n2h.round) assert_equal(3, n3h.round) assert_equal(-3, m2h.round) assert_equal(-4, m3h.round) end def test_zero_p assert_equal(true, BigDecimal.new("0").zero?) assert_equal(false, BigDecimal.new("1").zero?) assert_equal(true, BigDecimal.new("0E200000000000000").zero?) end def test_nonzero_p assert_equal(nil, BigDecimal.new("0").nonzero?) assert_equal(BigDecimal.new("1"), BigDecimal.new("1").nonzero?) end def test_double_fig assert_kind_of(Integer, BigDecimal.double_fig) end def test_cmp n1 = BigDecimal.new("1") n2 = BigDecimal.new("2") assert_equal( 0, n1 <=> n1) assert_equal( 1, n2 <=> n1) assert_equal(-1, n1 <=> n2) assert_operator(n1, :==, n1) assert_operator(n1, :!=, n2) assert_operator(n1, :<, n2) assert_operator(n1, :<=, n1) assert_operator(n1, :<=, n2) assert_operator(n2, :>, n1) assert_operator(n2, :>=, n1) assert_operator(n1, :>=, n1) assert_operator(BigDecimal.new("-0"), :==, BigDecimal.new("0")) assert_operator(BigDecimal.new("0"), :<, BigDecimal.new("1")) assert_operator(BigDecimal.new("1"), :>, BigDecimal.new("0")) assert_operator(BigDecimal.new("1"), :>, BigDecimal.new("-1")) assert_operator(BigDecimal.new("-1"), :<, BigDecimal.new("1")) assert_operator(BigDecimal.new((2**100).to_s), :>, BigDecimal.new("1")) assert_operator(BigDecimal.new("1"), :<, BigDecimal.new((2**100).to_s)) BigDecimal.mode(BigDecimal::EXCEPTION_OVERFLOW, false) inf = BigDecimal.new("Infinity") assert_operator(inf, :>, 1) assert_operator(1, :<, inf) assert_operator(BigDecimal("1E-1"), :==, 10**(-1), '#4825') assert_equal(0, BigDecimal("1E-1") <=> 10**(-1), '#4825') end def test_cmp_issue9192 bug9192 = '[ruby-core:58756] [#9192]' operators = { :== => :==, :< => :>, :> => :<, :<= => :>=, :>= => :<= } 5.upto(8) do |i| s = "706.0#{i}" d = BigDecimal(s) f = s.to_f operators.each do |op, inv| assert_equal(d.send(op, f), f.send(inv, d), "(BigDecimal(#{s.inspect}) #{op} #{s}) and (#{s} #{inv} BigDecimal(#{s.inspect})) is different #{bug9192}") end end end def test_cmp_nan n1 = BigDecimal.new("1") BigDecimal.mode(BigDecimal::EXCEPTION_NaN, false) assert_equal(nil, BigDecimal.new("NaN") <=> n1) assert_equal(false, BigDecimal.new("NaN") > n1) end def test_cmp_failing_coercion n1 = BigDecimal.new("1") assert_equal(nil, n1 <=> nil) assert_raise(ArgumentError){n1 > nil} end def test_cmp_coerce n1 = BigDecimal.new("1") n2 = BigDecimal.new("2") o1 = Object.new; def o1.coerce(x); [x, BigDecimal.new("1")]; end o2 = Object.new; def o2.coerce(x); [x, BigDecimal.new("2")]; end assert_equal( 0, n1 <=> o1) assert_equal( 1, n2 <=> o1) assert_equal(-1, n1 <=> o2) assert_operator(n1, :==, o1) assert_operator(n1, :!=, o2) assert_operator(n1, :<, o2) assert_operator(n1, :<=, o1) assert_operator(n1, :<=, o2) assert_operator(n2, :>, o1) assert_operator(n2, :>=, o1) assert_operator(n1, :>=, 1) end def test_cmp_bignum assert_operator(BigDecimal.new((2**100).to_s), :==, 2**100) end def test_cmp_data d = Time.now; def d.coerce(x); [x, x]; end assert_operator(BigDecimal.new((2**100).to_s), :==, d) end def test_precs a = BigDecimal.new("1").precs assert_instance_of(Array, a) assert_equal(2, a.size) assert_kind_of(Integer, a[0]) assert_kind_of(Integer, a[1]) end def test_hash a = [] b = BigDecimal.new("1") 10.times { a << b *= 10 } h = {} a.each_with_index {|x, i| h[x] = i } a.each_with_index do |x, i| assert_equal(i, h[x]) end end def test_marshal s = Marshal.dump(BigDecimal("1", 1)) assert_equal(BigDecimal("1", 1), Marshal.load(s)) # corrupt data s = s.gsub(/BigDecimal.*\z/m) {|x| x.gsub(/\d/m, "-") } assert_raise(TypeError) { Marshal.load(s) } end def test_finite_infinite_nan BigDecimal.mode(BigDecimal::EXCEPTION_OVERFLOW, false) BigDecimal.mode(BigDecimal::EXCEPTION_ZERODIVIDE, false) x = BigDecimal.new("0") assert_equal(true, x.finite?) assert_equal(nil, x.infinite?) assert_equal(false, x.nan?) y = 1 / x assert_equal(false, y.finite?) assert_equal(1, y.infinite?) assert_equal(false, y.nan?) y = -1 / x assert_equal(false, y.finite?) assert_equal(-1, y.infinite?) assert_equal(false, y.nan?) BigDecimal.mode(BigDecimal::EXCEPTION_NaN, false) y = 0 / x assert_equal(false, y.finite?) assert_equal(nil, y.infinite?) assert_equal(true, y.nan?) end def test_to_i BigDecimal.mode(BigDecimal::EXCEPTION_OVERFLOW, false) BigDecimal.mode(BigDecimal::EXCEPTION_NaN, false) x = BigDecimal.new("0") assert_kind_of(Integer, x.to_i) assert_equal(0, x.to_i) assert_raise(FloatDomainError){( 1 / x).to_i} assert_raise(FloatDomainError){(-1 / x).to_i} assert_raise(FloatDomainError) {( 0 / x).to_i} x = BigDecimal.new("1") assert_equal(1, x.to_i) x = BigDecimal.new((2**100).to_s) assert_equal(2**100, x.to_i) end def test_to_f BigDecimal.mode(BigDecimal::EXCEPTION_OVERFLOW, false) BigDecimal.mode(BigDecimal::EXCEPTION_NaN, false) BigDecimal.mode(BigDecimal::EXCEPTION_ZERODIVIDE, false) x = BigDecimal.new("0") assert_instance_of(Float, x.to_f) assert_equal(0.0, x.to_f) assert_equal( 1.0 / 0.0, ( 1 / x).to_f) assert_equal(-1.0 / 0.0, (-1 / x).to_f) assert_equal(true, ( 0 / x).to_f.nan?) x = BigDecimal.new("1") assert_equal(1.0, x.to_f) x = BigDecimal.new((2**100).to_s) assert_equal((2**100).to_f, x.to_f) x = BigDecimal.new("1" + "0" * 10000) assert_equal(0, BigDecimal.new("-0").to_f) BigDecimal.mode(BigDecimal::EXCEPTION_OVERFLOW, true) assert_raise(FloatDomainError) { x.to_f } BigDecimal.mode(BigDecimal::EXCEPTION_OVERFLOW, false) assert_kind_of(Float, x .to_f) assert_kind_of(Float, (-x).to_f) bug6944 = '[ruby-core:47342]' BigDecimal.mode(BigDecimal::EXCEPTION_UNDERFLOW, true) x = "1e#{Float::MIN_10_EXP - 2*Float::DIG}" assert_raise(FloatDomainError, x) {BigDecimal(x).to_f} x = "-#{x}" assert_raise(FloatDomainError, x) {BigDecimal(x).to_f} x = "1e#{Float::MIN_10_EXP - Float::DIG}" assert_nothing_raised(FloatDomainError, x) { assert_in_delta(0.0, BigDecimal(x).to_f, 10**Float::MIN_10_EXP, bug6944) } x = "-#{x}" assert_nothing_raised(FloatDomainError, x) { assert_in_delta(0.0, BigDecimal(x).to_f, 10**Float::MIN_10_EXP, bug6944) } BigDecimal.mode(BigDecimal::EXCEPTION_UNDERFLOW, false) x = "1e#{Float::MIN_10_EXP - 2*Float::DIG}" assert_equal( 0.0, BigDecimal(x).to_f, x) x = "-#{x}" assert_equal(-0.0, BigDecimal(x).to_f, x) x = "1e#{Float::MIN_10_EXP - Float::DIG}" assert_nothing_raised(FloatDomainError, x) { assert_in_delta(0.0, BigDecimal(x).to_f, 10**Float::MIN_10_EXP, bug6944) } x = "-#{x}" assert_nothing_raised(FloatDomainError, x) { assert_in_delta(0.0, BigDecimal(x).to_f, 10**Float::MIN_10_EXP, bug6944) } assert_equal( 0.0, BigDecimal( '9e-325').to_f) assert_equal( 0.0, BigDecimal( '10e-325').to_f) assert_equal(-0.0, BigDecimal( '-9e-325').to_f) assert_equal(-0.0, BigDecimal('-10e-325').to_f) end def test_coerce a, b = BigDecimal.new("1").coerce(1.0) assert_instance_of(BigDecimal, a) assert_instance_of(BigDecimal, b) assert_equal(2, 1 + BigDecimal.new("1"), '[ruby-core:25697]') a, b = BigDecimal("1").coerce(1.quo(10)) assert_equal(BigDecimal("0.1"), a, '[ruby-core:34318]') a, b = BigDecimal("0.11111").coerce(1.quo(3)) assert_equal(BigDecimal("0." + "3"*a.precs[0]), a) assert_nothing_raised(TypeError, '#7176') do BigDecimal.new('1') + Rational(1) end end def test_uplus x = BigDecimal.new("1") assert_equal(x, x.send(:+@)) end def test_add x = BigDecimal.new("1") assert_equal(BigDecimal.new("2"), x + x) assert_equal(1, BigDecimal.new("0") + 1) assert_equal(1, x + 0) assert_equal(BigDecimal::SIGN_POSITIVE_ZERO, (BigDecimal.new("0") + 0).sign) assert_equal(BigDecimal::SIGN_POSITIVE_ZERO, (BigDecimal.new("-0") + 0).sign) assert_equal(BigDecimal::SIGN_NEGATIVE_ZERO, (BigDecimal.new("-0") + BigDecimal.new("-0")).sign) x = BigDecimal.new((2**100).to_s) assert_equal(BigDecimal.new((2**100+1).to_s), x + 1) end def test_sub x = BigDecimal.new("1") assert_equal(BigDecimal.new("0"), x - x) assert_equal(-1, BigDecimal.new("0") - 1) assert_equal(1, x - 0) assert_equal(BigDecimal::SIGN_POSITIVE_ZERO, (BigDecimal.new("0") - 0).sign) assert_equal(BigDecimal::SIGN_NEGATIVE_ZERO, (BigDecimal.new("-0") - 0).sign) assert_equal(BigDecimal::SIGN_POSITIVE_ZERO, (BigDecimal.new("-0") - BigDecimal.new("-0")).sign) x = BigDecimal.new((2**100).to_s) assert_equal(BigDecimal.new((2**100-1).to_s), x - 1) end def test_sub_with_float assert_kind_of(BigDecimal, BigDecimal.new("3") - 1.0) end def test_sub_with_rational assert_kind_of(BigDecimal, BigDecimal.new("3") - 1.quo(3)) end def test_mult x = BigDecimal.new((2**100).to_s) assert_equal(BigDecimal.new((2**100 * 3).to_s), (x * 3).to_i) assert_equal(x, (x * 1).to_i) assert_equal(x, (BigDecimal("1") * x).to_i) assert_equal(BigDecimal.new((2**200).to_s), (x * x).to_i) end def test_mult_with_float assert_kind_of(BigDecimal, BigDecimal.new("3") * 1.5) end def test_mult_with_rational assert_kind_of(BigDecimal, BigDecimal.new("3") * 1.quo(3)) end def test_div x = BigDecimal.new((2**100).to_s) assert_equal(BigDecimal.new((2**100 / 3).to_s), (x / 3).to_i) assert_equal(BigDecimal::SIGN_POSITIVE_ZERO, (BigDecimal.new("0") / 1).sign) assert_equal(BigDecimal::SIGN_NEGATIVE_ZERO, (BigDecimal.new("-0") / 1).sign) assert_equal(2, BigDecimal.new("2") / 1) assert_equal(-2, BigDecimal.new("2") / -1) assert_equal(BigDecimal('1486.868686869'), BigDecimal('1472.0') / BigDecimal('0.99'), '[ruby-core:59365] [#9316]') assert_equal(4.124045235, BigDecimal('0.9932') / (700 * BigDecimal('0.344045') / BigDecimal('1000.0')), '[#9305]') end def test_div_with_float assert_kind_of(BigDecimal, BigDecimal.new("3") / 1.5) end def test_div_with_rational assert_kind_of(BigDecimal, BigDecimal.new("3") / 1.quo(3)) end def test_mod x = BigDecimal.new((2**100).to_s) assert_equal(1, x % 3) assert_equal(2, (-x) % 3) assert_equal(-2, x % -3) assert_equal(-1, (-x) % -3) end def test_mod_with_float assert_kind_of(BigDecimal, BigDecimal.new("3") % 1.5) end def test_mod_with_rational assert_kind_of(BigDecimal, BigDecimal.new("3") % 1.quo(3)) end def test_remainder x = BigDecimal.new((2**100).to_s) assert_equal(1, x.remainder(3)) assert_equal(-1, (-x).remainder(3)) assert_equal(1, x.remainder(-3)) assert_equal(-1, (-x).remainder(-3)) end def test_remainder_with_float assert_kind_of(BigDecimal, BigDecimal.new("3").remainder(1.5)) end def test_remainder_with_rational assert_kind_of(BigDecimal, BigDecimal.new("3").remainder(1.quo(3))) end def test_divmod x = BigDecimal.new((2**100).to_s) assert_equal([(x / 3).floor, 1], x.divmod(3)) assert_equal([(-x / 3).floor, 2], (-x).divmod(3)) assert_equal([0, 0], BigDecimal.new("0").divmod(2)) BigDecimal.mode(BigDecimal::EXCEPTION_NaN, false) assert_raise(ZeroDivisionError){BigDecimal.new("0").divmod(0)} end def test_add_bigdecimal x = BigDecimal.new((2**100).to_s) assert_equal(3000000000000000000000000000000, x.add(x, 1)) assert_equal(2500000000000000000000000000000, x.add(x, 2)) assert_equal(2540000000000000000000000000000, x.add(x, 3)) end def test_sub_bigdecimal x = BigDecimal.new((2**100).to_s) assert_equal(1000000000000000000000000000000, x.sub(1, 1)) assert_equal(1300000000000000000000000000000, x.sub(1, 2)) assert_equal(1270000000000000000000000000000, x.sub(1, 3)) end def test_mult_bigdecimal x = BigDecimal.new((2**100).to_s) assert_equal(4000000000000000000000000000000, x.mult(3, 1)) assert_equal(3800000000000000000000000000000, x.mult(3, 2)) assert_equal(3800000000000000000000000000000, x.mult(3, 3)) end def test_div_bigdecimal x = BigDecimal.new((2**100).to_s) assert_equal(422550200076076467165567735125, x.div(3)) assert_equal(400000000000000000000000000000, x.div(3, 1)) assert_equal(420000000000000000000000000000, x.div(3, 2)) assert_equal(423000000000000000000000000000, x.div(3, 3)) BigDecimal.save_exception_mode do BigDecimal.mode(BigDecimal::EXCEPTION_INFINITY, false) assert_equal(0, BigDecimal("0").div(BigDecimal("Infinity"))) end end def test_abs_bigdecimal x = BigDecimal.new((2**100).to_s) assert_equal(1267650600228229401496703205376, x.abs) x = BigDecimal.new("-" + (2**100).to_s) assert_equal(1267650600228229401496703205376, x.abs) x = BigDecimal.new("0") assert_equal(0, x.abs) BigDecimal.mode(BigDecimal::EXCEPTION_NaN, false) x = BigDecimal.new("NaN") assert_equal(true, x.abs.nan?) end def test_sqrt_bigdecimal x = BigDecimal.new("0.09") assert_in_delta(0.3, x.sqrt(1), 0.001) x = BigDecimal.new((2**100).to_s) y = BigDecimal("1125899906842624") e = y.exponent assert_equal(true, (x.sqrt(100) - y).abs < BigDecimal("1E#{e-100}")) assert_equal(true, (x.sqrt(200) - y).abs < BigDecimal("1E#{e-200}")) assert_equal(true, (x.sqrt(300) - y).abs < BigDecimal("1E#{e-300}")) x = BigDecimal.new("-" + (2**100).to_s) assert_raise(FloatDomainError) { x.sqrt(1) } x = BigDecimal.new((2**200).to_s) assert_equal(2**100, x.sqrt(1)) BigDecimal.mode(BigDecimal::EXCEPTION_OVERFLOW, false) BigDecimal.mode(BigDecimal::EXCEPTION_NaN, false) assert_raise(FloatDomainError) { BigDecimal.new("NaN").sqrt(1) } assert_equal(0, BigDecimal.new("0").sqrt(1)) assert_equal(1, BigDecimal.new("1").sqrt(1)) end def test_sqrt_5266 x = BigDecimal('2' + '0'*100) assert_equal('0.14142135623730950488016887242096980785696718753769480731', x.sqrt(56).to_s(56).split(' ')[0]) assert_equal('0.1414213562373095048801688724209698078569671875376948073', x.sqrt(55).to_s(55).split(' ')[0]) x = BigDecimal('2' + '0'*200) assert_equal('0.14142135623730950488016887242096980785696718753769480731766797379907324784621070388503875343276415727350138462', x.sqrt(110).to_s(110).split(' ')[0]) assert_equal('0.1414213562373095048801688724209698078569671875376948073176679737990732478462107038850387534327641572735013846', x.sqrt(109).to_s(109).split(' ')[0]) end def test_fix x = BigDecimal.new("1.1") assert_equal(1, x.fix) end def test_frac x = BigDecimal.new("1.1") assert_equal(0.1, x.frac) assert_equal(0.1, BigDecimal.new("0.1").frac) BigDecimal.mode(BigDecimal::EXCEPTION_NaN, false) assert_equal(true, BigDecimal.new("NaN").frac.nan?) end def test_round assert_equal(3, BigDecimal.new("3.14159").round) assert_equal(9, BigDecimal.new("8.7").round) assert_equal(3.142, BigDecimal.new("3.14159").round(3)) assert_equal(13300.0, BigDecimal.new("13345.234").round(-2)) x = BigDecimal.new("111.111") assert_equal(111 , x.round) assert_equal(111.1 , x.round(1)) assert_equal(111.11 , x.round(2)) assert_equal(111.111, x.round(3)) assert_equal(111.111, x.round(4)) assert_equal(110 , x.round(-1)) assert_equal(100 , x.round(-2)) assert_equal( 0 , x.round(-3)) assert_equal( 0 , x.round(-4)) x = BigDecimal.new("2.5") assert_equal(3, x.round(0, BigDecimal::ROUND_UP)) assert_equal(2, x.round(0, BigDecimal::ROUND_DOWN)) assert_equal(3, x.round(0, BigDecimal::ROUND_HALF_UP)) assert_equal(2, x.round(0, BigDecimal::ROUND_HALF_DOWN)) assert_equal(2, x.round(0, BigDecimal::ROUND_HALF_EVEN)) assert_equal(3, x.round(0, BigDecimal::ROUND_CEILING)) assert_equal(2, x.round(0, BigDecimal::ROUND_FLOOR)) assert_raise(ArgumentError) { x.round(0, 256) } ROUNDING_MODE_MAP.each do |const, sym| assert_equal(x.round(0, const), x.round(0, sym)) end bug3803 = '[ruby-core:32136]' 15.times do |n| x = BigDecimal.new("5#{'0'*n}1") assert_equal(10**(n+2), x.round(-(n+2), BigDecimal::ROUND_HALF_DOWN), bug3803) assert_equal(10**(n+2), x.round(-(n+2), BigDecimal::ROUND_HALF_EVEN), bug3803) x = BigDecimal.new("0.5#{'0'*n}1") assert_equal(1, x.round(0, BigDecimal::ROUND_HALF_DOWN), bug3803) assert_equal(1, x.round(0, BigDecimal::ROUND_HALF_EVEN), bug3803) x = BigDecimal.new("-0.5#{'0'*n}1") assert_equal(-1, x.round(0, BigDecimal::ROUND_HALF_DOWN), bug3803) assert_equal(-1, x.round(0, BigDecimal::ROUND_HALF_EVEN), bug3803) end end def test_truncate assert_equal(3, BigDecimal.new("3.14159").truncate) assert_equal(8, BigDecimal.new("8.7").truncate) assert_equal(3.141, BigDecimal.new("3.14159").truncate(3)) assert_equal(13300.0, BigDecimal.new("13345.234").truncate(-2)) end def test_floor assert_equal(3, BigDecimal.new("3.14159").floor) assert_equal(-10, BigDecimal.new("-9.1").floor) assert_equal(3.141, BigDecimal.new("3.14159").floor(3)) assert_equal(13300.0, BigDecimal.new("13345.234").floor(-2)) end def test_ceil assert_equal(4, BigDecimal.new("3.14159").ceil) assert_equal(-9, BigDecimal.new("-9.1").ceil) assert_equal(3.142, BigDecimal.new("3.14159").ceil(3)) assert_equal(13400.0, BigDecimal.new("13345.234").ceil(-2)) end def test_to_s assert_equal('-123.45678 90123 45678 9', BigDecimal.new('-123.45678901234567890').to_s('5F')) assert_equal('+123.45678901 23456789', BigDecimal.new('123.45678901234567890').to_s('+8F')) assert_equal(' 123.4567890123456789', BigDecimal.new('123.45678901234567890').to_s(' F')) assert_equal('0.1234567890123456789E3', BigDecimal.new('123.45678901234567890').to_s) assert_equal('0.12345 67890 12345 6789E3', BigDecimal.new('123.45678901234567890').to_s(5)) end def test_split x = BigDecimal.new('-123.45678901234567890') assert_equal([-1, "1234567890123456789", 10, 3], x.split) assert_equal([1, "0", 10, 0], BigDecimal.new("0").split) assert_equal([-1, "0", 10, 0], BigDecimal.new("-0").split) BigDecimal.mode(BigDecimal::EXCEPTION_OVERFLOW, false) BigDecimal.mode(BigDecimal::EXCEPTION_NaN, false) assert_equal([0, "NaN", 10, 0], BigDecimal.new("NaN").split) assert_equal([1, "Infinity", 10, 0], BigDecimal.new("Infinity").split) assert_equal([-1, "Infinity", 10, 0], BigDecimal.new("-Infinity").split) end def test_exponent x = BigDecimal.new('-123.45678901234567890') assert_equal(3, x.exponent) end def test_inspect x = BigDecimal.new("1234.5678") prec, maxprec = x.precs assert_match(/^#$/, x.inspect) end def test_power assert_nothing_raised(TypeError, '[ruby-core:47632]') do 1000.times { BigDecimal.new('1001.10')**0.75 } end end def test_power_with_nil assert_raise(TypeError) do BigDecimal(3) ** nil end end def test_power_of_nan BigDecimal.save_exception_mode do BigDecimal.mode(BigDecimal::EXCEPTION_NaN, false) assert_nan(BigDecimal::NAN ** 0) assert_nan(BigDecimal::NAN ** 1) assert_nan(BigDecimal::NAN ** 42) assert_nan(BigDecimal::NAN ** -42) assert_nan(BigDecimal::NAN ** 42.0) assert_nan(BigDecimal::NAN ** -42.0) assert_nan(BigDecimal::NAN ** BigDecimal(42)) assert_nan(BigDecimal::NAN ** BigDecimal(-42)) assert_nan(BigDecimal::NAN ** BigDecimal::INFINITY) BigDecimal.save_exception_mode do BigDecimal.mode(BigDecimal::EXCEPTION_INFINITY, false) assert_nan(BigDecimal::NAN ** (-BigDecimal::INFINITY)) end end end def test_power_with_Bignum BigDecimal.save_exception_mode do BigDecimal.mode(BigDecimal::EXCEPTION_INFINITY, false) assert_equal(0, BigDecimal(0) ** (2**100)) assert_positive_infinite(BigDecimal(0) ** -(2**100)) assert_positive_infinite((-BigDecimal(0)) ** -(2**100)) assert_negative_infinite((-BigDecimal(0)) ** -(2**100 + 1)) assert_equal(1, BigDecimal(1) ** (2**100)) assert_positive_infinite(BigDecimal(3) ** (2**100)) assert_positive_zero(BigDecimal(3) ** (-2**100)) assert_negative_infinite(BigDecimal(-3) ** (2**100)) assert_positive_infinite(BigDecimal(-3) ** (2**100 + 1)) assert_negative_zero(BigDecimal(-3) ** (-2**100)) assert_positive_zero(BigDecimal(-3) ** (-2**100 - 1)) assert_positive_zero(BigDecimal(0.5, Float::DIG) ** (2**100)) assert_positive_infinite(BigDecimal(0.5, Float::DIG) ** (-2**100)) assert_negative_zero(BigDecimal(-0.5, Float::DIG) ** (2**100)) assert_positive_zero(BigDecimal(-0.5, Float::DIG) ** (2**100 - 1)) assert_negative_infinite(BigDecimal(-0.5, Float::DIG) ** (-2**100)) assert_positive_infinite(BigDecimal(-0.5, Float::DIG) ** (-2**100 - 1)) end end def test_power_with_BigDecimal assert_nothing_raised do assert_in_delta(3 ** 3, BigDecimal(3) ** BigDecimal(3)) end end def test_power_of_finite_with_zero x = BigDecimal(1) assert_equal(1, x ** 0) assert_equal(1, x ** 0.quo(1)) assert_equal(1, x ** 0.0) assert_equal(1, x ** BigDecimal(0)) x = BigDecimal(42) assert_equal(1, x ** 0) assert_equal(1, x ** 0.quo(1)) assert_equal(1, x ** 0.0) assert_equal(1, x ** BigDecimal(0)) x = BigDecimal(-42) assert_equal(1, x ** 0) assert_equal(1, x ** 0.quo(1)) assert_equal(1, x ** 0.0) assert_equal(1, x ** BigDecimal(0)) end def test_power_of_three x = BigDecimal(3) assert_equal(81, x ** 4) assert_equal(1.quo(81), x ** -4) assert_in_delta(1.0/81, x ** -4) end def test_power_of_zero zero = BigDecimal(0) assert_equal(0, zero ** 4) assert_equal(0, zero ** 4.quo(1)) assert_equal(0, zero ** 4.0) assert_equal(0, zero ** BigDecimal(4)) assert_equal(1, zero ** 0) assert_equal(1, zero ** 0.quo(1)) assert_equal(1, zero ** 0.0) assert_equal(1, zero ** BigDecimal(0)) BigDecimal.save_exception_mode do BigDecimal.mode(BigDecimal::EXCEPTION_INFINITY, false) BigDecimal.mode(BigDecimal::EXCEPTION_OVERFLOW, false) assert_positive_infinite(zero ** -1) assert_positive_infinite(zero ** -1.quo(1)) assert_positive_infinite(zero ** -1.0) assert_positive_infinite(zero ** BigDecimal(-1)) m_zero = BigDecimal("-0") assert_negative_infinite(m_zero ** -1) assert_negative_infinite(m_zero ** -1.quo(1)) assert_negative_infinite(m_zero ** -1.0) assert_negative_infinite(m_zero ** BigDecimal(-1)) assert_positive_infinite(m_zero ** -2) assert_positive_infinite(m_zero ** -2.quo(1)) assert_positive_infinite(m_zero ** -2.0) assert_positive_infinite(m_zero ** BigDecimal(-2)) end end def test_power_of_positive_infinity BigDecimal.save_exception_mode do BigDecimal.mode(BigDecimal::EXCEPTION_OVERFLOW, false) assert_positive_infinite(BigDecimal::INFINITY ** 3) assert_positive_infinite(BigDecimal::INFINITY ** 3.quo(1)) assert_positive_infinite(BigDecimal::INFINITY ** 3.0) assert_positive_infinite(BigDecimal::INFINITY ** BigDecimal(3)) assert_positive_infinite(BigDecimal::INFINITY ** 2) assert_positive_infinite(BigDecimal::INFINITY ** 2.quo(1)) assert_positive_infinite(BigDecimal::INFINITY ** 2.0) assert_positive_infinite(BigDecimal::INFINITY ** BigDecimal(2)) assert_positive_infinite(BigDecimal::INFINITY ** 1) assert_positive_infinite(BigDecimal::INFINITY ** 1.quo(1)) assert_positive_infinite(BigDecimal::INFINITY ** 1.0) assert_positive_infinite(BigDecimal::INFINITY ** BigDecimal(1)) assert_equal(1, BigDecimal::INFINITY ** 0) assert_equal(1, BigDecimal::INFINITY ** 0.quo(1)) assert_equal(1, BigDecimal::INFINITY ** 0.0) assert_equal(1, BigDecimal::INFINITY ** BigDecimal(0)) assert_positive_zero(BigDecimal::INFINITY ** -1) assert_positive_zero(BigDecimal::INFINITY ** -1.quo(1)) assert_positive_zero(BigDecimal::INFINITY ** -1.0) assert_positive_zero(BigDecimal::INFINITY ** BigDecimal(-1)) assert_positive_zero(BigDecimal::INFINITY ** -2) assert_positive_zero(BigDecimal::INFINITY ** -2.0) assert_positive_zero(BigDecimal::INFINITY ** BigDecimal(-2)) end end def test_power_of_negative_infinity BigDecimal.save_exception_mode do BigDecimal.mode(BigDecimal::EXCEPTION_OVERFLOW, false) assert_negative_infinite((-BigDecimal::INFINITY) ** 3) assert_negative_infinite((-BigDecimal::INFINITY) ** 3.quo(1)) assert_negative_infinite((-BigDecimal::INFINITY) ** 3.0) assert_negative_infinite((-BigDecimal::INFINITY) ** BigDecimal(3)) assert_positive_infinite((-BigDecimal::INFINITY) ** 2) assert_positive_infinite((-BigDecimal::INFINITY) ** 2.quo(1)) assert_positive_infinite((-BigDecimal::INFINITY) ** 2.0) assert_positive_infinite((-BigDecimal::INFINITY) ** BigDecimal(2)) assert_negative_infinite((-BigDecimal::INFINITY) ** 1) assert_negative_infinite((-BigDecimal::INFINITY) ** 1.quo(1)) assert_negative_infinite((-BigDecimal::INFINITY) ** 1.0) assert_negative_infinite((-BigDecimal::INFINITY) ** BigDecimal(1)) assert_equal(1, (-BigDecimal::INFINITY) ** 0) assert_equal(1, (-BigDecimal::INFINITY) ** 0.quo(1)) assert_equal(1, (-BigDecimal::INFINITY) ** 0.0) assert_equal(1, (-BigDecimal::INFINITY) ** BigDecimal(0)) assert_negative_zero((-BigDecimal::INFINITY) ** -1) assert_negative_zero((-BigDecimal::INFINITY) ** -1.quo(1)) assert_negative_zero((-BigDecimal::INFINITY) ** -1.0) assert_negative_zero((-BigDecimal::INFINITY) ** BigDecimal(-1)) assert_positive_zero((-BigDecimal::INFINITY) ** -2) assert_positive_zero((-BigDecimal::INFINITY) ** -2.quo(1)) assert_positive_zero((-BigDecimal::INFINITY) ** -2.0) assert_positive_zero((-BigDecimal::INFINITY) ** BigDecimal(-2)) end end def test_power_without_prec pi = BigDecimal("3.14159265358979323846264338327950288419716939937511") e = BigDecimal("2.71828182845904523536028747135266249775724709369996") pow = BigDecimal("22.4591577183610454734271522045437350275893151339967843873233068") assert_equal(pow, pi.power(e)) end def test_power_with_prec pi = BigDecimal("3.14159265358979323846264338327950288419716939937511") e = BigDecimal("2.71828182845904523536028747135266249775724709369996") pow = BigDecimal("22.459157718361045473") assert_equal(pow, pi.power(e, 20)) b = BigDecimal('1.034482758620689655172413793103448275862068965517241379310344827586206896551724') assert_equal(BigDecimal('0.114523E1'), b.power(4, 5), '[Bug #8818] [ruby-core:56802]') end def test_limit BigDecimal.limit(1) x = BigDecimal.new("3") assert_equal(90, x ** 4) # OK? must it be 80? # 3 * 3 * 3 * 3 = 10 * 3 * 3 = 30 * 3 = 90 ??? assert_raise(ArgumentError) { BigDecimal.limit(-1) } bug7458 = '[ruby-core:50269] [#7458]' one = BigDecimal('1') epsilon = BigDecimal('0.7E-18') BigDecimal.save_limit do BigDecimal.limit(0) assert_equal(BigDecimal("1.0000000000000000007"), one + epsilon, "limit(0) #{bug7458}") 1.upto(18) do |lim| BigDecimal.limit(lim) assert_equal(BigDecimal("1.0"), one + epsilon, "limit(#{lim}) #{bug7458}") end BigDecimal.limit(19) assert_equal(BigDecimal("1.000000000000000001"), one + epsilon, "limit(19) #{bug7458}") BigDecimal.limit(20) assert_equal(BigDecimal("1.0000000000000000007"), one + epsilon, "limit(20) #{bug7458}") end end def test_sign BigDecimal.mode(BigDecimal::EXCEPTION_OVERFLOW, false) BigDecimal.mode(BigDecimal::EXCEPTION_NaN, false) BigDecimal.mode(BigDecimal::EXCEPTION_ZERODIVIDE, false) assert_equal(BigDecimal::SIGN_POSITIVE_ZERO, BigDecimal.new("0").sign) assert_equal(BigDecimal::SIGN_NEGATIVE_ZERO, BigDecimal.new("-0").sign) assert_equal(BigDecimal::SIGN_POSITIVE_FINITE, BigDecimal.new("1").sign) assert_equal(BigDecimal::SIGN_NEGATIVE_FINITE, BigDecimal.new("-1").sign) assert_equal(BigDecimal::SIGN_POSITIVE_INFINITE, (BigDecimal.new("1") / 0).sign) assert_equal(BigDecimal::SIGN_NEGATIVE_INFINITE, (BigDecimal.new("-1") / 0).sign) assert_equal(BigDecimal::SIGN_NaN, (BigDecimal.new("0") / 0).sign) end def test_inf BigDecimal.mode(BigDecimal::EXCEPTION_OVERFLOW, false) BigDecimal.mode(BigDecimal::EXCEPTION_NaN, false) inf = BigDecimal.new("Infinity") assert_equal(inf, inf + inf) assert_equal(true, (inf + (-inf)).nan?) assert_equal(true, (inf - inf).nan?) assert_equal(inf, inf - (-inf)) assert_equal(inf, inf * inf) assert_equal(true, (inf / inf).nan?) assert_equal(inf, inf + 1) assert_equal(inf, inf - 1) assert_equal(inf, inf * 1) assert_equal(true, (inf * 0).nan?) assert_equal(inf, inf / 1) assert_equal(inf, 1 + inf) assert_equal(-inf, 1 - inf) assert_equal(inf, 1 * inf) assert_equal(-inf, -1 * inf) assert_equal(true, (0 * inf).nan?) assert_equal(BigDecimal::SIGN_POSITIVE_ZERO, (1 / inf).sign) assert_equal(BigDecimal::SIGN_NEGATIVE_ZERO, (-1 / inf).sign) end def test_to_special_string BigDecimal.mode(BigDecimal::EXCEPTION_OVERFLOW, false) BigDecimal.mode(BigDecimal::EXCEPTION_NaN, false) nan = BigDecimal.new("NaN") assert_equal("NaN", nan.to_s) inf = BigDecimal.new("Infinity") assert_equal("Infinity", inf.to_s) assert_equal(" Infinity", inf.to_s(" ")) assert_equal("+Infinity", inf.to_s("+")) assert_equal("-Infinity", (-inf).to_s) pzero = BigDecimal.new("0") assert_equal("0.0", pzero.to_s) assert_equal(" 0.0", pzero.to_s(" ")) assert_equal("+0.0", pzero.to_s("+")) assert_equal("-0.0", (-pzero).to_s) end def test_to_string assert_equal("0.01", BigDecimal("0.01").to_s("F")) s = "0." + "0" * 100 + "1" assert_equal(s, BigDecimal(s).to_s("F")) s = "1" + "0" * 100 + ".0" assert_equal(s, BigDecimal(s).to_s("F")) end def test_ctov assert_equal(0.1, BigDecimal.new("1E-1")) assert_equal(10, BigDecimal.new("1E+1")) assert_equal(1, BigDecimal.new("+1")) BigDecimal.mode(BigDecimal::EXCEPTION_OVERFLOW, false) assert_equal(BigDecimal::SIGN_POSITIVE_INFINITE, BigDecimal.new("1E1" + "0" * 10000).sign) assert_equal(BigDecimal::SIGN_NEGATIVE_INFINITE, BigDecimal.new("-1E1" + "0" * 10000).sign) assert_equal(BigDecimal::SIGN_POSITIVE_ZERO, BigDecimal.new("1E-1" + "0" * 10000).sign) assert_equal(BigDecimal::SIGN_NEGATIVE_ZERO, BigDecimal.new("-1E-1" + "0" * 10000).sign) end def test_split_under_gc_stress bug3258 = '[ruby-dev:41213]' expect = 10.upto(20).map{|i|[1, "1", 10, i+1].inspect} assert_in_out_err(%w[-rbigdecimal --disable-gems], <<-EOS, expect, [], bug3258) GC.stress = true 10.upto(20) do |i| p BigDecimal.new("1"+"0"*i).split end EOS end def test_coerce_under_gc_stress assert_in_out_err(%w[-rbigdecimal --disable-gems], <<-EOS, [], []) expect = ":too_long_to_embed_as_string can't be coerced into BigDecimal" b = BigDecimal.new("1") GC.stress = true 10.times do begin b.coerce(:too_long_to_embed_as_string) rescue => e raise unless e.is_a?(TypeError) raise "'\#{expect}' is expected, but '\#{e.message}'" unless e.message == expect end end EOS end def test_INFINITY assert(BigDecimal::INFINITY.infinite?, "BigDecimal::INFINITY is not a infinity") end def test_NAN assert(BigDecimal::NAN.nan?, "BigDecimal::NAN is not NaN") end def test_exp_with_zerp_precision assert_raise(ArgumentError) do BigMath.exp(1, 0) end end def test_exp_with_negative_precision assert_raise(ArgumentError) do BigMath.exp(1, -42) end end def test_exp_with_complex assert_raise(ArgumentError) do BigMath.exp(Complex(1, 2), 20) end end def test_exp_with_negative_infinite BigDecimal.save_exception_mode do BigDecimal.mode(BigDecimal::EXCEPTION_INFINITY, false) assert_equal(0, BigMath.exp(-BigDecimal::INFINITY, 20)) end end def test_exp_with_positive_infinite BigDecimal.save_exception_mode do BigDecimal.mode(BigDecimal::EXCEPTION_INFINITY, false) assert(BigMath.exp(BigDecimal::INFINITY, 20) > 0) assert(BigMath.exp(BigDecimal::INFINITY, 20).infinite?) end end def test_exp_with_nan BigDecimal.save_exception_mode do BigDecimal.mode(BigDecimal::EXCEPTION_NaN, false) assert(BigMath.exp(BigDecimal::NAN, 20).nan?) end end def test_exp_with_1 assert_in_epsilon(Math::E, BigMath.exp(1, 20)) end def test_BigMath_exp prec = 20 assert_in_epsilon(Math.exp(20), BigMath.exp(BigDecimal("20"), prec)) assert_in_epsilon(Math.exp(40), BigMath.exp(BigDecimal("40"), prec)) assert_in_epsilon(Math.exp(-20), BigMath.exp(BigDecimal("-20"), prec)) assert_in_epsilon(Math.exp(-40), BigMath.exp(BigDecimal("-40"), prec)) end def test_BigMath_exp_with_float prec = 20 assert_in_epsilon(Math.exp(20), BigMath.exp(20.0, prec)) assert_in_epsilon(Math.exp(40), BigMath.exp(40.0, prec)) assert_in_epsilon(Math.exp(-20), BigMath.exp(-20.0, prec)) assert_in_epsilon(Math.exp(-40), BigMath.exp(-40.0, prec)) end def test_BigMath_exp_with_fixnum prec = 20 assert_in_epsilon(Math.exp(20), BigMath.exp(20, prec)) assert_in_epsilon(Math.exp(40), BigMath.exp(40, prec)) assert_in_epsilon(Math.exp(-20), BigMath.exp(-20, prec)) assert_in_epsilon(Math.exp(-40), BigMath.exp(-40, prec)) end def test_BigMath_exp_with_rational prec = 20 assert_in_epsilon(Math.exp(20), BigMath.exp(Rational(40,2), prec)) assert_in_epsilon(Math.exp(40), BigMath.exp(Rational(80,2), prec)) assert_in_epsilon(Math.exp(-20), BigMath.exp(Rational(-40,2), prec)) assert_in_epsilon(Math.exp(-40), BigMath.exp(Rational(-80,2), prec)) end def test_BigMath_exp_under_gc_stress assert_in_out_err(%w[-rbigdecimal --disable-gems], <<-EOS, [], []) expect = ":too_long_to_embed_as_string can't be coerced into BigDecimal" 10.times do begin BigMath.exp(:too_long_to_embed_as_string, 6) rescue => e raise unless e.is_a?(ArgumentError) raise "'\#{expect}' is expected, but '\#{e.message}'" unless e.message == expect end end EOS end def test_BigMath_log_with_string assert_raise(ArgumentError) do BigMath.log("foo", 20) end end def test_BigMath_log_with_nil assert_raise(ArgumentError) do BigMath.log(nil, 20) end end def test_BigMath_log_with_non_integer_precision assert_raise(ArgumentError) do BigMath.log(1, 0.5) end end def test_BigMath_log_with_nil_precision assert_raise(ArgumentError) do BigMath.log(1, nil) end end def test_BigMath_log_with_complex assert_raise(Math::DomainError) do BigMath.log(Complex(1, 2), 20) end end def test_BigMath_log_with_zero_arg assert_raise(Math::DomainError) do BigMath.log(0, 20) end end def test_BigMath_log_with_negative_arg assert_raise(Math::DomainError) do BigMath.log(-1, 20) end end def test_BigMath_log_with_zero_precision assert_raise(ArgumentError) do BigMath.log(1, 0) end end def test_BigMath_log_with_negative_precision assert_raise(ArgumentError) do BigMath.log(1, -42) end end def test_BigMath_log_with_negative_infinite BigDecimal.save_exception_mode do BigDecimal.mode(BigDecimal::EXCEPTION_INFINITY, false) assert_raise(Math::DomainError) do BigMath.log(-BigDecimal::INFINITY, 20) end end end def test_BigMath_log_with_positive_infinite BigDecimal.save_exception_mode do BigDecimal.mode(BigDecimal::EXCEPTION_INFINITY, false) assert(BigMath.log(BigDecimal::INFINITY, 20) > 0) assert(BigMath.log(BigDecimal::INFINITY, 20).infinite?) end end def test_BigMath_log_with_nan BigDecimal.save_exception_mode do BigDecimal.mode(BigDecimal::EXCEPTION_NaN, false) assert(BigMath.log(BigDecimal::NAN, 20).nan?) end end def test_BigMath_log_with_float_nan BigDecimal.save_exception_mode do BigDecimal.mode(BigDecimal::EXCEPTION_NaN, false) assert(BigMath.log(Float::NAN, 20).nan?) end end def test_BigMath_log_with_1 assert_in_delta(0.0, BigMath.log(1, 20)) assert_in_delta(0.0, BigMath.log(1.0, 20)) assert_in_delta(0.0, BigMath.log(BigDecimal(1), 20)) end def test_BigMath_log_with_exp_1 assert_in_delta(1.0, BigMath.log(BigMath.E(10), 10)) end def test_BigMath_log_with_2 assert_in_delta(Math.log(2), BigMath.log(2, 20)) assert_in_delta(Math.log(2), BigMath.log(2.0, 20)) assert_in_delta(Math.log(2), BigMath.log(BigDecimal(2), 20)) end def test_BigMath_log_with_square_of_E assert_in_delta(2, BigMath.log(BigMath.E(20)**2, 20)) end def test_BigMath_log_with_high_precision_case e = BigDecimal('2.71828182845904523536028747135266249775724709369996') e_3 = e.mult(e, 50).mult(e, 50) log_3 = BigMath.log(e_3, 50) assert_in_delta(3, log_3, 0.0000000000_0000000000_0000000000_0000000000_0000000001) end def test_BigMath_log_with_42 assert_in_delta(Math.log(42), BigMath.log(42, 20)) assert_in_delta(Math.log(42), BigMath.log(42.0, 20)) assert_in_delta(Math.log(42), BigMath.log(BigDecimal(42), 20)) end def test_BigMath_log_with_101 # this is mainly a performance test (should be very fast, not the 0.3 s) assert_in_delta(Math.log(101), BigMath.log(101, 20), 1E-15) end def test_BigMath_log_with_reciprocal_of_42 assert_in_delta(Math.log(1e-42), BigMath.log(1e-42, 20)) assert_in_delta(Math.log(1e-42), BigMath.log(BigDecimal("1e-42"), 20)) end def test_BigMath_log_under_gc_stress assert_in_out_err(%w[-rbigdecimal --disable-gems], <<-EOS, [], []) expect = ":too_long_to_embed_as_string can't be coerced into BigDecimal" 10.times do begin BigMath.log(:too_long_to_embed_as_string, 6) rescue => e raise unless e.is_a?(ArgumentError) raise "'\#{expect}' is expected, but '\#{e.message}'" unless e.message == expect end end EOS end def test_dup [1, -1, 2**100, -2**100].each do |i| x = BigDecimal(i) assert_equal(x, x.dup) end end def test_dup_subclass c = Class.new(BigDecimal) x = c.new(1) y = x.dup assert_equal(1, y) assert_kind_of(c, y) end def test_to_d bug6093 = '[ruby-core:42969]' code = "exit(BigDecimal.new('10.0') == 10.0.to_d)" assert_ruby_status(%w[-rbigdecimal -rbigdecimal/util -rmathn -], code, bug6093) end def test_bug6406 assert_in_out_err(%w[-rbigdecimal --disable-gems], <<-EOS, [], []) Thread.current.keys.to_s EOS end end