mirror of
https://github.com/SerenityOS/serenity
synced 2024-11-05 17:46:52 +00:00
7ab37ee22c
A lot of places were relying on AK/Traits.h to give it strnlen, memcmp, memcpy and other related declarations. In the quest to remove inclusion of LibC headers from Kernel files, deal with all the fallout of this included-everywhere header including less things.
386 lines
13 KiB
C++
386 lines
13 KiB
C++
/*
|
|
* Copyright (c) 2018-2020, Andreas Kling <kling@serenityos.org>
|
|
*
|
|
* SPDX-License-Identifier: BSD-2-Clause
|
|
*/
|
|
|
|
#pragma once
|
|
|
|
#include <AK/Array.h>
|
|
#include <AK/Assertions.h>
|
|
#include <AK/Platform.h>
|
|
#include <AK/Types.h>
|
|
|
|
#if defined(AK_OS_SERENITY) && defined(KERNEL)
|
|
# include <Kernel/API/POSIX/sys/time.h>
|
|
# include <Kernel/API/POSIX/time.h>
|
|
#else
|
|
# include <sys/time.h>
|
|
# include <time.h>
|
|
#endif
|
|
|
|
namespace AK {
|
|
|
|
// Concept to detect types which look like timespec without requiring the type.
|
|
template<typename T>
|
|
concept TimeSpecType = requires(T t) {
|
|
t.tv_sec;
|
|
t.tv_nsec;
|
|
};
|
|
|
|
constexpr bool is_leap_year(int year)
|
|
{
|
|
return year % 4 == 0 && (year % 100 != 0 || year % 400 == 0);
|
|
}
|
|
|
|
// Month and day start at 1. Month must be >= 1 and <= 12.
|
|
// The return value is 0-indexed, that is 0 is Sunday, 1 is Monday, etc.
|
|
// Day may be negative or larger than the number of days
|
|
// in the given month.
|
|
unsigned day_of_week(int year, unsigned month, int day);
|
|
|
|
// Month and day start at 1. Month must be >= 1 and <= 12.
|
|
// The return value is 0-indexed, that is Jan 1 is day 0.
|
|
// Day may be negative or larger than the number of days
|
|
// in the given month. If day is negative enough, the result
|
|
// can be negative.
|
|
constexpr int day_of_year(int year, unsigned month, int day)
|
|
{
|
|
VERIFY(month >= 1 && month <= 12);
|
|
|
|
constexpr Array seek_table = { 0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334 };
|
|
int day_of_year = seek_table[month - 1] + day - 1;
|
|
|
|
if (is_leap_year(year) && month >= 3)
|
|
day_of_year++;
|
|
|
|
return day_of_year;
|
|
}
|
|
|
|
// Month starts at 1. Month must be >= 1 and <= 12.
|
|
int days_in_month(int year, unsigned month);
|
|
|
|
constexpr int days_in_year(int year)
|
|
{
|
|
return 365 + (is_leap_year(year) ? 1 : 0);
|
|
}
|
|
|
|
namespace Detail {
|
|
// Integer division rounding towards negative infinity.
|
|
// TODO: This feels like there should be an easier way to do this.
|
|
template<int divisor>
|
|
constexpr i64 floor_div_by(i64 dividend)
|
|
{
|
|
static_assert(divisor > 1);
|
|
int is_negative = dividend < 0;
|
|
return (dividend + is_negative) / divisor - is_negative;
|
|
}
|
|
|
|
// Counts how many integers n are in the interval [begin, end) with n % positive_mod == 0.
|
|
// NOTE: "end" is not considered to be part of the range, hence "[begin, end)".
|
|
template<int positive_mod>
|
|
constexpr i64 mod_zeros_in_range(i64 begin, i64 end)
|
|
{
|
|
return floor_div_by<positive_mod>(end - 1) - floor_div_by<positive_mod>(begin - 1);
|
|
}
|
|
}
|
|
|
|
constexpr i64 years_to_days_since_epoch(int year)
|
|
{
|
|
int begin_year, end_year, leap_sign;
|
|
if (year < 1970) {
|
|
begin_year = year;
|
|
end_year = 1970;
|
|
leap_sign = -1;
|
|
} else {
|
|
begin_year = 1970;
|
|
end_year = year;
|
|
leap_sign = +1;
|
|
}
|
|
i64 year_i64 = year;
|
|
// This duplicates the logic of 'is_leap_year', with the advantage of not needing any loops.
|
|
// Given that the definition of leap years is not expected to change, this should be a good trade-off.
|
|
i64 days = 365 * (year_i64 - 1970);
|
|
i64 extra_leap_days = 0;
|
|
extra_leap_days += Detail::mod_zeros_in_range<4>(begin_year, end_year);
|
|
extra_leap_days -= Detail::mod_zeros_in_range<100>(begin_year, end_year);
|
|
extra_leap_days += Detail::mod_zeros_in_range<400>(begin_year, end_year);
|
|
return days + extra_leap_days * leap_sign;
|
|
}
|
|
|
|
constexpr i64 days_since_epoch(int year, int month, int day)
|
|
{
|
|
return years_to_days_since_epoch(year) + day_of_year(year, month, day);
|
|
}
|
|
|
|
constexpr i64 seconds_since_epoch_to_year(i64 seconds)
|
|
{
|
|
constexpr double seconds_per_year = 60.0 * 60.0 * 24.0 * 365.2425;
|
|
|
|
// NOTE: We are not using floor() from <math.h> to avoid LibC / DynamicLoader dependency issues.
|
|
auto round_down = [](double value) -> i64 {
|
|
auto as_i64 = static_cast<i64>(value);
|
|
|
|
if ((value == as_i64) || (as_i64 >= 0))
|
|
return as_i64;
|
|
return as_i64 - 1;
|
|
};
|
|
|
|
auto years_since_epoch = static_cast<double>(seconds) / seconds_per_year;
|
|
return 1970 + round_down(years_since_epoch);
|
|
}
|
|
|
|
/*
|
|
* Represents a time amount in a "safe" way.
|
|
* Minimum: 0 seconds, 0 nanoseconds
|
|
* Maximum: 2**63-1 seconds, 999'999'999 nanoseconds
|
|
* If any operation (e.g. 'from_timeval' or operator-) would over- or underflow, the closest legal value is returned instead.
|
|
* Inputs (e.g. to 'from_timespec') are allowed to be in non-normal form (e.g. "1 second, 2'012'345'678 nanoseconds" or "1 second, -2 microseconds").
|
|
* Outputs (e.g. from 'to_timeval') are always in normal form.
|
|
*/
|
|
class Time {
|
|
public:
|
|
Time() = default;
|
|
Time(Time const&) = default;
|
|
Time& operator=(Time const&) = default;
|
|
|
|
Time(Time&& other)
|
|
: m_seconds(exchange(other.m_seconds, 0))
|
|
, m_nanoseconds(exchange(other.m_nanoseconds, 0))
|
|
{
|
|
}
|
|
Time& operator=(Time&& other)
|
|
{
|
|
if (this != &other) {
|
|
m_seconds = exchange(other.m_seconds, 0);
|
|
m_nanoseconds = exchange(other.m_nanoseconds, 0);
|
|
}
|
|
return *this;
|
|
}
|
|
|
|
private:
|
|
// This must be part of the header in order to make the various 'from_*' functions constexpr.
|
|
// However, sane_mod can only deal with a limited range of values for 'denominator', so this can't be made public.
|
|
ALWAYS_INLINE static constexpr i64 sane_mod(i64& numerator, i64 denominator)
|
|
{
|
|
VERIFY(2 <= denominator && denominator <= 1'000'000'000);
|
|
// '%' in C/C++ does not work in the obvious way:
|
|
// For example, -9 % 7 is -2, not +5.
|
|
// However, we want a representation like "(-2)*7 + (+5)".
|
|
i64 dividend = numerator / denominator;
|
|
numerator %= denominator;
|
|
if (numerator < 0) {
|
|
// Does not overflow: different signs.
|
|
numerator += denominator;
|
|
// Does not underflow: denominator >= 2.
|
|
dividend -= 1;
|
|
}
|
|
return dividend;
|
|
}
|
|
ALWAYS_INLINE static constexpr i32 sane_mod(i32& numerator, i32 denominator)
|
|
{
|
|
i64 numerator_64 = numerator;
|
|
i64 dividend = sane_mod(numerator_64, denominator);
|
|
// Does not underflow: numerator can only become smaller.
|
|
numerator = static_cast<i32>(numerator_64);
|
|
// Does not overflow: Will be smaller than original value of 'numerator'.
|
|
return static_cast<i32>(dividend);
|
|
}
|
|
|
|
public:
|
|
[[nodiscard]] constexpr static Time from_timestamp(i32 year, u8 month, u8 day, u8 hour, u8 minute, u8 second, u16 millisecond)
|
|
{
|
|
constexpr auto milliseconds_per_day = 86'400'000;
|
|
constexpr auto milliseconds_per_hour = 3'600'000;
|
|
constexpr auto milliseconds_per_minute = 60'000;
|
|
constexpr auto milliseconds_per_second = 1'000;
|
|
|
|
i64 milliseconds_since_epoch = days_since_epoch(year, month, day);
|
|
milliseconds_since_epoch *= milliseconds_per_day;
|
|
|
|
milliseconds_since_epoch += hour * milliseconds_per_hour;
|
|
milliseconds_since_epoch += minute * milliseconds_per_minute;
|
|
milliseconds_since_epoch += second * milliseconds_per_second;
|
|
milliseconds_since_epoch += millisecond;
|
|
|
|
return from_milliseconds(milliseconds_since_epoch);
|
|
}
|
|
|
|
[[nodiscard]] constexpr static Time from_seconds(i64 seconds) { return Time(seconds, 0); }
|
|
[[nodiscard]] constexpr static Time from_nanoseconds(i64 nanoseconds)
|
|
{
|
|
i64 seconds = sane_mod(nanoseconds, 1'000'000'000);
|
|
return Time(seconds, nanoseconds);
|
|
}
|
|
[[nodiscard]] constexpr static Time from_microseconds(i64 microseconds)
|
|
{
|
|
i64 seconds = sane_mod(microseconds, 1'000'000);
|
|
return Time(seconds, microseconds * 1'000);
|
|
}
|
|
[[nodiscard]] constexpr static Time from_milliseconds(i64 milliseconds)
|
|
{
|
|
i64 seconds = sane_mod(milliseconds, 1'000);
|
|
return Time(seconds, milliseconds * 1'000'000);
|
|
}
|
|
[[nodiscard]] static Time from_ticks(clock_t, time_t);
|
|
[[nodiscard]] static Time from_timespec(const struct timespec&);
|
|
[[nodiscard]] static Time from_timeval(const struct timeval&);
|
|
// We don't pull in <stdint.h> for the pretty min/max definitions because this file is also included in the Kernel
|
|
[[nodiscard]] constexpr static Time min() { return Time(-__INT64_MAX__ - 1LL, 0); };
|
|
[[nodiscard]] constexpr static Time zero() { return Time(0, 0); };
|
|
[[nodiscard]] constexpr static Time max() { return Time(__INT64_MAX__, 999'999'999); };
|
|
|
|
#ifndef KERNEL
|
|
[[nodiscard]] static Time now_realtime();
|
|
[[nodiscard]] static Time now_realtime_coarse();
|
|
[[nodiscard]] static Time now_monotonic();
|
|
[[nodiscard]] static Time now_monotonic_coarse();
|
|
#endif
|
|
|
|
// Truncates towards zero (2.8s to 2s, -2.8s to -2s).
|
|
[[nodiscard]] i64 to_truncated_seconds() const;
|
|
[[nodiscard]] i64 to_truncated_milliseconds() const;
|
|
[[nodiscard]] i64 to_truncated_microseconds() const;
|
|
// Rounds away from zero (2.3s to 3s, -2.3s to -3s).
|
|
[[nodiscard]] i64 to_seconds() const;
|
|
[[nodiscard]] i64 to_milliseconds() const;
|
|
[[nodiscard]] i64 to_microseconds() const;
|
|
[[nodiscard]] i64 to_nanoseconds() const;
|
|
[[nodiscard]] timespec to_timespec() const;
|
|
// Rounds towards -inf (it was the easiest to implement).
|
|
[[nodiscard]] timeval to_timeval() const;
|
|
|
|
[[nodiscard]] bool is_zero() const { return (m_seconds == 0) && (m_nanoseconds == 0); }
|
|
[[nodiscard]] bool is_negative() const { return m_seconds < 0; }
|
|
|
|
Time operator+(Time const& other) const;
|
|
Time& operator+=(Time const& other);
|
|
Time operator-(Time const& other) const;
|
|
Time& operator-=(Time const& other);
|
|
|
|
constexpr bool operator==(Time const& other) const = default;
|
|
constexpr int operator<=>(Time const& other) const
|
|
{
|
|
if (int cmp = (m_seconds > other.m_seconds ? 1 : m_seconds < other.m_seconds ? -1
|
|
: 0);
|
|
cmp != 0)
|
|
return cmp;
|
|
if (int cmp = (m_nanoseconds > other.m_nanoseconds ? 1 : m_nanoseconds < other.m_nanoseconds ? -1
|
|
: 0);
|
|
cmp != 0)
|
|
return cmp;
|
|
return 0;
|
|
}
|
|
|
|
private:
|
|
constexpr explicit Time(i64 seconds, u32 nanoseconds)
|
|
: m_seconds(seconds)
|
|
, m_nanoseconds(nanoseconds)
|
|
{
|
|
}
|
|
|
|
[[nodiscard]] static Time from_half_sanitized(i64 seconds, i32 extra_seconds, u32 nanoseconds);
|
|
|
|
i64 m_seconds { 0 };
|
|
u32 m_nanoseconds { 0 }; // Always less than 1'000'000'000
|
|
};
|
|
|
|
template<typename TimevalType>
|
|
inline void timeval_sub(TimevalType const& a, TimevalType const& b, TimevalType& result)
|
|
{
|
|
result.tv_sec = a.tv_sec - b.tv_sec;
|
|
result.tv_usec = a.tv_usec - b.tv_usec;
|
|
if (result.tv_usec < 0) {
|
|
--result.tv_sec;
|
|
result.tv_usec += 1'000'000;
|
|
}
|
|
}
|
|
|
|
template<typename TimevalType>
|
|
inline void timeval_add(TimevalType const& a, TimevalType const& b, TimevalType& result)
|
|
{
|
|
result.tv_sec = a.tv_sec + b.tv_sec;
|
|
result.tv_usec = a.tv_usec + b.tv_usec;
|
|
if (result.tv_usec >= 1'000'000) {
|
|
++result.tv_sec;
|
|
result.tv_usec -= 1'000'000;
|
|
}
|
|
}
|
|
|
|
template<typename TimespecType>
|
|
inline void timespec_sub(TimespecType const& a, TimespecType const& b, TimespecType& result)
|
|
{
|
|
result.tv_sec = a.tv_sec - b.tv_sec;
|
|
result.tv_nsec = a.tv_nsec - b.tv_nsec;
|
|
if (result.tv_nsec < 0) {
|
|
--result.tv_sec;
|
|
result.tv_nsec += 1'000'000'000;
|
|
}
|
|
}
|
|
|
|
template<typename TimespecType>
|
|
inline void timespec_add(TimespecType const& a, TimespecType const& b, TimespecType& result)
|
|
{
|
|
result.tv_sec = a.tv_sec + b.tv_sec;
|
|
result.tv_nsec = a.tv_nsec + b.tv_nsec;
|
|
if (result.tv_nsec >= 1000'000'000) {
|
|
++result.tv_sec;
|
|
result.tv_nsec -= 1000'000'000;
|
|
}
|
|
}
|
|
|
|
template<typename TimespecType, typename TimevalType>
|
|
inline void timespec_add_timeval(TimespecType const& a, TimevalType const& b, TimespecType& result)
|
|
{
|
|
result.tv_sec = a.tv_sec + b.tv_sec;
|
|
result.tv_nsec = a.tv_nsec + b.tv_usec * 1000;
|
|
if (result.tv_nsec >= 1000'000'000) {
|
|
++result.tv_sec;
|
|
result.tv_nsec -= 1000'000'000;
|
|
}
|
|
}
|
|
|
|
template<typename TimevalType, typename TimespecType>
|
|
inline void timeval_to_timespec(TimevalType const& tv, TimespecType& ts)
|
|
{
|
|
ts.tv_sec = tv.tv_sec;
|
|
ts.tv_nsec = tv.tv_usec * 1000;
|
|
}
|
|
|
|
template<typename TimespecType, typename TimevalType>
|
|
inline void timespec_to_timeval(TimespecType const& ts, TimevalType& tv)
|
|
{
|
|
tv.tv_sec = ts.tv_sec;
|
|
tv.tv_usec = ts.tv_nsec / 1000;
|
|
}
|
|
|
|
// To use these, add a ``using namespace AK::TimeLiterals`` at block or file scope
|
|
namespace TimeLiterals {
|
|
|
|
constexpr Time operator""_ns(unsigned long long nanoseconds) { return Time::from_nanoseconds(static_cast<i64>(nanoseconds)); }
|
|
constexpr Time operator""_us(unsigned long long microseconds) { return Time::from_microseconds(static_cast<i64>(microseconds)); }
|
|
constexpr Time operator""_ms(unsigned long long milliseconds) { return Time::from_milliseconds(static_cast<i64>(milliseconds)); }
|
|
constexpr Time operator""_sec(unsigned long long seconds) { return Time::from_seconds(static_cast<i64>(seconds)); }
|
|
|
|
}
|
|
|
|
}
|
|
|
|
#if USING_AK_GLOBALLY
|
|
using AK::day_of_week;
|
|
using AK::day_of_year;
|
|
using AK::days_in_month;
|
|
using AK::days_in_year;
|
|
using AK::days_since_epoch;
|
|
using AK::is_leap_year;
|
|
using AK::seconds_since_epoch_to_year;
|
|
using AK::Time;
|
|
using AK::timespec_add;
|
|
using AK::timespec_add_timeval;
|
|
using AK::timespec_sub;
|
|
using AK::timespec_to_timeval;
|
|
using AK::timeval_add;
|
|
using AK::timeval_sub;
|
|
using AK::timeval_to_timespec;
|
|
using AK::years_to_days_since_epoch;
|
|
#endif
|