ladybird/Userland/Libraries/LibJS/Runtime/Date.cpp
Andreas Kling 72c9f56c66 LibJS: Make Heap::allocate<T>() infallible
Stop worrying about tiny OOMs. Work towards #20449.

While going through these, I also changed the function signature in many
places where returning ThrowCompletionOr<T> is no longer necessary.
2023-08-13 15:38:42 +02:00

633 lines
26 KiB
C++
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

/*
* Copyright (c) 2020-2023, Linus Groh <linusg@serenityos.org>
* Copyright (c) 2022, Tim Flynn <trflynn89@serenityos.org>
*
* SPDX-License-Identifier: BSD-2-Clause
*/
#include <AK/NumericLimits.h>
#include <AK/StringBuilder.h>
#include <AK/Time.h>
#include <LibJS/Runtime/AbstractOperations.h>
#include <LibJS/Runtime/Date.h>
#include <LibJS/Runtime/GlobalObject.h>
#include <LibJS/Runtime/Temporal/ISO8601.h>
#include <LibTimeZone/TimeZone.h>
#include <time.h>
namespace JS {
static Crypto::SignedBigInteger const s_one_billion_bigint { 1'000'000'000 };
static Crypto::SignedBigInteger const s_one_million_bigint { 1'000'000 };
static Crypto::SignedBigInteger const s_one_thousand_bigint { 1'000 };
Crypto::SignedBigInteger const ns_per_day_bigint { static_cast<i64>(ns_per_day) };
NonnullGCPtr<Date> Date::create(Realm& realm, double date_value)
{
return realm.heap().allocate<Date>(realm, date_value, realm.intrinsics().date_prototype());
}
Date::Date(double date_value, Object& prototype)
: Object(ConstructWithPrototypeTag::Tag, prototype)
, m_date_value(date_value)
{
}
ErrorOr<String> Date::iso_date_string() const
{
int year = year_from_time(m_date_value);
StringBuilder builder;
if (year < 0)
builder.appendff("-{:06}", -year);
else if (year > 9999)
builder.appendff("+{:06}", year);
else
builder.appendff("{:04}", year);
builder.append('-');
builder.appendff("{:02}", month_from_time(m_date_value) + 1);
builder.append('-');
builder.appendff("{:02}", date_from_time(m_date_value));
builder.append('T');
builder.appendff("{:02}", hour_from_time(m_date_value));
builder.append(':');
builder.appendff("{:02}", min_from_time(m_date_value));
builder.append(':');
builder.appendff("{:02}", sec_from_time(m_date_value));
builder.append('.');
builder.appendff("{:03}", ms_from_time(m_date_value));
builder.append('Z');
return builder.to_string();
}
// DayWithinYear(t), https://tc39.es/ecma262/#eqn-DayWithinYear
u16 day_within_year(double t)
{
if (!Value(t).is_finite_number())
return 0;
// Day(t) - DayFromYear(YearFromTime(t))
return static_cast<u16>(day(t) - day_from_year(year_from_time(t)));
}
// DateFromTime(t), https://tc39.es/ecma262/#sec-date-number
u8 date_from_time(double t)
{
switch (month_from_time(t)) {
// DayWithinYear(t) + 1𝔽 if MonthFromTime(t) = +0𝔽
case 0:
return day_within_year(t) + 1;
// DayWithinYear(t) - 30𝔽 if MonthFromTime(t) = 1𝔽
case 1:
return day_within_year(t) - 30;
// DayWithinYear(t) - 58𝔽 - InLeapYear(t) if MonthFromTime(t) = 2𝔽
case 2:
return day_within_year(t) - 58 - in_leap_year(t);
// DayWithinYear(t) - 89𝔽 - InLeapYear(t) if MonthFromTime(t) = 3𝔽
case 3:
return day_within_year(t) - 89 - in_leap_year(t);
// DayWithinYear(t) - 119𝔽 - InLeapYear(t) if MonthFromTime(t) = 4𝔽
case 4:
return day_within_year(t) - 119 - in_leap_year(t);
// DayWithinYear(t) - 150𝔽 - InLeapYear(t) if MonthFromTime(t) = 5𝔽
case 5:
return day_within_year(t) - 150 - in_leap_year(t);
// DayWithinYear(t) - 180𝔽 - InLeapYear(t) if MonthFromTime(t) = 6𝔽
case 6:
return day_within_year(t) - 180 - in_leap_year(t);
// DayWithinYear(t) - 211𝔽 - InLeapYear(t) if MonthFromTime(t) = 7𝔽
case 7:
return day_within_year(t) - 211 - in_leap_year(t);
// DayWithinYear(t) - 242𝔽 - InLeapYear(t) if MonthFromTime(t) = 8𝔽
case 8:
return day_within_year(t) - 242 - in_leap_year(t);
// DayWithinYear(t) - 272𝔽 - InLeapYear(t) if MonthFromTime(t) = 9𝔽
case 9:
return day_within_year(t) - 272 - in_leap_year(t);
// DayWithinYear(t) - 303𝔽 - InLeapYear(t) if MonthFromTime(t) = 10𝔽
case 10:
return day_within_year(t) - 303 - in_leap_year(t);
// DayWithinYear(t) - 333𝔽 - InLeapYear(t) if MonthFromTime(t) = 11𝔽
case 11:
return day_within_year(t) - 333 - in_leap_year(t);
default:
VERIFY_NOT_REACHED();
}
}
// DaysInYear(y), https://tc39.es/ecma262/#eqn-DaysInYear
u16 days_in_year(i32 y)
{
// 365𝔽 if ((y) modulo 4) ≠ 0
if (y % 4 != 0)
return 365;
// 366𝔽 if ((y) modulo 4) = 0 and ((y) modulo 100) ≠ 0
if (y % 4 == 0 && y % 100 != 0)
return 366;
// 365𝔽 if ((y) modulo 100) = 0 and ((y) modulo 400) ≠ 0
if (y % 100 == 0 && y % 400 != 0)
return 365;
// 366𝔽 if ((y) modulo 400) = 0
if (y % 400 == 0)
return 366;
VERIFY_NOT_REACHED();
}
// DayFromYear(y), https://tc39.es/ecma262/#eqn-DaysFromYear
double day_from_year(i32 y)
{
// 𝔽(365 × ((y) - 1970) + floor(((y) - 1969) / 4) - floor(((y) - 1901) / 100) + floor(((y) - 1601) / 400))
return 365.0 * (y - 1970) + floor((y - 1969) / 4.0) - floor((y - 1901) / 100.0) + floor((y - 1601) / 400.0);
}
// TimeFromYear(y), https://tc39.es/ecma262/#eqn-TimeFromYear
double time_from_year(i32 y)
{
// msPerDay × DayFromYear(y)
return ms_per_day * day_from_year(y);
}
// YearFromTime(t), https://tc39.es/ecma262/#eqn-YearFromTime
i32 year_from_time(double t)
{
// the largest integral Number y (closest to +∞) such that TimeFromYear(y) ≤ t
if (!Value(t).is_finite_number())
return NumericLimits<i32>::max();
// Approximation using average number of milliseconds per year. We might have to adjust this guess afterwards.
auto year = static_cast<i32>(t / (365.2425 * ms_per_day) + 1970);
auto year_t = time_from_year(year);
if (year_t > t)
year--;
else if (year_t + days_in_year(year) * ms_per_day <= t)
year++;
return year;
}
// InLeapYear(t), https://tc39.es/ecma262/#eqn-InLeapYear
bool in_leap_year(double t)
{
// +0𝔽 if DaysInYear(YearFromTime(t)) = 365𝔽
// 1𝔽 if DaysInYear(YearFromTime(t)) = 366𝔽
return days_in_year(year_from_time(t)) == 366;
}
// MonthFromTime(t), https://tc39.es/ecma262/#eqn-MonthFromTime
u8 month_from_time(double t)
{
auto in_leap_year = JS::in_leap_year(t);
auto day_within_year = JS::day_within_year(t);
// +0𝔽 if +0𝔽 ≤ DayWithinYear(t) < 31𝔽
if (day_within_year < 31)
return 0;
// 1𝔽 if 31𝔽 ≤ DayWithinYear(t) < 59𝔽 + InLeapYear(t)
if (31 <= day_within_year && day_within_year < 59 + in_leap_year)
return 1;
// 2𝔽 if 59𝔽 + InLeapYear(t) ≤ DayWithinYear(t) < 90𝔽 + InLeapYear(t)
if (59 + in_leap_year <= day_within_year && day_within_year < 90 + in_leap_year)
return 2;
// 3𝔽 if 90𝔽 + InLeapYear(t) ≤ DayWithinYear(t) < 120𝔽 + InLeapYear(t)
if (90 + in_leap_year <= day_within_year && day_within_year < 120 + in_leap_year)
return 3;
// 4𝔽 if 120𝔽 + InLeapYear(t) ≤ DayWithinYear(t) < 151𝔽 + InLeapYear(t)
if (120 + in_leap_year <= day_within_year && day_within_year < 151 + in_leap_year)
return 4;
// 5𝔽 if 151𝔽 + InLeapYear(t) ≤ DayWithinYear(t) < 181𝔽 + InLeapYear(t)
if (151 + in_leap_year <= day_within_year && day_within_year < 181 + in_leap_year)
return 5;
// 6𝔽 if 181𝔽 + InLeapYear(t) ≤ DayWithinYear(t) < 212𝔽 + InLeapYear(t)
if (181 + in_leap_year <= day_within_year && day_within_year < 212 + in_leap_year)
return 6;
// 7𝔽 if 212𝔽 + InLeapYear(t) ≤ DayWithinYear(t) < 243𝔽 + InLeapYear(t)
if (212 + in_leap_year <= day_within_year && day_within_year < 243 + in_leap_year)
return 7;
// 8𝔽 if 243𝔽 + InLeapYear(t) ≤ DayWithinYear(t) < 273𝔽 + InLeapYear(t)
if (243 + in_leap_year <= day_within_year && day_within_year < 273 + in_leap_year)
return 8;
// 9𝔽 if 273𝔽 + InLeapYear(t) ≤ DayWithinYear(t) < 304𝔽 + InLeapYear(t)
if (273 + in_leap_year <= day_within_year && day_within_year < 304 + in_leap_year)
return 9;
// 10𝔽 if 304𝔽 + InLeapYear(t) ≤ DayWithinYear(t) < 334𝔽 + InLeapYear(t)
if (304 + in_leap_year <= day_within_year && day_within_year < 334 + in_leap_year)
return 10;
// 11𝔽 if 334𝔽 + InLeapYear(t) ≤ DayWithinYear(t) < 365𝔽 + InLeapYear(t)
if (334 + in_leap_year <= day_within_year && day_within_year < 365 + in_leap_year)
return 11;
VERIFY_NOT_REACHED();
}
// HourFromTime(t), https://tc39.es/ecma262/#eqn-HourFromTime
u8 hour_from_time(double t)
{
if (!Value(t).is_finite_number())
return 0;
// 𝔽(floor((t / msPerHour)) modulo HoursPerDay)
return static_cast<u8>(modulo(floor(t / ms_per_hour), hours_per_day));
}
// MinFromTime(t), https://tc39.es/ecma262/#eqn-MinFromTime
u8 min_from_time(double t)
{
if (!Value(t).is_finite_number())
return 0;
// 𝔽(floor((t / msPerMinute)) modulo MinutesPerHour)
return static_cast<u8>(modulo(floor(t / ms_per_minute), minutes_per_hour));
}
// SecFromTime(t), https://tc39.es/ecma262/#eqn-SecFromTime
u8 sec_from_time(double t)
{
if (!Value(t).is_finite_number())
return 0;
// 𝔽(floor((t / msPerSecond)) modulo SecondsPerMinute)
return static_cast<u8>(modulo(floor(t / ms_per_second), seconds_per_minute));
}
// msFromTime(t), https://tc39.es/ecma262/#eqn-msFromTime
u16 ms_from_time(double t)
{
if (!Value(t).is_finite_number())
return 0;
// 𝔽((t) modulo (msPerSecond))
return static_cast<u16>(modulo(t, ms_per_second));
}
// 21.4.1.6 Week Day, https://tc39.es/ecma262/#sec-week-day
u8 week_day(double t)
{
if (!Value(t).is_finite_number())
return 0;
// 𝔽((Day(t) + 4𝔽) modulo 7)
return static_cast<u8>(modulo(day(t) + 4, 7));
}
// 21.4.1.7 GetUTCEpochNanoseconds ( year, month, day, hour, minute, second, millisecond, microsecond, nanosecond ), https://tc39.es/ecma262/#sec-getutcepochnanoseconds
Crypto::SignedBigInteger get_utc_epoch_nanoseconds(i32 year, u8 month, u8 day, u8 hour, u8 minute, u8 second, u16 millisecond, u16 microsecond, u16 nanosecond)
{
// 1. Let date be MakeDay(𝔽(year), 𝔽(month - 1), 𝔽(day)).
auto date = make_day(year, month - 1, day);
// 2. Let time be MakeTime(𝔽(hour), 𝔽(minute), 𝔽(second), 𝔽(millisecond)).
auto time = make_time(hour, minute, second, millisecond);
// 3. Let ms be MakeDate(date, time).
auto ms = make_date(date, time);
// 4. Assert: ms is an integral Number.
VERIFY(ms == trunc(ms));
// 5. Return ((ms) × 10^6 + microsecond × 10^3 + nanosecond).
auto result = Crypto::SignedBigInteger { ms }.multiplied_by(s_one_million_bigint);
result = result.plus(Crypto::SignedBigInteger { static_cast<i32>(microsecond) }.multiplied_by(s_one_thousand_bigint));
result = result.plus(Crypto::SignedBigInteger { static_cast<i32>(nanosecond) });
return result;
}
static i64 clip_bigint_to_sane_time(Crypto::SignedBigInteger const& value)
{
static Crypto::SignedBigInteger const min_bigint { NumericLimits<i64>::min() };
static Crypto::SignedBigInteger const max_bigint { NumericLimits<i64>::max() };
// The provided epoch (nano)seconds value is potentially out of range for AK::Duration and subsequently
// get_time_zone_offset(). We can safely assume that the TZDB has no useful information that far
// into the past and future anyway, so clamp it to the i64 range.
if (value < min_bigint)
return NumericLimits<i64>::min();
if (value > max_bigint)
return NumericLimits<i64>::max();
// FIXME: Can we do this without string conversion?
return value.to_base_deprecated(10).to_int<i64>().value();
}
// 21.4.1.8 GetNamedTimeZoneEpochNanoseconds ( timeZoneIdentifier, year, month, day, hour, minute, second, millisecond, microsecond, nanosecond ), https://tc39.es/ecma262/#sec-getnamedtimezoneepochnanoseconds
Vector<Crypto::SignedBigInteger> get_named_time_zone_epoch_nanoseconds(StringView time_zone_identifier, i32 year, u8 month, u8 day, u8 hour, u8 minute, u8 second, u16 millisecond, u16 microsecond, u16 nanosecond)
{
auto local_nanoseconds = get_utc_epoch_nanoseconds(year, month, day, hour, minute, second, millisecond, microsecond, nanosecond);
auto local_time = UnixDateTime::from_nanoseconds_since_epoch(clip_bigint_to_sane_time(local_nanoseconds));
// FIXME: LibTimeZone does not behave exactly as the spec expects. It does not consider repeated or skipped time points.
auto offset = TimeZone::get_time_zone_offset(time_zone_identifier, local_time);
// Can only fail if the time zone identifier is invalid, which cannot be the case here.
VERIFY(offset.has_value());
return { local_nanoseconds.minus(Crypto::SignedBigInteger { offset->seconds }.multiplied_by(s_one_billion_bigint)) };
}
// 21.4.1.9 GetNamedTimeZoneOffsetNanoseconds ( timeZoneIdentifier, epochNanoseconds ), https://tc39.es/ecma262/#sec-getnamedtimezoneoffsetnanoseconds
i64 get_named_time_zone_offset_nanoseconds(StringView time_zone_identifier, Crypto::SignedBigInteger const& epoch_nanoseconds)
{
// Only called with validated time zone identifier as argument.
auto time_zone = TimeZone::time_zone_from_string(time_zone_identifier);
VERIFY(time_zone.has_value());
// Since UnixDateTime::from_seconds_since_epoch() and UnixDateTime::from_nanoseconds_since_epoch() both take an i64, converting to
// seconds first gives us a greater range. The TZDB doesn't have sub-second offsets.
auto seconds = epoch_nanoseconds.divided_by(s_one_billion_bigint).quotient;
auto time = UnixDateTime::from_seconds_since_epoch(clip_bigint_to_sane_time(seconds));
auto offset = TimeZone::get_time_zone_offset(*time_zone, time);
VERIFY(offset.has_value());
return offset->seconds * 1'000'000'000;
}
// 21.4.1.10 DefaultTimeZone ( ), https://tc39.es/ecma262/#sec-defaulttimezone
// 6.4.3 DefaultTimeZone ( ), https://tc39.es/ecma402/#sup-defaulttimezone
StringView default_time_zone()
{
return TimeZone::current_time_zone();
}
// 21.4.1.11 LocalTime ( t ), https://tc39.es/ecma262/#sec-localtime
double local_time(double time)
{
// 1. Let localTimeZone be DefaultTimeZone().
auto local_time_zone = default_time_zone();
double offset_nanoseconds { 0 };
// 2. If IsTimeZoneOffsetString(localTimeZone) is true, then
if (is_time_zone_offset_string(local_time_zone)) {
// a. Let offsetNs be ParseTimeZoneOffsetString(localTimeZone).
offset_nanoseconds = parse_time_zone_offset_string(local_time_zone);
}
// 3. Else,
else {
// a. Let offsetNs be GetNamedTimeZoneOffsetNanoseconds(localTimeZone, ((t) × 10^6)).
auto time_bigint = Crypto::SignedBigInteger { time }.multiplied_by(s_one_million_bigint);
offset_nanoseconds = get_named_time_zone_offset_nanoseconds(local_time_zone, time_bigint);
}
// 4. Let offsetMs be truncate(offsetNs / 10^6).
auto offset_milliseconds = trunc(offset_nanoseconds / 1e6);
// 5. Return t + 𝔽(offsetMs).
return time + offset_milliseconds;
}
// 21.4.1.12 UTC ( t ), https://tc39.es/ecma262/#sec-utc-t
double utc_time(double time)
{
// 1. Let localTimeZone be DefaultTimeZone().
auto local_time_zone = default_time_zone();
double offset_nanoseconds { 0 };
// 2. If IsTimeZoneOffsetString(localTimeZone) is true, then
if (is_time_zone_offset_string(local_time_zone)) {
// a. Let offsetNs be ParseTimeZoneOffsetString(localTimeZone).
offset_nanoseconds = parse_time_zone_offset_string(local_time_zone);
}
// 3. Else,
else {
// a. Let possibleInstants be GetNamedTimeZoneEpochNanoseconds(localTimeZone, (YearFromTime(t)), (MonthFromTime(t)) + 1, (DateFromTime(t)), (HourFromTime(t)), (MinFromTime(t)), (SecFromTime(t)), (msFromTime(t)), 0, 0).
auto possible_instants = get_named_time_zone_epoch_nanoseconds(local_time_zone, year_from_time(time), month_from_time(time) + 1, date_from_time(time), hour_from_time(time), min_from_time(time), sec_from_time(time), ms_from_time(time), 0, 0);
// b. NOTE: The following steps ensure that when t represents local time repeating multiple times at a negative time zone transition (e.g. when the daylight saving time ends or the time zone offset is decreased due to a time zone rule change) or skipped local time at a positive time zone transition (e.g. when the daylight saving time starts or the time zone offset is increased due to a time zone rule change), t is interpreted using the time zone offset before the transition.
Crypto::SignedBigInteger disambiguated_instant;
// c. If possibleInstants is not empty, then
if (!possible_instants.is_empty()) {
// i. Let disambiguatedInstant be possibleInstants[0].
disambiguated_instant = move(possible_instants.first());
}
// d. Else,
else {
// i. NOTE: t represents a local time skipped at a positive time zone transition (e.g. due to daylight saving time starting or a time zone rule change increasing the UTC offset).
// ii. Let possibleInstantsBefore be GetNamedTimeZoneEpochNanoseconds(localTimeZone, (YearFromTime(tBefore)), (MonthFromTime(tBefore)) + 1, (DateFromTime(tBefore)), (HourFromTime(tBefore)), (MinFromTime(tBefore)), (SecFromTime(tBefore)), (msFromTime(tBefore)), 0, 0), where tBefore is the largest integral Number < t for which possibleInstantsBefore is not empty (i.e., tBefore represents the last local time before the transition).
// iii. Let disambiguatedInstant be the last element of possibleInstantsBefore.
// FIXME: This branch currently cannot be reached with our implementation, because LibTimeZone does not handle skipped time points.
// When GetNamedTimeZoneEpochNanoseconds is updated to use a LibTimeZone API which does handle them, implement these steps.
VERIFY_NOT_REACHED();
}
// e. Let offsetNs be GetNamedTimeZoneOffsetNanoseconds(localTimeZone, disambiguatedInstant).
offset_nanoseconds = get_named_time_zone_offset_nanoseconds(local_time_zone, disambiguated_instant);
}
// 4. Let offsetMs be truncate(offsetNs / 10^6).
auto offset_milliseconds = trunc(offset_nanoseconds / 1e6);
// 5. Return t - 𝔽(offsetMs).
return time - offset_milliseconds;
}
// 21.4.1.14 MakeTime ( hour, min, sec, ms ), https://tc39.es/ecma262/#sec-maketime
double make_time(double hour, double min, double sec, double ms)
{
// 1. If hour is not finite or min is not finite or sec is not finite or ms is not finite, return NaN.
if (!isfinite(hour) || !isfinite(min) || !isfinite(sec) || !isfinite(ms))
return NAN;
// 2. Let h be 𝔽(! ToIntegerOrInfinity(hour)).
auto h = to_integer_or_infinity(hour);
// 3. Let m be 𝔽(! ToIntegerOrInfinity(min)).
auto m = to_integer_or_infinity(min);
// 4. Let s be 𝔽(! ToIntegerOrInfinity(sec)).
auto s = to_integer_or_infinity(sec);
// 5. Let milli be 𝔽(! ToIntegerOrInfinity(ms)).
auto milli = to_integer_or_infinity(ms);
// 6. Let t be ((h * msPerHour + m * msPerMinute) + s * msPerSecond) + milli, performing the arithmetic according to IEEE 754-2019 rules (that is, as if using the ECMAScript operators * and +).
// NOTE: C++ arithmetic abides by IEEE 754 rules
auto t = ((h * ms_per_hour + m * ms_per_minute) + s * ms_per_second) + milli;
// 7. Return t.
return t;
}
// Day(t), https://tc39.es/ecma262/#eqn-Day
double day(double time_value)
{
return floor(time_value / ms_per_day);
}
// TimeWithinDay(t), https://tc39.es/ecma262/#eqn-TimeWithinDay
double time_within_day(double time)
{
// 𝔽((t) modulo (msPerDay))
return modulo(time, ms_per_day);
}
// 21.4.1.15 MakeDay ( year, month, date ), https://tc39.es/ecma262/#sec-makeday
double make_day(double year, double month, double date)
{
// 1. If year is not finite or month is not finite or date is not finite, return NaN.
if (!isfinite(year) || !isfinite(month) || !isfinite(date))
return NAN;
// 2. Let y be 𝔽(! ToIntegerOrInfinity(year)).
auto y = to_integer_or_infinity(year);
// 3. Let m be 𝔽(! ToIntegerOrInfinity(month)).
auto m = to_integer_or_infinity(month);
// 4. Let dt be 𝔽(! ToIntegerOrInfinity(date)).
auto dt = to_integer_or_infinity(date);
// 5. Let ym be y + 𝔽(floor((m) / 12)).
auto ym = y + floor(m / 12);
// 6. If ym is not finite, return NaN.
if (!isfinite(ym))
return NAN;
// 7. Let mn be 𝔽((m) modulo 12).
auto mn = modulo(m, 12);
// 8. Find a finite time value t such that YearFromTime(t) is ym and MonthFromTime(t) is mn and DateFromTime(t) is 1𝔽; but if this is not possible (because some argument is out of range), return NaN.
if (!AK::is_within_range<int>(ym) || !AK::is_within_range<int>(mn + 1))
return NAN;
auto t = days_since_epoch(static_cast<int>(ym), static_cast<int>(mn) + 1, 1) * ms_per_day;
// 9. Return Day(t) + dt - 1𝔽.
return day(static_cast<double>(t)) + dt - 1;
}
// 21.4.1.16 MakeDate ( day, time ), https://tc39.es/ecma262/#sec-makedate
double make_date(double day, double time)
{
// 1. If day is not finite or time is not finite, return NaN.
if (!isfinite(day) || !isfinite(time))
return NAN;
// 2. Let tv be day × msPerDay + time.
auto tv = day * ms_per_day + time;
// 3. If tv is not finite, return NaN.
if (!isfinite(tv))
return NAN;
// 4. Return tv.
return tv;
}
// 21.4.1.17 TimeClip ( time ), https://tc39.es/ecma262/#sec-timeclip
double time_clip(double time)
{
// 1. If time is not finite, return NaN.
if (!isfinite(time))
return NAN;
// 2. If abs((time)) > 8.64 × 10^15, return NaN.
if (fabs(time) > 8.64E15)
return NAN;
// 3. Return 𝔽(! ToIntegerOrInfinity(time)).
return to_integer_or_infinity(time);
}
// 21.4.1.19.1 IsTimeZoneOffsetString ( offsetString ), https://tc39.es/ecma262/#sec-istimezoneoffsetstring
bool is_time_zone_offset_string(StringView offset_string)
{
// 1. Let parseResult be ParseText(StringToCodePoints(offsetString), UTCOffset).
auto parse_result = Temporal::parse_iso8601(Temporal::Production::TimeZoneNumericUTCOffset, offset_string);
// 2. If parseResult is a List of errors, return false.
// 3. Return true.
return parse_result.has_value();
}
// 21.4.1.19.2 ParseTimeZoneOffsetString ( offsetString ), https://tc39.es/ecma262/#sec-parsetimezoneoffsetstring
double parse_time_zone_offset_string(StringView offset_string)
{
// 1. Let parseResult be ParseText(StringToCodePoints(offsetString), UTCOffset).
auto parse_result = Temporal::parse_iso8601(Temporal::Production::TimeZoneNumericUTCOffset, offset_string);
// 2. Assert: parseResult is not a List of errors.
VERIFY(parse_result.has_value());
// 3. Assert: parseResult contains a TemporalSign Parse Node.
VERIFY(parse_result->time_zone_utc_offset_sign.has_value());
// 4. Let parsedSign be the source text matched by the TemporalSign Parse Node contained within parseResult.
auto parsed_sign = *parse_result->time_zone_utc_offset_sign;
i8 sign { 0 };
// 5. If parsedSign is the single code point U+002D (HYPHEN-MINUS) or U+2212 (MINUS SIGN), then
if (parsed_sign.is_one_of("-"sv, "\xE2\x88\x92"sv)) {
// a. Let sign be -1.
sign = -1;
}
// 6. Else,
else {
// a. Let sign be 1.
sign = 1;
}
// 7. NOTE: Applications of StringToNumber below do not lose precision, since each of the parsed values is guaranteed to be a sufficiently short string of decimal digits.
// 8. Assert: parseResult contains an Hour Parse Node.
VERIFY(parse_result->time_zone_utc_offset_hour.has_value());
// 9. Let parsedHours be the source text matched by the Hour Parse Node contained within parseResult.
auto parsed_hours = *parse_result->time_zone_utc_offset_hour;
// 10. Let hours be (StringToNumber(CodePointsToString(parsedHours))).
auto hours = string_to_number(parsed_hours);
double minutes { 0 };
double seconds { 0 };
double nanoseconds { 0 };
// 11. If parseResult does not contain a MinuteSecond Parse Node, then
if (!parse_result->time_zone_utc_offset_minute.has_value()) {
// a. Let minutes be 0.
minutes = 0;
}
// 12. Else,
else {
// a. Let parsedMinutes be the source text matched by the first MinuteSecond Parse Node contained within parseResult.
auto parsed_minutes = *parse_result->time_zone_utc_offset_minute;
// b. Let minutes be (StringToNumber(CodePointsToString(parsedMinutes))).
minutes = string_to_number(parsed_minutes);
}
// 13. If parseResult does not contain two MinuteSecond Parse Nodes, then
if (!parse_result->time_zone_utc_offset_second.has_value()) {
// a. Let seconds be 0.
seconds = 0;
}
// 14. Else,
else {
// a. Let parsedSeconds be the source text matched by the second secondSecond Parse Node contained within parseResult.
auto parsed_seconds = *parse_result->time_zone_utc_offset_second;
// b. Let seconds be (StringToNumber(CodePointsToString(parsedSeconds))).
seconds = string_to_number(parsed_seconds);
}
// 15. If parseResult does not contain a TemporalDecimalFraction Parse Node, then
if (!parse_result->time_zone_utc_offset_fraction.has_value()) {
// a. Let nanoseconds be 0.
nanoseconds = 0;
}
// 16. Else,
else {
// a. Let parsedFraction be the source text matched by the TemporalDecimalFraction Parse Node contained within parseResult.
auto parsed_fraction = *parse_result->time_zone_utc_offset_fraction;
// b. Let fraction be the string-concatenation of CodePointsToString(parsedFraction) and "000000000".
auto fraction = DeprecatedString::formatted("{}000000000", parsed_fraction);
// c. Let nanosecondsString be the substring of fraction from 1 to 10.
auto nanoseconds_string = fraction.substring_view(1, 9);
// d. Let nanoseconds be (StringToNumber(nanosecondsString)).
nanoseconds = string_to_number(nanoseconds_string);
}
// 17. Return sign × (((hours × 60 + minutes) × 60 + seconds) × 10^9 + nanoseconds).
// NOTE: Using scientific notation (1e9) ensures the result of this expression is a double,
// which is important - otherwise it's all integers and the result overflows!
return sign * (((hours * 60 + minutes) * 60 + seconds) * 1e9 + nanoseconds);
}
}