platform_bionic/tests/time_test.cpp

1485 lines
39 KiB
C++
Raw Normal View History

/*
* Copyright (C) 2013 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include <time.h>
#include <errno.h>
#include <gtest/gtest.h>
#include <pthread.h>
#include <signal.h>
#include <sys/cdefs.h>
#include <sys/syscall.h>
#include <sys/types.h>
#include <sys/wait.h>
#include <unistd.h>
#include <atomic>
#include <chrono>
Expose tzalloc()/localtime_rz()/mktime_z()/tzfree(). * Rationale The question often comes up of how to use multiple time zones in C code. If you're single-threaded, you can just use setenv() to manipulate $TZ. toybox does this, for example. But that's not thread-safe in two distinct ways: firstly, getenv() is not thread-safe with respect to modifications to the environment (and between the way putenv() is specified and the existence of environ, it's not obvious how to fully fix that), and secondly the _caller_ needs to ensure that no other threads are using tzset() or any function that behaves "as if" tzset() was called (which is neither easy to determine nor easy to ensure). This isn't a bigger problem because most of the time the right answer is to stop pretending that libc is at all suitable for any i18n, and switch to icu4c instead. (The NDK icu4c headers do not include ucal_*, so this is not a realistic option for most applications.) But what if you're somewhere in between? Like the rust chrono library, for example? What then? Currently their "least worst" option is to reinvent the entire wheel and read our tzdata files. Which isn't a great solution for anyone, for obvious maintainability reasons. So it's probably time we broke the catch-22 here and joined NetBSD in offering a less broken API than standard C has for the last 40 years. Sure, any would-be caller will have to have a separate "is this Android?" and even "is this API level >= 35?" path, but that will fix itself sometime in the 2030s when developers can just assume "yes, it is", whereas if we keep putting off exposing anything, this problem never gets solved. (No-one's bothered to try to implement the std::chrono::time_zone functionality in libc++ yet, but they'll face a similar problem if/when they do.) * Implementation The good news is that tzcode already implements these functions, so there's relatively little here. I've chosen not to expose `struct state` because `struct __timezone_t` makes for clearer error messages, given that compiler diagnostics will show the underlying type name (`struct __timezone_t*`) rather than the typedef name (`timezone_t`) that's used in calling code. I've moved us over to FreeBSD's wcsftime() rather than keep the OpenBSD one building --- I've long wanted to only have one implementation here, and FreeBSD is already doing the "convert back and forth, calling the non-wide function in the middle" dance that I'd hoped to get round to doing myself someday. This should mean that our strftime() and wcsftime() behaviors can't easily diverge in future, plus macOS/iOS are mostly FreeBSD, so any bugs will likely be interoperable with the other major mobile operating system, so there's something nice for everyone there! The FreeBSD wcsftime() implementation includes a wcsftime_l() implementation, so that's one stub we can remove. The flip side of that is that it uses mbsrtowcs_l() and wcsrtombs_l() which we didn't previously have. So expose those as aliases of mbsrtowcs() and wcsrtombs(). Bug: https://github.com/chronotope/chrono/issues/499 Test: treehugger Change-Id: Iee1b9d763ead15eef3d2c33666b3403b68940c3c
2023-06-15 22:17:08 +02:00
#include <thread>
#include "SignalUtils.h"
#include "android-base/file.h"
#include "android-base/strings.h"
#include "utils.h"
using namespace std::chrono_literals;
TEST(time, time) {
// Acquire time
time_t p1, t1 = time(&p1);
// valid?
ASSERT_NE(static_cast<time_t>(0), t1);
ASSERT_NE(static_cast<time_t>(-1), t1);
ASSERT_EQ(p1, t1);
// Acquire time one+ second later
usleep(1010000);
time_t p2, t2 = time(&p2);
// valid?
ASSERT_NE(static_cast<time_t>(0), t2);
ASSERT_NE(static_cast<time_t>(-1), t2);
ASSERT_EQ(p2, t2);
// Expect time progression
ASSERT_LT(p1, p2);
ASSERT_LE(t2 - t1, static_cast<time_t>(2));
// Expect nullptr call to produce same results
ASSERT_LE(t2, time(nullptr));
ASSERT_LE(time(nullptr) - t2, static_cast<time_t>(1));
}
TEST(time, gmtime) {
time_t t = 0;
tm* broken_down = gmtime(&t);
ASSERT_TRUE(broken_down != nullptr);
ASSERT_EQ(0, broken_down->tm_sec);
ASSERT_EQ(0, broken_down->tm_min);
ASSERT_EQ(0, broken_down->tm_hour);
ASSERT_EQ(1, broken_down->tm_mday);
ASSERT_EQ(0, broken_down->tm_mon);
ASSERT_EQ(1970, broken_down->tm_year + 1900);
}
TEST(time, gmtime_r) {
struct tm tm = {};
time_t t = 0;
struct tm* broken_down = gmtime_r(&t, &tm);
ASSERT_EQ(broken_down, &tm);
ASSERT_EQ(0, broken_down->tm_sec);
ASSERT_EQ(0, broken_down->tm_min);
ASSERT_EQ(0, broken_down->tm_hour);
ASSERT_EQ(1, broken_down->tm_mday);
ASSERT_EQ(0, broken_down->tm_mon);
ASSERT_EQ(1970, broken_down->tm_year + 1900);
}
TEST(time, mktime_TZ_as_UTC_and_offset) {
struct tm tm = {.tm_year = 70, .tm_mon = 0, .tm_mday = 1};
// This TZ value is not a valid Olson ID and is not present in tzdata file,
// but is a valid TZ string according to POSIX standard.
setenv("TZ", "UTC+08:00:00", 1);
tzset();
ASSERT_EQ(static_cast<time_t>(8 * 60 * 60), mktime(&tm));
}
static void* gmtime_no_stack_overflow_14313703_fn(void*) {
const char* original_tz = getenv("TZ");
// Ensure we'll actually have to enter tzload by using a timezone that doesn't exist.
setenv("TZ", "gmtime_stack_overflow_14313703", 1);
tzset();
if (original_tz != nullptr) {
setenv("TZ", original_tz, 1);
}
tzset();
return nullptr;
}
TEST(time, gmtime_no_stack_overflow_14313703) {
// Is it safe to call tzload on a thread with a small stack?
// http://b/14313703
// https://code.google.com/p/android/issues/detail?id=61130
pthread_attr_t a;
ASSERT_EQ(0, pthread_attr_init(&a));
ASSERT_EQ(0, pthread_attr_setstacksize(&a, PTHREAD_STACK_MIN));
pthread_t t;
ASSERT_EQ(0, pthread_create(&t, &a, gmtime_no_stack_overflow_14313703_fn, nullptr));
ASSERT_EQ(0, pthread_join(t, nullptr));
}
TEST(time, mktime_empty_TZ) {
// tzcode used to have a bug where it didn't reinitialize some internal state.
// Choose a time where DST is set.
struct tm t;
memset(&t, 0, sizeof(tm));
t.tm_year = 1980 - 1900;
t.tm_mon = 6;
t.tm_mday = 2;
setenv("TZ", "America/Los_Angeles", 1);
tzset();
ASSERT_EQ(static_cast<time_t>(331372800U), mktime(&t));
memset(&t, 0, sizeof(tm));
t.tm_year = 1980 - 1900;
t.tm_mon = 6;
t.tm_mday = 2;
setenv("TZ", "", 1); // Implies UTC.
tzset();
ASSERT_EQ(static_cast<time_t>(331344000U), mktime(&t));
}
TEST(time, mktime_10310929) {
struct tm tm = {.tm_year = 2100 - 1900, .tm_mon = 2, .tm_mday = 10};
#if !defined(__LP64__)
// 32-bit bionic has a signed 32-bit time_t.
ASSERT_EQ(-1, mktime(&tm));
ASSERT_ERRNO(EOVERFLOW);
#else
// Everyone else should be using a signed 64-bit time_t.
ASSERT_GE(sizeof(time_t) * 8, 64U);
setenv("TZ", "America/Los_Angeles", 1);
tzset();
errno = 0;
// On the date/time specified by tm America/Los_Angeles
// follows DST. But tm_isdst is set to 0, which forces
// mktime to interpret that time as local standard, hence offset
// is 8 hours, not 7.
ASSERT_EQ(static_cast<time_t>(4108348800U), mktime(&tm));
ASSERT_ERRNO(0);
#endif
}
TEST(time, mktime_EOVERFLOW) {
setenv("TZ", "UTC", 1);
struct tm t;
memset(&t, 0, sizeof(tm));
// LP32 year range is 1901-2038, so this year is guaranteed not to overflow.
t.tm_year = 2016 - 1900;
t.tm_mon = 2;
t.tm_mday = 10;
errno = 0;
ASSERT_NE(static_cast<time_t>(-1), mktime(&t));
ASSERT_ERRNO(0);
// This will overflow for LP32.
t.tm_year = INT_MAX;
errno = 0;
#if !defined(__LP64__)
ASSERT_EQ(static_cast<time_t>(-1), mktime(&t));
ASSERT_ERRNO(EOVERFLOW);
#else
ASSERT_EQ(static_cast<time_t>(67768036166016000U), mktime(&t));
ASSERT_ERRNO(0);
#endif
// This will overflow for LP32 or LP64.
// tm_year is int, this t struct points to INT_MAX + 1 no matter what TZ is.
t.tm_year = INT_MAX;
t.tm_mon = 11;
t.tm_mday = 45;
errno = 0;
ASSERT_EQ(static_cast<time_t>(-1), mktime(&t));
ASSERT_ERRNO(EOVERFLOW);
}
TEST(time, mktime_invalid_tm_TZ_combination) {
setenv("TZ", "UTC", 1);
struct tm t;
memset(&t, 0, sizeof(tm));
t.tm_year = 2022 - 1900;
t.tm_mon = 11;
t.tm_mday = 31;
// UTC does not observe DST
t.tm_isdst = 1;
errno = 0;
EXPECT_EQ(static_cast<time_t>(-1), mktime(&t));
// mktime sets errno to EOVERFLOW if result is unrepresentable.
EXPECT_ERRNO(EOVERFLOW);
}
// Transitions in the tzdata file are generated up to the year 2100. Testing
// that dates beyond that are handled properly too.
TEST(time, mktime_after_2100) {
struct tm tm = {.tm_year = 2150 - 1900, .tm_mon = 2, .tm_mday = 10, .tm_isdst = -1};
#if !defined(__LP64__)
// 32-bit bionic has a signed 32-bit time_t.
ASSERT_EQ(-1, mktime(&tm));
ASSERT_ERRNO(EOVERFLOW);
#else
setenv("TZ", "Europe/London", 1);
tzset();
errno = 0;
ASSERT_EQ(static_cast<time_t>(5686156800U), mktime(&tm));
ASSERT_ERRNO(0);
#endif
}
TEST(time, strftime) {
setenv("TZ", "UTC", 1);
struct tm t;
memset(&t, 0, sizeof(tm));
t.tm_year = 200;
t.tm_mon = 2;
t.tm_mday = 10;
char buf[64];
// Seconds since the epoch.
#if defined(__BIONIC__) || defined(__LP64__) // Not 32-bit glibc.
EXPECT_EQ(10U, strftime(buf, sizeof(buf), "%s", &t));
EXPECT_STREQ("4108320000", buf);
#endif
// Date and time as text.
EXPECT_EQ(24U, strftime(buf, sizeof(buf), "%c", &t));
EXPECT_STREQ("Sun Mar 10 00:00:00 2100", buf);
}
TEST(time, strftime_second_before_epoch) {
setenv("TZ", "UTC", 1);
struct tm t;
memset(&t, 0, sizeof(tm));
t.tm_year = 1969 - 1900;
t.tm_mon = 11;
t.tm_mday = 31;
t.tm_hour = 23;
t.tm_min = 59;
t.tm_sec = 59;
char buf[64];
EXPECT_EQ(2U, strftime(buf, sizeof(buf), "%s", &t));
EXPECT_STREQ("-1", buf);
}
TEST(time, strftime_Z_null_tm_zone) {
// Netflix on Nexus Player wouldn't start (http://b/25170306).
struct tm t;
memset(&t, 0, sizeof(tm));
char buf[64];
setenv("TZ", "America/Los_Angeles", 1);
tzset();
t.tm_isdst = 0; // "0 if Daylight Savings Time is not in effect".
EXPECT_EQ(5U, strftime(buf, sizeof(buf), "<%Z>", &t));
EXPECT_STREQ("<PST>", buf);
#if defined(__BIONIC__) // glibc 2.19 only copes with tm_isdst being 0 and 1.
t.tm_isdst = 2; // "positive if Daylight Savings Time is in effect"
EXPECT_EQ(5U, strftime(buf, sizeof(buf), "<%Z>", &t));
EXPECT_STREQ("<PDT>", buf);
t.tm_isdst = -123; // "and negative if the information is not available".
EXPECT_EQ(2U, strftime(buf, sizeof(buf), "<%Z>", &t));
EXPECT_STREQ("<>", buf);
#endif
setenv("TZ", "UTC", 1);
tzset();
t.tm_isdst = 0;
EXPECT_EQ(5U, strftime(buf, sizeof(buf), "<%Z>", &t));
EXPECT_STREQ("<UTC>", buf);
#if defined(__BIONIC__) // glibc 2.19 thinks UTC DST is "UTC".
t.tm_isdst = 1; // UTC has no DST.
EXPECT_EQ(2U, strftime(buf, sizeof(buf), "<%Z>", &t));
EXPECT_STREQ("<>", buf);
#endif
}
// According to C language specification the only tm struct field needed to
// find out replacement for %z and %Z in strftime is tm_isdst. Which is
// wrong, as timezones change their standard offset and even DST savings.
// tzcode deviates from C language specification and requires tm struct either
// to be output of localtime-like functions or to be modified by mktime call
// before passing to strftime. See tz mailing discussion for more details
// https://mm.icann.org/pipermail/tz/2022-July/031674.html
// But we are testing case when tm.tm_zone is null, which means that tm struct
// is not coming from localtime and is neither modified by mktime. That's why
// we are comparing against +0000, even though America/Los_Angeles never
// observes it.
TEST(time, strftime_z_null_tm_zone) {
char str[64];
struct tm tm = {.tm_year = 109, .tm_mon = 4, .tm_mday = 2, .tm_isdst = 0};
setenv("TZ", "America/Los_Angeles", 1);
tzset();
tm.tm_zone = NULL;
size_t result = strftime(str, sizeof(str), "%z", &tm);
EXPECT_EQ(5U, result);
EXPECT_STREQ("+0000", str);
tm.tm_isdst = 1;
result = strftime(str, sizeof(str), "%z", &tm);
EXPECT_EQ(5U, result);
EXPECT_STREQ("+0000", str);
setenv("TZ", "UTC", 1);
tzset();
tm.tm_isdst = 0;
result = strftime(str, sizeof(str), "%z", &tm);
EXPECT_EQ(5U, result);
EXPECT_STREQ("+0000", str);
tm.tm_isdst = 1;
result = strftime(str, sizeof(str), "%z", &tm);
EXPECT_EQ(5U, result);
EXPECT_STREQ("+0000", str);
}
TEST(time, strftime_z_Europe_Lisbon) {
char str[64];
// During 1992-1996 Europe/Lisbon standard offset was 1 hour.
// tm_isdst is not set as it will be overridden by mktime call anyway.
struct tm tm = {.tm_year = 1996 - 1900, .tm_mon = 2, .tm_mday = 13};
setenv("TZ", "Europe/Lisbon", 1);
tzset();
// tzcode's strftime implementation for %z relies on prior mktime call.
// At the moment of writing %z value is taken from tm_gmtoff. So without
// mktime call %z is replaced with +0000.
// See https://mm.icann.org/pipermail/tz/2022-July/031674.html
mktime(&tm);
size_t result = strftime(str, sizeof(str), "%z", &tm);
EXPECT_EQ(5U, result);
EXPECT_STREQ("+0100", str);
// Now standard offset is 0.
tm = {.tm_year = 2022 - 1900, .tm_mon = 2, .tm_mday = 13};
mktime(&tm);
result = strftime(str, sizeof(str), "%z", &tm);
EXPECT_EQ(5U, result);
EXPECT_STREQ("+0000", str);
}
TEST(time, strftime_l) {
locale_t cloc = newlocale(LC_ALL, "C.UTF-8", nullptr);
locale_t old_locale = uselocale(cloc);
setenv("TZ", "UTC", 1);
struct tm t;
memset(&t, 0, sizeof(tm));
t.tm_year = 200;
t.tm_mon = 2;
t.tm_mday = 10;
// Date and time as text.
char buf[64];
EXPECT_EQ(24U, strftime_l(buf, sizeof(buf), "%c", &t, cloc));
EXPECT_STREQ("Sun Mar 10 00:00:00 2100", buf);
uselocale(old_locale);
freelocale(cloc);
}
TEST(time, strptime) {
setenv("TZ", "UTC", 1);
struct tm t;
char buf[64];
memset(&t, 0, sizeof(t));
strptime("11:14", "%R", &t);
strftime(buf, sizeof(buf), "%H:%M", &t);
EXPECT_STREQ("11:14", buf);
memset(&t, 0, sizeof(t));
strptime("09:41:53", "%T", &t);
strftime(buf, sizeof(buf), "%H:%M:%S", &t);
EXPECT_STREQ("09:41:53", buf);
}
TEST(time, strptime_l) {
#if !defined(ANDROID_HOST_MUSL)
setenv("TZ", "UTC", 1);
struct tm t;
char buf[64];
memset(&t, 0, sizeof(t));
strptime_l("11:14", "%R", &t, LC_GLOBAL_LOCALE);
strftime_l(buf, sizeof(buf), "%H:%M", &t, LC_GLOBAL_LOCALE);
EXPECT_STREQ("11:14", buf);
memset(&t, 0, sizeof(t));
strptime_l("09:41:53", "%T", &t, LC_GLOBAL_LOCALE);
strftime_l(buf, sizeof(buf), "%H:%M:%S", &t, LC_GLOBAL_LOCALE);
EXPECT_STREQ("09:41:53", buf);
#else
GTEST_SKIP() << "musl doesn't support strptime_l";
#endif
}
TEST(time, strptime_F) {
setenv("TZ", "UTC", 1);
struct tm tm = {};
ASSERT_EQ('\0', *strptime("2019-03-26", "%F", &tm));
EXPECT_EQ(119, tm.tm_year);
EXPECT_EQ(2, tm.tm_mon);
EXPECT_EQ(26, tm.tm_mday);
}
TEST(time, strptime_P_p) {
setenv("TZ", "UTC", 1);
// For parsing, %P and %p are the same: case doesn't matter.
struct tm tm = {.tm_hour = 12};
ASSERT_EQ('\0', *strptime("AM", "%p", &tm));
EXPECT_EQ(0, tm.tm_hour);
tm = {.tm_hour = 12};
ASSERT_EQ('\0', *strptime("am", "%p", &tm));
EXPECT_EQ(0, tm.tm_hour);
tm = {.tm_hour = 12};
ASSERT_EQ('\0', *strptime("AM", "%P", &tm));
EXPECT_EQ(0, tm.tm_hour);
tm = {.tm_hour = 12};
ASSERT_EQ('\0', *strptime("am", "%P", &tm));
EXPECT_EQ(0, tm.tm_hour);
}
TEST(time, strptime_u) {
setenv("TZ", "UTC", 1);
struct tm tm = {};
ASSERT_EQ('\0', *strptime("2", "%u", &tm));
EXPECT_EQ(2, tm.tm_wday);
}
TEST(time, strptime_v) {
setenv("TZ", "UTC", 1);
struct tm tm = {};
ASSERT_EQ('\0', *strptime("26-Mar-1980", "%v", &tm));
EXPECT_EQ(80, tm.tm_year);
EXPECT_EQ(2, tm.tm_mon);
EXPECT_EQ(26, tm.tm_mday);
}
TEST(time, strptime_V_G_g) {
setenv("TZ", "UTC", 1);
// %V (ISO-8601 week number), %G (year of week number, without century), and
// %g (year of week number) have no effect when parsed, and are supported
// solely so that it's possible for strptime(3) to parse everything that
// strftime(3) can output.
struct tm tm = {};
ASSERT_EQ('\0', *strptime("1 2 3", "%V %G %g", &tm));
struct tm zero = {};
EXPECT_TRUE(memcmp(&tm, &zero, sizeof(tm)) == 0);
}
TEST(time, strptime_Z) {
#if defined(__BIONIC__)
// glibc doesn't handle %Z at all.
// The BSDs only handle hard-coded "GMT" and "UTC", plus whatever two strings
// are in the global `tzname` (which correspond to the current $TZ).
struct tm tm;
setenv("TZ", "Europe/Berlin", 1);
// "GMT" always works.
tm = {};
ASSERT_EQ('\0', *strptime("GMT", "%Z", &tm));
EXPECT_STREQ("GMT", tm.tm_zone);
EXPECT_EQ(0, tm.tm_isdst);
EXPECT_EQ(0, tm.tm_gmtoff);
// As does "UTC".
tm = {};
ASSERT_EQ('\0', *strptime("UTC", "%Z", &tm));
EXPECT_STREQ("UTC", tm.tm_zone);
EXPECT_EQ(0, tm.tm_isdst);
EXPECT_EQ(0, tm.tm_gmtoff);
// Europe/Berlin is known as "CET" when there's no DST.
tm = {};
ASSERT_EQ('\0', *strptime("CET", "%Z", &tm));
EXPECT_STREQ("CET", tm.tm_zone);
EXPECT_EQ(0, tm.tm_isdst);
EXPECT_EQ(3600, tm.tm_gmtoff);
// Europe/Berlin is known as "CEST" when there's no DST.
tm = {};
ASSERT_EQ('\0', *strptime("CEST", "%Z", &tm));
EXPECT_STREQ("CEST", tm.tm_zone);
EXPECT_EQ(1, tm.tm_isdst);
EXPECT_EQ(3600, tm.tm_gmtoff);
// And as long as we're in Europe/Berlin, those are the only timezone
// abbreviations that are recognized.
tm = {};
ASSERT_TRUE(strptime("PDT", "%Z", &tm) == nullptr);
#endif
}
TEST(time, strptime_z) {
struct tm tm;
setenv("TZ", "Europe/Berlin", 1);
// "UT" is what RFC822 called UTC.
tm = {};
ASSERT_EQ('\0', *strptime("UT", "%z", &tm));
EXPECT_STREQ("UTC", tm.tm_zone);
EXPECT_EQ(0, tm.tm_isdst);
EXPECT_EQ(0, tm.tm_gmtoff);
// "GMT" is RFC822's other name for UTC.
tm = {};
ASSERT_EQ('\0', *strptime("GMT", "%z", &tm));
EXPECT_STREQ("UTC", tm.tm_zone);
EXPECT_EQ(0, tm.tm_isdst);
EXPECT_EQ(0, tm.tm_gmtoff);
// "Z" ("Zulu") is a synonym for UTC.
tm = {};
ASSERT_EQ('\0', *strptime("Z", "%z", &tm));
EXPECT_STREQ("UTC", tm.tm_zone);
EXPECT_EQ(0, tm.tm_isdst);
EXPECT_EQ(0, tm.tm_gmtoff);
// "PST"/"PDT" and the other common US zone abbreviations are all supported.
tm = {};
ASSERT_EQ('\0', *strptime("PST", "%z", &tm));
EXPECT_STREQ("PST", tm.tm_zone);
EXPECT_EQ(0, tm.tm_isdst);
EXPECT_EQ(-28800, tm.tm_gmtoff);
tm = {};
ASSERT_EQ('\0', *strptime("PDT", "%z", &tm));
EXPECT_STREQ("PDT", tm.tm_zone);
EXPECT_EQ(1, tm.tm_isdst);
EXPECT_EQ(-25200, tm.tm_gmtoff);
// +-hh
tm = {};
ASSERT_EQ('\0', *strptime("+01", "%z", &tm));
EXPECT_EQ(3600, tm.tm_gmtoff);
EXPECT_TRUE(tm.tm_zone == nullptr);
EXPECT_EQ(0, tm.tm_isdst);
// +-hhmm
tm = {};
ASSERT_EQ('\0', *strptime("+0130", "%z", &tm));
EXPECT_EQ(5400, tm.tm_gmtoff);
EXPECT_TRUE(tm.tm_zone == nullptr);
EXPECT_EQ(0, tm.tm_isdst);
// +-hh:mm
tm = {};
ASSERT_EQ('\0', *strptime("+01:30", "%z", &tm));
EXPECT_EQ(5400, tm.tm_gmtoff);
EXPECT_TRUE(tm.tm_zone == nullptr);
EXPECT_EQ(0, tm.tm_isdst);
}
void SetTime(timer_t t, time_t value_s, time_t value_ns, time_t interval_s, time_t interval_ns) {
itimerspec ts;
ts.it_value.tv_sec = value_s;
ts.it_value.tv_nsec = value_ns;
ts.it_interval.tv_sec = interval_s;
ts.it_interval.tv_nsec = interval_ns;
ASSERT_EQ(0, timer_settime(t, 0, &ts, nullptr));
}
static void NoOpNotifyFunction(sigval) {
}
TEST(time, timer_create) {
sigevent se;
memset(&se, 0, sizeof(se));
se.sigev_notify = SIGEV_THREAD;
se.sigev_notify_function = NoOpNotifyFunction;
timer_t timer_id;
ASSERT_EQ(0, timer_create(CLOCK_MONOTONIC, &se, &timer_id));
pid_t pid = fork();
ASSERT_NE(-1, pid) << strerror(errno);
if (pid == 0) {
// Timers are not inherited by the child.
ASSERT_EQ(-1, timer_delete(timer_id));
ASSERT_ERRNO(EINVAL);
_exit(0);
}
AssertChildExited(pid, 0);
ASSERT_EQ(0, timer_delete(timer_id));
}
static int timer_create_SIGEV_SIGNAL_signal_handler_invocation_count;
static void timer_create_SIGEV_SIGNAL_signal_handler(int signal_number) {
++timer_create_SIGEV_SIGNAL_signal_handler_invocation_count;
ASSERT_EQ(SIGUSR1, signal_number);
}
TEST(time, timer_create_SIGEV_SIGNAL) {
sigevent se;
memset(&se, 0, sizeof(se));
se.sigev_notify = SIGEV_SIGNAL;
se.sigev_signo = SIGUSR1;
timer_t timer_id;
ASSERT_EQ(0, timer_create(CLOCK_MONOTONIC, &se, &timer_id));
timer_create_SIGEV_SIGNAL_signal_handler_invocation_count = 0;
ScopedSignalHandler ssh(SIGUSR1, timer_create_SIGEV_SIGNAL_signal_handler);
ASSERT_EQ(0, timer_create_SIGEV_SIGNAL_signal_handler_invocation_count);
itimerspec ts;
ts.it_value.tv_sec = 0;
ts.it_value.tv_nsec = 1;
ts.it_interval.tv_sec = 0;
ts.it_interval.tv_nsec = 0;
ASSERT_EQ(0, timer_settime(timer_id, 0, &ts, nullptr));
usleep(500000);
ASSERT_EQ(1, timer_create_SIGEV_SIGNAL_signal_handler_invocation_count);
}
struct Counter {
private:
std::atomic<int> value;
timer_t timer_id;
sigevent se;
bool timer_valid;
void Create() {
ASSERT_FALSE(timer_valid);
ASSERT_EQ(0, timer_create(CLOCK_REALTIME, &se, &timer_id));
timer_valid = true;
}
public:
explicit Counter(void (*fn)(sigval)) : value(0), timer_valid(false) {
memset(&se, 0, sizeof(se));
se.sigev_notify = SIGEV_THREAD;
se.sigev_notify_function = fn;
se.sigev_value.sival_ptr = this;
Create();
}
void DeleteTimer() {
ASSERT_TRUE(timer_valid);
ASSERT_EQ(0, timer_delete(timer_id));
timer_valid = false;
}
~Counter() {
if (timer_valid) {
DeleteTimer();
}
}
int Value() const {
return value;
}
void SetTime(time_t value_s, time_t value_ns, time_t interval_s, time_t interval_ns) {
::SetTime(timer_id, value_s, value_ns, interval_s, interval_ns);
}
bool ValueUpdated() {
int current_value = value;
time_t start = time(nullptr);
while (current_value == value && (time(nullptr) - start) < 5) {
}
return current_value != value;
}
static void CountNotifyFunction(sigval value) {
Counter* cd = reinterpret_cast<Counter*>(value.sival_ptr);
++cd->value;
}
static void CountAndDisarmNotifyFunction(sigval value) {
Counter* cd = reinterpret_cast<Counter*>(value.sival_ptr);
++cd->value;
// Setting the initial expiration time to 0 disarms the timer.
cd->SetTime(0, 0, 1, 0);
}
};
TEST(time, timer_settime_0) {
Counter counter(Counter::CountAndDisarmNotifyFunction);
ASSERT_EQ(0, counter.Value());
counter.SetTime(0, 500000000, 1, 0);
sleep(1);
// The count should just be 1 because we disarmed the timer the first time it fired.
ASSERT_EQ(1, counter.Value());
}
TEST(time, timer_settime_repeats) {
Counter counter(Counter::CountNotifyFunction);
ASSERT_EQ(0, counter.Value());
counter.SetTime(0, 1, 0, 10);
ASSERT_TRUE(counter.ValueUpdated());
ASSERT_TRUE(counter.ValueUpdated());
ASSERT_TRUE(counter.ValueUpdated());
counter.DeleteTimer();
// Add a sleep as other threads may be calling the callback function when the timer is deleted.
usleep(500000);
}
static int timer_create_NULL_signal_handler_invocation_count;
static void timer_create_NULL_signal_handler(int signal_number) {
++timer_create_NULL_signal_handler_invocation_count;
ASSERT_EQ(SIGALRM, signal_number);
}
TEST(time, timer_create_NULL) {
// A NULL sigevent* is equivalent to asking for SIGEV_SIGNAL for SIGALRM.
timer_t timer_id;
ASSERT_EQ(0, timer_create(CLOCK_MONOTONIC, nullptr, &timer_id));
timer_create_NULL_signal_handler_invocation_count = 0;
ScopedSignalHandler ssh(SIGALRM, timer_create_NULL_signal_handler);
ASSERT_EQ(0, timer_create_NULL_signal_handler_invocation_count);
SetTime(timer_id, 0, 1, 0, 0);
usleep(500000);
ASSERT_EQ(1, timer_create_NULL_signal_handler_invocation_count);
}
static int GetThreadCount() {
std::string status;
if (android::base::ReadFileToString("/proc/self/status", &status)) {
for (const auto& line : android::base::Split(status, "\n")) {
int thread_count;
if (sscanf(line.c_str(), "Threads: %d", &thread_count) == 1) {
return thread_count;
}
}
}
return -1;
}
TEST(time, timer_create_EINVAL) {
const clockid_t kInvalidClock = 16;
// A SIGEV_SIGNAL timer failure is easy; that's the kernel's problem.
timer_t timer_id;
ASSERT_EQ(-1, timer_create(kInvalidClock, nullptr, &timer_id));
ASSERT_ERRNO(EINVAL);
// A SIGEV_THREAD timer failure is more interesting because we have a thread
// to clean up (https://issuetracker.google.com/340125671).
sigevent se = {};
se.sigev_notify = SIGEV_THREAD;
se.sigev_notify_function = NoOpNotifyFunction;
ASSERT_EQ(-1, timer_create(kInvalidClock, &se, &timer_id));
ASSERT_ERRNO(EINVAL);
// timer_create() doesn't guarantee that the thread will be dead _before_
// it returns because that would require extra synchronization that's
// unnecessary in the normal (successful) case. A timeout here means we
// leaked a thread.
while (GetThreadCount() > 1) {
}
}
TEST(time, timer_create_multiple) {
Counter counter1(Counter::CountNotifyFunction);
Counter counter2(Counter::CountNotifyFunction);
Counter counter3(Counter::CountNotifyFunction);
ASSERT_EQ(0, counter1.Value());
ASSERT_EQ(0, counter2.Value());
ASSERT_EQ(0, counter3.Value());
counter2.SetTime(0, 500000000, 0, 0);
sleep(1);
EXPECT_EQ(0, counter1.Value());
EXPECT_EQ(1, counter2.Value());
EXPECT_EQ(0, counter3.Value());
}
// Test to verify that disarming a repeatable timer disables the callbacks.
TEST(time, timer_disarm_terminates) {
Counter counter(Counter::CountNotifyFunction);
ASSERT_EQ(0, counter.Value());
counter.SetTime(0, 1, 0, 1);
ASSERT_TRUE(counter.ValueUpdated());
ASSERT_TRUE(counter.ValueUpdated());
ASSERT_TRUE(counter.ValueUpdated());
counter.SetTime(0, 0, 0, 0);
// Add a sleep as the kernel may have pending events when the timer is disarmed.
usleep(500000);
int value = counter.Value();
usleep(500000);
// Verify the counter has not been incremented.
ASSERT_EQ(value, counter.Value());
}
// Test to verify that deleting a repeatable timer disables the callbacks.
TEST(time, timer_delete_terminates) {
Counter counter(Counter::CountNotifyFunction);
ASSERT_EQ(0, counter.Value());
counter.SetTime(0, 1, 0, 1);
ASSERT_TRUE(counter.ValueUpdated());
ASSERT_TRUE(counter.ValueUpdated());
ASSERT_TRUE(counter.ValueUpdated());
counter.DeleteTimer();
// Add a sleep as other threads may be calling the callback function when the timer is deleted.
usleep(500000);
int value = counter.Value();
usleep(500000);
// Verify the counter has not been incremented.
ASSERT_EQ(value, counter.Value());
}
struct TimerDeleteData {
timer_t timer_id;
Be more strict about using invalid `pthread_t`s. Another release, another attempt to remove the global thread list. But this time, let's admit that it's not going away. We can switch to using a read/write lock for the global thread list, and to aborting rather than quietly returning ESRCH if we're given an invalid pthread_t. This change affects pthread_detach, pthread_getcpuclockid, pthread_getschedparam/pthread_setschedparam, pthread_join, and pthread_kill: instead of returning ESRCH when passed an invalid pthread_t, if you're targeting O or above, they'll abort with the message "attempt to use invalid pthread_t". Note that this doesn't change behavior as much as you might think: the old lookup only held the global thread list lock for the duration of the lookup, so there was still a race between that and the dereference in the caller, given that callers actually need the tid to pass to some syscall or other, and sometimes update fields in the pthread_internal_t struct too. (This patch replaces such users with calls to pthread_gettid_np, which at least makes the TOCTOU window smaller.) We can't check thread->tid against 0 to see whether a pthread_t is still valid because a dead thread gets its thread struct unmapped along with its stack, so the dereference isn't safe. Taking the affected functions one by one: * pthread_getcpuclockid and pthread_getschedparam/pthread_setschedparam should be fine. Unsafe calls to those seem highly unlikely. * Unsafe pthread_detach callers probably want to switch to pthread_attr_setdetachstate instead, or using pthread_detach(pthread_self()) from the new thread's start routine rather than doing the detach in the parent. * pthread_join calls should be safe anyway, because a joinable thread won't actually exit and unmap until it's joined. If you're joining an unjoinable thread, the fix is to stop marking it detached. If you're joining an already-joined thread, you need to rethink your design. * Unsafe pthread_kill calls aren't portably fixable. (And are obviously inherently non-portable as-is.) The best alternative on Android is to use pthread_gettid_np at some point that you know the thread to be alive, and then call kill/tgkill directly. That's still not completely safe because if you're too late, the tid may have been reused, but then your code is inherently unsafe anyway. Bug: http://b/19636317 Test: ran tests Change-Id: I0372c4428e8a7f1c3af5c9334f5d9c25f2c73f21
2017-02-14 02:59:29 +01:00
pid_t tid;
volatile bool complete;
};
static void TimerDeleteCallback(sigval value) {
TimerDeleteData* tdd = reinterpret_cast<TimerDeleteData*>(value.sival_ptr);
Be more strict about using invalid `pthread_t`s. Another release, another attempt to remove the global thread list. But this time, let's admit that it's not going away. We can switch to using a read/write lock for the global thread list, and to aborting rather than quietly returning ESRCH if we're given an invalid pthread_t. This change affects pthread_detach, pthread_getcpuclockid, pthread_getschedparam/pthread_setschedparam, pthread_join, and pthread_kill: instead of returning ESRCH when passed an invalid pthread_t, if you're targeting O or above, they'll abort with the message "attempt to use invalid pthread_t". Note that this doesn't change behavior as much as you might think: the old lookup only held the global thread list lock for the duration of the lookup, so there was still a race between that and the dereference in the caller, given that callers actually need the tid to pass to some syscall or other, and sometimes update fields in the pthread_internal_t struct too. (This patch replaces such users with calls to pthread_gettid_np, which at least makes the TOCTOU window smaller.) We can't check thread->tid against 0 to see whether a pthread_t is still valid because a dead thread gets its thread struct unmapped along with its stack, so the dereference isn't safe. Taking the affected functions one by one: * pthread_getcpuclockid and pthread_getschedparam/pthread_setschedparam should be fine. Unsafe calls to those seem highly unlikely. * Unsafe pthread_detach callers probably want to switch to pthread_attr_setdetachstate instead, or using pthread_detach(pthread_self()) from the new thread's start routine rather than doing the detach in the parent. * pthread_join calls should be safe anyway, because a joinable thread won't actually exit and unmap until it's joined. If you're joining an unjoinable thread, the fix is to stop marking it detached. If you're joining an already-joined thread, you need to rethink your design. * Unsafe pthread_kill calls aren't portably fixable. (And are obviously inherently non-portable as-is.) The best alternative on Android is to use pthread_gettid_np at some point that you know the thread to be alive, and then call kill/tgkill directly. That's still not completely safe because if you're too late, the tid may have been reused, but then your code is inherently unsafe anyway. Bug: http://b/19636317 Test: ran tests Change-Id: I0372c4428e8a7f1c3af5c9334f5d9c25f2c73f21
2017-02-14 02:59:29 +01:00
tdd->tid = gettid();
timer_delete(tdd->timer_id);
tdd->complete = true;
}
TEST(time, timer_delete_from_timer_thread) {
TimerDeleteData tdd;
sigevent se;
memset(&se, 0, sizeof(se));
se.sigev_notify = SIGEV_THREAD;
se.sigev_notify_function = TimerDeleteCallback;
se.sigev_value.sival_ptr = &tdd;
tdd.complete = false;
ASSERT_EQ(0, timer_create(CLOCK_REALTIME, &se, &tdd.timer_id));
itimerspec ts;
ts.it_value.tv_sec = 1;
ts.it_value.tv_nsec = 0;
ts.it_interval.tv_sec = 0;
ts.it_interval.tv_nsec = 0;
ASSERT_EQ(0, timer_settime(tdd.timer_id, 0, &ts, nullptr));
time_t cur_time = time(nullptr);
while (!tdd.complete && (time(nullptr) - cur_time) < 5);
ASSERT_TRUE(tdd.complete);
#if defined(__BIONIC__)
// Since bionic timers are implemented by creating a thread to handle the
// callback, verify that the thread actually completes.
cur_time = time(NULL);
Be more strict about using invalid `pthread_t`s. Another release, another attempt to remove the global thread list. But this time, let's admit that it's not going away. We can switch to using a read/write lock for the global thread list, and to aborting rather than quietly returning ESRCH if we're given an invalid pthread_t. This change affects pthread_detach, pthread_getcpuclockid, pthread_getschedparam/pthread_setschedparam, pthread_join, and pthread_kill: instead of returning ESRCH when passed an invalid pthread_t, if you're targeting O or above, they'll abort with the message "attempt to use invalid pthread_t". Note that this doesn't change behavior as much as you might think: the old lookup only held the global thread list lock for the duration of the lookup, so there was still a race between that and the dereference in the caller, given that callers actually need the tid to pass to some syscall or other, and sometimes update fields in the pthread_internal_t struct too. (This patch replaces such users with calls to pthread_gettid_np, which at least makes the TOCTOU window smaller.) We can't check thread->tid against 0 to see whether a pthread_t is still valid because a dead thread gets its thread struct unmapped along with its stack, so the dereference isn't safe. Taking the affected functions one by one: * pthread_getcpuclockid and pthread_getschedparam/pthread_setschedparam should be fine. Unsafe calls to those seem highly unlikely. * Unsafe pthread_detach callers probably want to switch to pthread_attr_setdetachstate instead, or using pthread_detach(pthread_self()) from the new thread's start routine rather than doing the detach in the parent. * pthread_join calls should be safe anyway, because a joinable thread won't actually exit and unmap until it's joined. If you're joining an unjoinable thread, the fix is to stop marking it detached. If you're joining an already-joined thread, you need to rethink your design. * Unsafe pthread_kill calls aren't portably fixable. (And are obviously inherently non-portable as-is.) The best alternative on Android is to use pthread_gettid_np at some point that you know the thread to be alive, and then call kill/tgkill directly. That's still not completely safe because if you're too late, the tid may have been reused, but then your code is inherently unsafe anyway. Bug: http://b/19636317 Test: ran tests Change-Id: I0372c4428e8a7f1c3af5c9334f5d9c25f2c73f21
2017-02-14 02:59:29 +01:00
while ((kill(tdd.tid, 0) != -1 || errno != ESRCH) && (time(NULL) - cur_time) < 5);
ASSERT_EQ(-1, kill(tdd.tid, 0));
ASSERT_ERRNO(ESRCH);
#endif
}
// Musl doesn't define __NR_clock_gettime on 32-bit architectures.
#if !defined(__NR_clock_gettime)
#define __NR_clock_gettime __NR_clock_gettime32
#endif
TEST(time, clock_gettime) {
// Try to ensure that our vdso clock_gettime is working.
timespec ts0;
timespec ts1;
timespec ts2;
ASSERT_EQ(0, clock_gettime(CLOCK_MONOTONIC, &ts0));
ASSERT_EQ(0, syscall(__NR_clock_gettime, CLOCK_MONOTONIC, &ts1));
ASSERT_EQ(0, clock_gettime(CLOCK_MONOTONIC, &ts2));
// Check we have a nice monotonic timestamp sandwich.
ASSERT_LE(ts0.tv_sec, ts1.tv_sec);
if (ts0.tv_sec == ts1.tv_sec) {
ASSERT_LE(ts0.tv_nsec, ts1.tv_nsec);
}
ASSERT_LE(ts1.tv_sec, ts2.tv_sec);
if (ts1.tv_sec == ts2.tv_sec) {
ASSERT_LE(ts1.tv_nsec, ts2.tv_nsec);
}
}
TEST(time, clock_gettime_CLOCK_REALTIME) {
timespec ts;
ASSERT_EQ(0, clock_gettime(CLOCK_REALTIME, &ts));
}
TEST(time, clock_gettime_CLOCK_MONOTONIC) {
timespec ts;
ASSERT_EQ(0, clock_gettime(CLOCK_MONOTONIC, &ts));
}
TEST(time, clock_gettime_CLOCK_PROCESS_CPUTIME_ID) {
timespec ts;
ASSERT_EQ(0, clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &ts));
}
TEST(time, clock_gettime_CLOCK_THREAD_CPUTIME_ID) {
timespec ts;
ASSERT_EQ(0, clock_gettime(CLOCK_THREAD_CPUTIME_ID, &ts));
}
TEST(time, clock_gettime_CLOCK_BOOTTIME) {
timespec ts;
ASSERT_EQ(0, clock_gettime(CLOCK_BOOTTIME, &ts));
}
TEST(time, clock_gettime_unknown) {
errno = 0;
timespec ts;
ASSERT_EQ(-1, clock_gettime(-1, &ts));
ASSERT_ERRNO(EINVAL);
}
TEST(time, clock_getres_CLOCK_REALTIME) {
timespec ts;
ASSERT_EQ(0, clock_getres(CLOCK_REALTIME, &ts));
ASSERT_EQ(1, ts.tv_nsec);
ASSERT_EQ(0, ts.tv_sec);
}
TEST(time, clock_getres_CLOCK_MONOTONIC) {
timespec ts;
ASSERT_EQ(0, clock_getres(CLOCK_MONOTONIC, &ts));
ASSERT_EQ(1, ts.tv_nsec);
ASSERT_EQ(0, ts.tv_sec);
}
TEST(time, clock_getres_CLOCK_PROCESS_CPUTIME_ID) {
timespec ts;
ASSERT_EQ(0, clock_getres(CLOCK_PROCESS_CPUTIME_ID, &ts));
}
TEST(time, clock_getres_CLOCK_THREAD_CPUTIME_ID) {
timespec ts;
ASSERT_EQ(0, clock_getres(CLOCK_THREAD_CPUTIME_ID, &ts));
}
TEST(time, clock_getres_CLOCK_BOOTTIME) {
timespec ts;
ASSERT_EQ(0, clock_getres(CLOCK_BOOTTIME, &ts));
ASSERT_EQ(1, ts.tv_nsec);
ASSERT_EQ(0, ts.tv_sec);
}
TEST(time, clock_getres_unknown) {
errno = 0;
timespec ts = { -1, -1 };
ASSERT_EQ(-1, clock_getres(-1, &ts));
ASSERT_ERRNO(EINVAL);
ASSERT_EQ(-1, ts.tv_nsec);
ASSERT_EQ(-1, ts.tv_sec);
}
TEST(time, clock_getres_null_resolution) {
ASSERT_EQ(0, clock_getres(CLOCK_REALTIME, nullptr));
}
TEST(time, clock) {
// clock(3) is hard to test, but a 1s sleep should cost less than 10ms on average.
static const clock_t N = 5;
static const clock_t mean_limit_ms = 10;
clock_t t0 = clock();
for (size_t i = 0; i < N; ++i) {
sleep(1);
}
clock_t t1 = clock();
ASSERT_LT(t1 - t0, N * mean_limit_ms * (CLOCKS_PER_SEC / 1000));
}
static pid_t GetInvalidPid() {
std::unique_ptr<FILE, decltype(&fclose)> fp{fopen("/proc/sys/kernel/pid_max", "r"), fclose};
long pid_max;
fscanf(fp.get(), "%ld", &pid_max);
return static_cast<pid_t>(pid_max + 1);
}
TEST(time, clock_getcpuclockid_current) {
clockid_t clockid;
ASSERT_EQ(0, clock_getcpuclockid(getpid(), &clockid));
timespec ts;
ASSERT_EQ(0, clock_gettime(clockid, &ts));
}
TEST(time, clock_getcpuclockid_parent) {
clockid_t clockid;
ASSERT_EQ(0, clock_getcpuclockid(getppid(), &clockid));
timespec ts;
ASSERT_EQ(0, clock_gettime(clockid, &ts));
}
TEST(time, clock_getcpuclockid_ESRCH) {
// We can't use -1 for invalid pid here, because clock_getcpuclockid() can't detect it.
errno = 0;
// If this fails, your kernel needs commit e1b6b6ce to be backported.
clockid_t clockid;
ASSERT_EQ(ESRCH, clock_getcpuclockid(GetInvalidPid(), &clockid)) << "\n"
<< "Please ensure that the following kernel patches or their replacements have been applied:\n"
<< "* https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/"
<< "commit/?id=e1b6b6ce55a0a25c8aa8af019095253b2133a41a\n"
<< "* https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/"
<< "commit/?id=c80ed088a519da53f27b798a69748eaabc66aadf\n";
ASSERT_ERRNO(0);
}
TEST(time, clock_settime) {
errno = 0;
timespec ts;
ASSERT_EQ(-1, clock_settime(-1, &ts));
ASSERT_ERRNO(EINVAL);
}
TEST(time, clock_nanosleep_EINVAL) {
timespec in;
timespec out;
ASSERT_EQ(EINVAL, clock_nanosleep(-1, 0, &in, &out));
}
TEST(time, clock_nanosleep_thread_cputime_id) {
timespec in;
in.tv_sec = 1;
in.tv_nsec = 0;
ASSERT_EQ(EINVAL, clock_nanosleep(CLOCK_THREAD_CPUTIME_ID, 0, &in, nullptr));
}
TEST(time, clock_nanosleep) {
auto t0 = std::chrono::steady_clock::now();
const timespec ts = {.tv_nsec = 5000000};
ASSERT_EQ(0, clock_nanosleep(CLOCK_MONOTONIC, 0, &ts, nullptr));
auto t1 = std::chrono::steady_clock::now();
ASSERT_GE(t1-t0, 5000000ns);
}
TEST(time, nanosleep) {
auto t0 = std::chrono::steady_clock::now();
const timespec ts = {.tv_nsec = 5000000};
ASSERT_EQ(0, nanosleep(&ts, nullptr));
auto t1 = std::chrono::steady_clock::now();
ASSERT_GE(t1-t0, 5000000ns);
}
TEST(time, nanosleep_EINVAL) {
timespec ts = {.tv_sec = -1};
errno = 0;
ASSERT_EQ(-1, nanosleep(&ts, nullptr));
ASSERT_ERRNO(EINVAL);
}
TEST(time, bug_31938693) {
// User-visible symptoms in N:
// http://b/31938693
// https://code.google.com/p/android/issues/detail?id=225132
// Actual underlying bug (the code change, not the tzdata upgrade that first exposed the bug):
// http://b/31848040
// This isn't a great test, because very few timezones were actually affected, and there's
// no real logic to which ones were affected: it was just a coincidence of the data that came
// after them in the tzdata file.
time_t t = 1475619727;
struct tm tm;
setenv("TZ", "America/Los_Angeles", 1);
tzset();
ASSERT_TRUE(localtime_r(&t, &tm) != nullptr);
EXPECT_EQ(15, tm.tm_hour);
setenv("TZ", "Europe/London", 1);
tzset();
ASSERT_TRUE(localtime_r(&t, &tm) != nullptr);
EXPECT_EQ(23, tm.tm_hour);
setenv("TZ", "America/Atka", 1);
tzset();
ASSERT_TRUE(localtime_r(&t, &tm) != nullptr);
EXPECT_EQ(13, tm.tm_hour);
setenv("TZ", "Pacific/Apia", 1);
tzset();
ASSERT_TRUE(localtime_r(&t, &tm) != nullptr);
EXPECT_EQ(12, tm.tm_hour);
setenv("TZ", "Pacific/Honolulu", 1);
tzset();
ASSERT_TRUE(localtime_r(&t, &tm) != nullptr);
EXPECT_EQ(12, tm.tm_hour);
setenv("TZ", "Asia/Magadan", 1);
tzset();
ASSERT_TRUE(localtime_r(&t, &tm) != nullptr);
EXPECT_EQ(9, tm.tm_hour);
}
TEST(time, bug_31339449) {
// POSIX says localtime acts as if it calls tzset.
// tzset does two things:
// 1. it sets the timezone ctime/localtime/mktime/strftime will use.
// 2. it sets the global `tzname`.
// POSIX says localtime_r need not set `tzname` (2).
// Q: should localtime_r set the timezone (1)?
// Upstream tzcode (and glibc) answer "no", everyone else answers "yes".
// Pick a time, any time...
time_t t = 1475619727;
// Call tzset with a specific timezone.
setenv("TZ", "America/Atka", 1);
tzset();
// If we change the timezone and call localtime, localtime should use the new timezone.
setenv("TZ", "America/Los_Angeles", 1);
struct tm* tm_p = localtime(&t);
EXPECT_EQ(15, tm_p->tm_hour);
// Reset the timezone back.
setenv("TZ", "America/Atka", 1);
tzset();
#if defined(__BIONIC__)
// If we change the timezone again and call localtime_r, localtime_r should use the new timezone.
setenv("TZ", "America/Los_Angeles", 1);
struct tm tm = {};
localtime_r(&t, &tm);
EXPECT_EQ(15, tm.tm_hour);
#else
// The BSDs agree with us, but glibc gets this wrong.
#endif
}
TEST(time, asctime) {
const struct tm tm = {};
ASSERT_STREQ("Sun Jan 0 00:00:00 1900\n", asctime(&tm));
}
TEST(time, asctime_r) {
const struct tm tm = {};
char buf[256];
ASSERT_EQ(buf, asctime_r(&tm, buf));
ASSERT_STREQ("Sun Jan 0 00:00:00 1900\n", buf);
}
TEST(time, ctime) {
setenv("TZ", "UTC", 1);
const time_t t = 0;
ASSERT_STREQ("Thu Jan 1 00:00:00 1970\n", ctime(&t));
}
TEST(time, ctime_r) {
setenv("TZ", "UTC", 1);
const time_t t = 0;
char buf[256];
ASSERT_EQ(buf, ctime_r(&t, buf));
ASSERT_STREQ("Thu Jan 1 00:00:00 1970\n", buf);
}
// https://issuetracker.google.com/37128336
TEST(time, strftime_strptime_s) {
char buf[32];
const struct tm tm0 = { .tm_year = 1982-1900, .tm_mon = 0, .tm_mday = 1 };
setenv("TZ", "America/Los_Angeles", 1);
strftime(buf, sizeof(buf), "<%s>", &tm0);
EXPECT_STREQ("<378720000>", buf);
setenv("TZ", "UTC", 1);
strftime(buf, sizeof(buf), "<%s>", &tm0);
EXPECT_STREQ("<378691200>", buf);
struct tm tm;
setenv("TZ", "America/Los_Angeles", 1);
tzset();
memset(&tm, 0xff, sizeof(tm));
char* p = strptime("378720000x", "%s", &tm);
ASSERT_EQ('x', *p);
EXPECT_EQ(0, tm.tm_sec);
EXPECT_EQ(0, tm.tm_min);
EXPECT_EQ(0, tm.tm_hour);
EXPECT_EQ(1, tm.tm_mday);
EXPECT_EQ(0, tm.tm_mon);
EXPECT_EQ(82, tm.tm_year);
EXPECT_EQ(5, tm.tm_wday);
EXPECT_EQ(0, tm.tm_yday);
EXPECT_EQ(0, tm.tm_isdst);
setenv("TZ", "UTC", 1);
tzset();
memset(&tm, 0xff, sizeof(tm));
p = strptime("378691200x", "%s", &tm);
ASSERT_EQ('x', *p);
EXPECT_EQ(0, tm.tm_sec);
EXPECT_EQ(0, tm.tm_min);
EXPECT_EQ(0, tm.tm_hour);
EXPECT_EQ(1, tm.tm_mday);
EXPECT_EQ(0, tm.tm_mon);
EXPECT_EQ(82, tm.tm_year);
EXPECT_EQ(5, tm.tm_wday);
EXPECT_EQ(0, tm.tm_yday);
EXPECT_EQ(0, tm.tm_isdst);
}
TEST(time, strptime_s_nothing) {
struct tm tm;
ASSERT_EQ(nullptr, strptime("x", "%s", &tm));
}
TEST(time, timespec_get) {
#if defined(__BIONIC__)
timespec ts = {};
ASSERT_EQ(TIME_UTC, timespec_get(&ts, TIME_UTC));
ASSERT_EQ(TIME_MONOTONIC, timespec_get(&ts, TIME_MONOTONIC));
ASSERT_EQ(TIME_ACTIVE, timespec_get(&ts, TIME_ACTIVE));
ASSERT_EQ(TIME_THREAD_ACTIVE, timespec_get(&ts, TIME_THREAD_ACTIVE));
#else
GTEST_SKIP() << "glibc doesn't have timespec_get until 2.21";
#endif
}
TEST(time, timespec_get_invalid) {
#if defined(__BIONIC__)
timespec ts = {};
ASSERT_EQ(0, timespec_get(&ts, 123));
#else
GTEST_SKIP() << "glibc doesn't have timespec_get until 2.21";
#endif
}
TEST(time, timespec_getres) {
#if defined(__BIONIC__)
timespec ts = {};
ASSERT_EQ(TIME_UTC, timespec_getres(&ts, TIME_UTC));
ASSERT_EQ(1, ts.tv_nsec);
ASSERT_EQ(0, ts.tv_sec);
#else
GTEST_SKIP() << "glibc doesn't have timespec_get until 2.21";
#endif
}
TEST(time, timespec_getres_invalid) {
#if defined(__BIONIC__)
timespec ts = {};
ASSERT_EQ(0, timespec_getres(&ts, 123));
#else
GTEST_SKIP() << "glibc doesn't have timespec_get until 2.21";
#endif
}
TEST(time, difftime) {
ASSERT_EQ(1.0, difftime(1, 0));
ASSERT_EQ(-1.0, difftime(0, 1));
}
Expose tzalloc()/localtime_rz()/mktime_z()/tzfree(). * Rationale The question often comes up of how to use multiple time zones in C code. If you're single-threaded, you can just use setenv() to manipulate $TZ. toybox does this, for example. But that's not thread-safe in two distinct ways: firstly, getenv() is not thread-safe with respect to modifications to the environment (and between the way putenv() is specified and the existence of environ, it's not obvious how to fully fix that), and secondly the _caller_ needs to ensure that no other threads are using tzset() or any function that behaves "as if" tzset() was called (which is neither easy to determine nor easy to ensure). This isn't a bigger problem because most of the time the right answer is to stop pretending that libc is at all suitable for any i18n, and switch to icu4c instead. (The NDK icu4c headers do not include ucal_*, so this is not a realistic option for most applications.) But what if you're somewhere in between? Like the rust chrono library, for example? What then? Currently their "least worst" option is to reinvent the entire wheel and read our tzdata files. Which isn't a great solution for anyone, for obvious maintainability reasons. So it's probably time we broke the catch-22 here and joined NetBSD in offering a less broken API than standard C has for the last 40 years. Sure, any would-be caller will have to have a separate "is this Android?" and even "is this API level >= 35?" path, but that will fix itself sometime in the 2030s when developers can just assume "yes, it is", whereas if we keep putting off exposing anything, this problem never gets solved. (No-one's bothered to try to implement the std::chrono::time_zone functionality in libc++ yet, but they'll face a similar problem if/when they do.) * Implementation The good news is that tzcode already implements these functions, so there's relatively little here. I've chosen not to expose `struct state` because `struct __timezone_t` makes for clearer error messages, given that compiler diagnostics will show the underlying type name (`struct __timezone_t*`) rather than the typedef name (`timezone_t`) that's used in calling code. I've moved us over to FreeBSD's wcsftime() rather than keep the OpenBSD one building --- I've long wanted to only have one implementation here, and FreeBSD is already doing the "convert back and forth, calling the non-wide function in the middle" dance that I'd hoped to get round to doing myself someday. This should mean that our strftime() and wcsftime() behaviors can't easily diverge in future, plus macOS/iOS are mostly FreeBSD, so any bugs will likely be interoperable with the other major mobile operating system, so there's something nice for everyone there! The FreeBSD wcsftime() implementation includes a wcsftime_l() implementation, so that's one stub we can remove. The flip side of that is that it uses mbsrtowcs_l() and wcsrtombs_l() which we didn't previously have. So expose those as aliases of mbsrtowcs() and wcsrtombs(). Bug: https://github.com/chronotope/chrono/issues/499 Test: treehugger Change-Id: Iee1b9d763ead15eef3d2c33666b3403b68940c3c
2023-06-15 22:17:08 +02:00
TEST(time, tzfree_null) {
#if defined(__BIONIC__)
Expose tzalloc()/localtime_rz()/mktime_z()/tzfree(). * Rationale The question often comes up of how to use multiple time zones in C code. If you're single-threaded, you can just use setenv() to manipulate $TZ. toybox does this, for example. But that's not thread-safe in two distinct ways: firstly, getenv() is not thread-safe with respect to modifications to the environment (and between the way putenv() is specified and the existence of environ, it's not obvious how to fully fix that), and secondly the _caller_ needs to ensure that no other threads are using tzset() or any function that behaves "as if" tzset() was called (which is neither easy to determine nor easy to ensure). This isn't a bigger problem because most of the time the right answer is to stop pretending that libc is at all suitable for any i18n, and switch to icu4c instead. (The NDK icu4c headers do not include ucal_*, so this is not a realistic option for most applications.) But what if you're somewhere in between? Like the rust chrono library, for example? What then? Currently their "least worst" option is to reinvent the entire wheel and read our tzdata files. Which isn't a great solution for anyone, for obvious maintainability reasons. So it's probably time we broke the catch-22 here and joined NetBSD in offering a less broken API than standard C has for the last 40 years. Sure, any would-be caller will have to have a separate "is this Android?" and even "is this API level >= 35?" path, but that will fix itself sometime in the 2030s when developers can just assume "yes, it is", whereas if we keep putting off exposing anything, this problem never gets solved. (No-one's bothered to try to implement the std::chrono::time_zone functionality in libc++ yet, but they'll face a similar problem if/when they do.) * Implementation The good news is that tzcode already implements these functions, so there's relatively little here. I've chosen not to expose `struct state` because `struct __timezone_t` makes for clearer error messages, given that compiler diagnostics will show the underlying type name (`struct __timezone_t*`) rather than the typedef name (`timezone_t`) that's used in calling code. I've moved us over to FreeBSD's wcsftime() rather than keep the OpenBSD one building --- I've long wanted to only have one implementation here, and FreeBSD is already doing the "convert back and forth, calling the non-wide function in the middle" dance that I'd hoped to get round to doing myself someday. This should mean that our strftime() and wcsftime() behaviors can't easily diverge in future, plus macOS/iOS are mostly FreeBSD, so any bugs will likely be interoperable with the other major mobile operating system, so there's something nice for everyone there! The FreeBSD wcsftime() implementation includes a wcsftime_l() implementation, so that's one stub we can remove. The flip side of that is that it uses mbsrtowcs_l() and wcsrtombs_l() which we didn't previously have. So expose those as aliases of mbsrtowcs() and wcsrtombs(). Bug: https://github.com/chronotope/chrono/issues/499 Test: treehugger Change-Id: Iee1b9d763ead15eef3d2c33666b3403b68940c3c
2023-06-15 22:17:08 +02:00
tzfree(nullptr);
#else
GTEST_SKIP() << "glibc doesn't have timezone_t";
#endif
}
TEST(time, localtime_rz) {
#if defined(__BIONIC__)
Expose tzalloc()/localtime_rz()/mktime_z()/tzfree(). * Rationale The question often comes up of how to use multiple time zones in C code. If you're single-threaded, you can just use setenv() to manipulate $TZ. toybox does this, for example. But that's not thread-safe in two distinct ways: firstly, getenv() is not thread-safe with respect to modifications to the environment (and between the way putenv() is specified and the existence of environ, it's not obvious how to fully fix that), and secondly the _caller_ needs to ensure that no other threads are using tzset() or any function that behaves "as if" tzset() was called (which is neither easy to determine nor easy to ensure). This isn't a bigger problem because most of the time the right answer is to stop pretending that libc is at all suitable for any i18n, and switch to icu4c instead. (The NDK icu4c headers do not include ucal_*, so this is not a realistic option for most applications.) But what if you're somewhere in between? Like the rust chrono library, for example? What then? Currently their "least worst" option is to reinvent the entire wheel and read our tzdata files. Which isn't a great solution for anyone, for obvious maintainability reasons. So it's probably time we broke the catch-22 here and joined NetBSD in offering a less broken API than standard C has for the last 40 years. Sure, any would-be caller will have to have a separate "is this Android?" and even "is this API level >= 35?" path, but that will fix itself sometime in the 2030s when developers can just assume "yes, it is", whereas if we keep putting off exposing anything, this problem never gets solved. (No-one's bothered to try to implement the std::chrono::time_zone functionality in libc++ yet, but they'll face a similar problem if/when they do.) * Implementation The good news is that tzcode already implements these functions, so there's relatively little here. I've chosen not to expose `struct state` because `struct __timezone_t` makes for clearer error messages, given that compiler diagnostics will show the underlying type name (`struct __timezone_t*`) rather than the typedef name (`timezone_t`) that's used in calling code. I've moved us over to FreeBSD's wcsftime() rather than keep the OpenBSD one building --- I've long wanted to only have one implementation here, and FreeBSD is already doing the "convert back and forth, calling the non-wide function in the middle" dance that I'd hoped to get round to doing myself someday. This should mean that our strftime() and wcsftime() behaviors can't easily diverge in future, plus macOS/iOS are mostly FreeBSD, so any bugs will likely be interoperable with the other major mobile operating system, so there's something nice for everyone there! The FreeBSD wcsftime() implementation includes a wcsftime_l() implementation, so that's one stub we can remove. The flip side of that is that it uses mbsrtowcs_l() and wcsrtombs_l() which we didn't previously have. So expose those as aliases of mbsrtowcs() and wcsrtombs(). Bug: https://github.com/chronotope/chrono/issues/499 Test: treehugger Change-Id: Iee1b9d763ead15eef3d2c33666b3403b68940c3c
2023-06-15 22:17:08 +02:00
setenv("TZ", "America/Los_Angeles", 1);
tzset();
auto AssertTmEq = [](const struct tm& rhs, int hour) {
ASSERT_EQ(93, rhs.tm_year);
ASSERT_EQ(0, rhs.tm_mon);
ASSERT_EQ(1, rhs.tm_mday);
ASSERT_EQ(hour, rhs.tm_hour);
ASSERT_EQ(0, rhs.tm_min);
ASSERT_EQ(0, rhs.tm_sec);
};
const time_t t = 725875200;
// Spam localtime_r() while we use localtime_rz().
std::atomic<bool> done = false;
std::thread thread{[&] {
while (!done) {
struct tm tm {};
ASSERT_EQ(&tm, localtime_r(&t, &tm));
AssertTmEq(tm, 0);
}
}};
struct tm tm;
timezone_t london{tzalloc("Europe/London")};
tm = {};
ASSERT_EQ(&tm, localtime_rz(london, &t, &tm));
AssertTmEq(tm, 8);
timezone_t seoul{tzalloc("Asia/Seoul")};
tm = {};
ASSERT_EQ(&tm, localtime_rz(seoul, &t, &tm));
AssertTmEq(tm, 17);
// Just check that mktime()'s timezone didn't change.
Expose tzalloc()/localtime_rz()/mktime_z()/tzfree(). * Rationale The question often comes up of how to use multiple time zones in C code. If you're single-threaded, you can just use setenv() to manipulate $TZ. toybox does this, for example. But that's not thread-safe in two distinct ways: firstly, getenv() is not thread-safe with respect to modifications to the environment (and between the way putenv() is specified and the existence of environ, it's not obvious how to fully fix that), and secondly the _caller_ needs to ensure that no other threads are using tzset() or any function that behaves "as if" tzset() was called (which is neither easy to determine nor easy to ensure). This isn't a bigger problem because most of the time the right answer is to stop pretending that libc is at all suitable for any i18n, and switch to icu4c instead. (The NDK icu4c headers do not include ucal_*, so this is not a realistic option for most applications.) But what if you're somewhere in between? Like the rust chrono library, for example? What then? Currently their "least worst" option is to reinvent the entire wheel and read our tzdata files. Which isn't a great solution for anyone, for obvious maintainability reasons. So it's probably time we broke the catch-22 here and joined NetBSD in offering a less broken API than standard C has for the last 40 years. Sure, any would-be caller will have to have a separate "is this Android?" and even "is this API level >= 35?" path, but that will fix itself sometime in the 2030s when developers can just assume "yes, it is", whereas if we keep putting off exposing anything, this problem never gets solved. (No-one's bothered to try to implement the std::chrono::time_zone functionality in libc++ yet, but they'll face a similar problem if/when they do.) * Implementation The good news is that tzcode already implements these functions, so there's relatively little here. I've chosen not to expose `struct state` because `struct __timezone_t` makes for clearer error messages, given that compiler diagnostics will show the underlying type name (`struct __timezone_t*`) rather than the typedef name (`timezone_t`) that's used in calling code. I've moved us over to FreeBSD's wcsftime() rather than keep the OpenBSD one building --- I've long wanted to only have one implementation here, and FreeBSD is already doing the "convert back and forth, calling the non-wide function in the middle" dance that I'd hoped to get round to doing myself someday. This should mean that our strftime() and wcsftime() behaviors can't easily diverge in future, plus macOS/iOS are mostly FreeBSD, so any bugs will likely be interoperable with the other major mobile operating system, so there's something nice for everyone there! The FreeBSD wcsftime() implementation includes a wcsftime_l() implementation, so that's one stub we can remove. The flip side of that is that it uses mbsrtowcs_l() and wcsrtombs_l() which we didn't previously have. So expose those as aliases of mbsrtowcs() and wcsrtombs(). Bug: https://github.com/chronotope/chrono/issues/499 Test: treehugger Change-Id: Iee1b9d763ead15eef3d2c33666b3403b68940c3c
2023-06-15 22:17:08 +02:00
tm = {};
ASSERT_EQ(&tm, localtime_r(&t, &tm));
ASSERT_EQ(0, tm.tm_hour);
AssertTmEq(tm, 0);
done = true;
thread.join();
tzfree(london);
tzfree(seoul);
#else
GTEST_SKIP() << "glibc doesn't have timezone_t";
#endif
}
TEST(time, mktime_z) {
#if defined(__BIONIC__)
Expose tzalloc()/localtime_rz()/mktime_z()/tzfree(). * Rationale The question often comes up of how to use multiple time zones in C code. If you're single-threaded, you can just use setenv() to manipulate $TZ. toybox does this, for example. But that's not thread-safe in two distinct ways: firstly, getenv() is not thread-safe with respect to modifications to the environment (and between the way putenv() is specified and the existence of environ, it's not obvious how to fully fix that), and secondly the _caller_ needs to ensure that no other threads are using tzset() or any function that behaves "as if" tzset() was called (which is neither easy to determine nor easy to ensure). This isn't a bigger problem because most of the time the right answer is to stop pretending that libc is at all suitable for any i18n, and switch to icu4c instead. (The NDK icu4c headers do not include ucal_*, so this is not a realistic option for most applications.) But what if you're somewhere in between? Like the rust chrono library, for example? What then? Currently their "least worst" option is to reinvent the entire wheel and read our tzdata files. Which isn't a great solution for anyone, for obvious maintainability reasons. So it's probably time we broke the catch-22 here and joined NetBSD in offering a less broken API than standard C has for the last 40 years. Sure, any would-be caller will have to have a separate "is this Android?" and even "is this API level >= 35?" path, but that will fix itself sometime in the 2030s when developers can just assume "yes, it is", whereas if we keep putting off exposing anything, this problem never gets solved. (No-one's bothered to try to implement the std::chrono::time_zone functionality in libc++ yet, but they'll face a similar problem if/when they do.) * Implementation The good news is that tzcode already implements these functions, so there's relatively little here. I've chosen not to expose `struct state` because `struct __timezone_t` makes for clearer error messages, given that compiler diagnostics will show the underlying type name (`struct __timezone_t*`) rather than the typedef name (`timezone_t`) that's used in calling code. I've moved us over to FreeBSD's wcsftime() rather than keep the OpenBSD one building --- I've long wanted to only have one implementation here, and FreeBSD is already doing the "convert back and forth, calling the non-wide function in the middle" dance that I'd hoped to get round to doing myself someday. This should mean that our strftime() and wcsftime() behaviors can't easily diverge in future, plus macOS/iOS are mostly FreeBSD, so any bugs will likely be interoperable with the other major mobile operating system, so there's something nice for everyone there! The FreeBSD wcsftime() implementation includes a wcsftime_l() implementation, so that's one stub we can remove. The flip side of that is that it uses mbsrtowcs_l() and wcsrtombs_l() which we didn't previously have. So expose those as aliases of mbsrtowcs() and wcsrtombs(). Bug: https://github.com/chronotope/chrono/issues/499 Test: treehugger Change-Id: Iee1b9d763ead15eef3d2c33666b3403b68940c3c
2023-06-15 22:17:08 +02:00
setenv("TZ", "America/Los_Angeles", 1);
tzset();
// Spam mktime() while we use mktime_z().
std::atomic<bool> done = false;
std::thread thread{[&done] {
while (!done) {
struct tm tm {
.tm_year = 93, .tm_mday = 1
};
ASSERT_EQ(725875200, mktime(&tm));
}
}};
struct tm tm;
timezone_t london{tzalloc("Europe/London")};
tm = {.tm_year = 93, .tm_mday = 1};
ASSERT_EQ(725846400, mktime_z(london, &tm));
timezone_t seoul{tzalloc("Asia/Seoul")};
tm = {.tm_year = 93, .tm_mday = 1};
ASSERT_EQ(725814000, mktime_z(seoul, &tm));
// Just check that mktime()'s timezone didn't change.
Expose tzalloc()/localtime_rz()/mktime_z()/tzfree(). * Rationale The question often comes up of how to use multiple time zones in C code. If you're single-threaded, you can just use setenv() to manipulate $TZ. toybox does this, for example. But that's not thread-safe in two distinct ways: firstly, getenv() is not thread-safe with respect to modifications to the environment (and between the way putenv() is specified and the existence of environ, it's not obvious how to fully fix that), and secondly the _caller_ needs to ensure that no other threads are using tzset() or any function that behaves "as if" tzset() was called (which is neither easy to determine nor easy to ensure). This isn't a bigger problem because most of the time the right answer is to stop pretending that libc is at all suitable for any i18n, and switch to icu4c instead. (The NDK icu4c headers do not include ucal_*, so this is not a realistic option for most applications.) But what if you're somewhere in between? Like the rust chrono library, for example? What then? Currently their "least worst" option is to reinvent the entire wheel and read our tzdata files. Which isn't a great solution for anyone, for obvious maintainability reasons. So it's probably time we broke the catch-22 here and joined NetBSD in offering a less broken API than standard C has for the last 40 years. Sure, any would-be caller will have to have a separate "is this Android?" and even "is this API level >= 35?" path, but that will fix itself sometime in the 2030s when developers can just assume "yes, it is", whereas if we keep putting off exposing anything, this problem never gets solved. (No-one's bothered to try to implement the std::chrono::time_zone functionality in libc++ yet, but they'll face a similar problem if/when they do.) * Implementation The good news is that tzcode already implements these functions, so there's relatively little here. I've chosen not to expose `struct state` because `struct __timezone_t` makes for clearer error messages, given that compiler diagnostics will show the underlying type name (`struct __timezone_t*`) rather than the typedef name (`timezone_t`) that's used in calling code. I've moved us over to FreeBSD's wcsftime() rather than keep the OpenBSD one building --- I've long wanted to only have one implementation here, and FreeBSD is already doing the "convert back and forth, calling the non-wide function in the middle" dance that I'd hoped to get round to doing myself someday. This should mean that our strftime() and wcsftime() behaviors can't easily diverge in future, plus macOS/iOS are mostly FreeBSD, so any bugs will likely be interoperable with the other major mobile operating system, so there's something nice for everyone there! The FreeBSD wcsftime() implementation includes a wcsftime_l() implementation, so that's one stub we can remove. The flip side of that is that it uses mbsrtowcs_l() and wcsrtombs_l() which we didn't previously have. So expose those as aliases of mbsrtowcs() and wcsrtombs(). Bug: https://github.com/chronotope/chrono/issues/499 Test: treehugger Change-Id: Iee1b9d763ead15eef3d2c33666b3403b68940c3c
2023-06-15 22:17:08 +02:00
tm = {.tm_year = 93, .tm_mday = 1};
ASSERT_EQ(725875200, mktime(&tm));
done = true;
thread.join();
tzfree(london);
tzfree(seoul);
#else
GTEST_SKIP() << "glibc doesn't have timezone_t";
#endif
}
TEST(time, tzalloc_nullptr) {
#if defined(__BIONIC__)
// tzalloc(nullptr) returns the system timezone.
timezone_t default_tz = tzalloc(nullptr);
ASSERT_NE(nullptr, default_tz);
// Check that mktime_z() with the default timezone matches mktime().
// This assumes that the system timezone doesn't change during the test,
// but that should be unlikely, and we don't have much choice if we
// want to write a test at all.
// We unset $TZ before calling mktime() because mktime() honors $TZ.
unsetenv("TZ");
struct tm tm = {.tm_year = 93, .tm_mday = 1};
time_t t = mktime(&tm);
ASSERT_EQ(t, mktime_z(default_tz, &tm));
// Check that changing $TZ doesn't affect the tzalloc() default in
// the same way it would the mktime() default.
setenv("TZ", "America/Los_Angeles", 1);
tzset();
ASSERT_EQ(t, mktime_z(default_tz, &tm));
setenv("TZ", "Europe/London", 1);
tzset();
ASSERT_EQ(t, mktime_z(default_tz, &tm));
setenv("TZ", "Asia/Seoul", 1);
tzset();
ASSERT_EQ(t, mktime_z(default_tz, &tm));
tzfree(default_tz);
#else
GTEST_SKIP() << "glibc doesn't have timezone_t";
#endif
}
TEST(time, tzalloc_unique_ptr) {
#if defined(__BIONIC__)
std::unique_ptr<std::remove_pointer_t<timezone_t>, decltype(&tzfree)> tz{tzalloc("Asia/Seoul"),
tzfree};
#else
GTEST_SKIP() << "glibc doesn't have timezone_t";
#endif
}