Year 2038 problem

The latest time since 1 January 1970 that can be stored using a signed 32-bit integer is 03:14:07 on Tuesday, 19 January 2038 (2³¹-1 = 2,147,483,647 seconds after 1 January 1970).[1]

Programs that attempt to increment the time beyond this date will cause the value to be stored internally as a very large negative number, which these systems will interpret as having occurred at 20:45:52 on Friday, 13 December 1901 rather than 19 January 2038. This is caused by integer overflow, during which the counter runs out of usable digit bits, and flips the sign bit instead. This reports a maximally negative number, and continues to count up, towards zero, and then up through the positive integers again. Resulting erroneous calculations on such systems are likely to cause problems for users and other reliant parties.

Programs that work with future dates will begin to run into problems sooner; for example, a program that works with dates 10 years in the future will need to be fixed no later than 19 January 2028.

In May 2006, reports surfaced of an early manifestation of the Y2038 problem in the AOLserver software. The software was designed with a kludge to handle a database request that should "never" time out. Rather than specifically handling this special case, the initial design simply specified an arbitrary time-out date in the future. The default configuration for the server specified that the request should time out after one billion seconds. One billion seconds (approximately 32 years) after 01:27:28 UTC on 13 May 2006 is beyond the 2038 cutoff date. Thus, after this time, the time-out calculation overflowed and returned a date that was actually in the past, causing the software to crash. When the problem was discovered, AOLServer operators had to edit the configuration file and set the time-out to a lower value.[2][3]

Players of games or apps which are programmed to impose waiting periods[4] are running into this problem when they attempt to work around the waiting period on devices which harbor the coding, by manually setting their devices to a date past 19 January 2038, but are unable to do so, since a 32-bit Unix time format is being used.

Embedded systems that use dates for either computation or diagnostic logging are most likely to be affected by the 2038 problem.[5]

Many transportation systems from flight to automobiles use embedded systems extensively. In automotive systems, this may include anti-lock braking system (ABS), electronic stability control (ESC/ESP), traction control (TCS) and automatic four-wheel drive; aircraft may use inertial guidance systems and GPS receivers.[note 1] However, this does not imply that all these systems will suffer from the Y2038 problem, since many such systems do not require access to dates. For those that do, those systems which only track the difference between times/dates and not absolute times/dates will, by the nature of the calculation, not experience a major problem. This is the case for automotive diagnostics based on legislative standards such as CARB (California Air Resources Board).[6]

Another major use of embedded systems is in communications devices, including cell phones and Internet appliances (routers, wireless access points, etc.) which rely on storing an accurate time and date and are increasingly based on UNIX-like operating systems. For example, the Y2038 problem makes some devices running 32-bit Android crash and not restart when the time is changed to that date.[7]

Despite the modern 18–24 month generational update in computer systems technology, embedded systems are designed to last the lifetime of the machine in which they are a component. It is conceivable that some of these systems may still be in use in 2038. It may be impractical or, in some cases, impossible to upgrade the software running these systems, ultimately requiring replacement if 32-bit time_t limitations are to be corrected.

MySQL database's built-in functions like UNIX_TIMESTAMP() will return 0 after 03:14:07 UTC on 19 January 2038.[8]

Early Mac OS X versions[9] are susceptible to the Year 2038 problem.

Many data structures in use today have 32-bit time representations embedded into their structure. A full list of these data structures is virtually impossible to derive but there are well-known data structures that have the Unix time problem:

• file systems (many file systems use only 32 bits to represent times in inodes)
• binary file formats (that use 32-bit time fields)
• databases (that have 32-bit time fields)
• database query languages, like SQL that have UNIX_TIMESTAMP()-like commands

Examples of systems using data structures that may contain 32-bit time representations include:

• embedded factory, refinery control and monitoring subsystems
• assorted medical devices
• assorted military devices

Any system making use of data structures containing 32-bit time representations will present risk. The degree of risk is dependent on the mode of failure.

The Network Time Protocol (NTP) has a related overflow issue, which manifests itself in 2036, rather than 2038. The 64-bit timestamps used by NTP consist of a 32-bit part for seconds and a 32-bit part for fractional second, giving NTP a time scale that rolls over every 2³² seconds (136 years) and a theoretical resolution of 2⁻³² seconds (233 picoseconds). NTP uses an epoch of 1 January 1900. The first rollover occurs in 2036, prior to the UNIX year 2038 problem.

Implementations should disambiguate NTP time using a knowledge of the approximate time from other sources. Since NTP only works with the differences between timestamps and never their absolute values, the wraparound is invisible in the calculations as long as the timestamps are within 68 years of each other. However, after a wraparound the clients can still face two problems:

1. They receive the date 1900-01-01 00:00:00UTC, not 2036-02-07 06:28:15 (plus or minus some leap seconds) as the new time.
2. When a client tries to adopt this time and store it in UNIX time format, as many embedded systems do, it will fail because UNIX time starts at 13 December 1901 (signed 32 bit integer) or 1 January 1970 (unsigned 32 bit integer).

This means that for NTP the rollover will be invisible for most running systems, since they will have the correct time to within a very small tolerance. However, systems that are starting up need to know the date within no more than 68 years. Given the large allowed error, it is not expected that this is too onerous a requirement. One suggested method is to set the clock to no earlier than the system build date or the release date of the current version of the NTP software. Many systems use a battery-powered hardware clock to avoid this problem.

Even so, future versions of NTP may extend the time representation to 128 bits: 64 bits for the second and 64 bits for the fractional-second. The current NTP4 format has support for Era Number and Era Offset, that when used properly should aid fixing date rollover issues. According to Mills, "The 64 bit value for the fraction is enough to resolve the amount of time it takes a photon to pass an electron at the speed of light. The 64 bit second value is enough to provide unambiguous time representation until the universe goes dim."[10][note 2]

There is no universal solution for the Year 2038 problem. Any change to the definition of the time_t data type would result in code compatibility problems in any application in which date and time representations are dependent on the nature of the signed 32-bit time_t integer. For example, changing time_t to an unsigned 32-bit integer, which would extend the range to 2106 (specifically, 06:28:15 UTC on Sunday, 7 February 2106), would adversely affect programs that store, retrieve, or manipulate dates prior to 1970, as such dates are represented by negative numbers. Increasing the size of the time_t type to 64-bit in an existing system would cause incompatible changes to the layout of structures and the binary interface of functions.

Most operating systems designed to run on 64-bit hardware already use signed 64-bit time_t integers. Using a signed 64-bit value introduces a new wraparound date that is over twenty times greater than the estimated age of the universe: approximately 292 billion years from now, at 15:30:08 UTC on Sunday, 4 December 292,277,026,596. The ability to make computations on dates is limited by the fact that tm_year uses a signed 32 bit integer value starting at 1900 for the year. This limits the year to a maximum of 2,147,485,547 (2,147,483,647 + 1900).[11]

FreeBSD uses 64-bit time_t for all 32-bit and 64-bit architectures except 32-bit i386, which uses unsigned 32-bit time_t instead.[12]

Starting with NetBSD version 6.0 (released in October 2012), the NetBSD operating system uses a 64-bit time_t for both 32-bit and 64-bit architectures. Applications that were compiled for an older NetBSD release with 32-bit time_t are supported via a binary compatibility layer, but such older applications will still suffer from the Year 2038 problem.[13]

OpenBSD since version 5.5, released in May 2014, also uses a 64-bit time_t for both 32-bit and 64-bit architectures. In contrast to NetBSD, there is no binary compatibility layer. Therefore, applications expecting a 32-bit time_t and applications using anything different from time_t to store time values may break.[14]

Linux originally used a 64-bit time_t for 64-bit architectures only; the pure 32-bit ABI was not changed due to backward compatibility.[15] Starting with version 5.6, 64-bit time_t is supported on 32-bit architectures, too. This was done primarily for the sake of embedded Linux systems.[16]

The x32 ABI for Linux (which defines an environment for programs with 32-bit addresses but running the processor in 64-bit mode) uses a 64-bit time_t. Since it was a new environment, there was no need for special compatibility precautions.[15]

Network File System version 4 has defined its time fields as struct nfstime4 {int64_t seconds; uint32_t nseconds;} since December 2000.[17] Values greater than zero for the seconds field denote dates after the 0-hour, January 1, 1970. Values less than zero for the seconds field denote dates before the 0-hour, January 1, 1970. In both cases, the nseconds (nanoseconds) field is to be added to the seconds field for the final time representation.

Alternative proposals have been made (some of which are already in use), such as storing either milliseconds or microseconds since an epoch (typically either 1 January 1970 or 1 January 2000) in a signed 64-bit integer, providing a minimum range of 300,000 years at microsecond resolution.[18][19] In particular, Java's use of 64-bit long integers everywhere to represent time as "milliseconds since 1 January 1970" will work correctly for the next 292 million years. Other proposals for new time representations provide different precisions, ranges, and sizes (almost always wider than 32 bits), as well as solving other related problems, such as the handling of leap seconds. In particular, TAI64[20] is an implementation of the International Atomic Time (TAI) standard, the current international real-time standard for defining a second and frame of reference.

• Deep Impact is believed to have been lost at the time its internal clock reached 2³² one-tenth seconds since 2000, on 11 August 2013, 00:38:49 UTC.
• John Titor, alleged time traveler who is sometimes related to the problem
• Time formatting and storage bugs
• Unix time
• Year 10,000 problem
• Year 2000 problem

• Y2038 Proofness Design glibc Wiki
• Entry in How Stuff Works
• The Project 2038 Frequently Asked Questions
• Critical and Significant Dates 2038
• A 2038-safe replacement for time.h on 32 bit systems
• The number glitch that can lead to catastrophe
• Clewett, James. "2,147,483,647 – The End of Time [Unix]". Numberphile. Brady Haran. Archived from the original on 22 May 2017. Retrieved 7 April 2013.