[eska]

This gives me flashbacks to the DOS filesystem timestamps. It's exactly the same mistake. By splitting the date into multiple fields, bits are wasted. If they hadn't tried to be smart and just made it one number, it would've been more precise with a wider range.

[Dylan16807]

I don't really see it as the same thing. Using multiple fields of bits works fine here. 60 possibilities for seconds and minutes fit very well into 6 bits each. A raw 24-bit integer gives you plus or minus 2330 hours, and if you don't reserve a bit the format they actually went with gives you plus or minus 2047 hours. Since it just needs to represent 999 hours and change, that's fine.

The problem is the weird non-bit packing they did before.

DOS is different, partly because there are fields that fit worse, and partly because 32 bits is just barely enough to store seconds in the first place. If you did a DOS-style packing with 34 or 38 bits it would work fine. And it would be able to represent leap seconds, making it arguably better than unix timestamps!

[falcolas]

FWIW, even keeping an interval in a single number still imposes limits. As proven by the epoch's 2038 problem.

[saturn_vk]

That's not really a problem though, at least from what i understand. even on 32bit linux systems, time_t is still 64bit. on windows it also seems to be 64bit

[nwallin]

The linux kernel has the option CONFIG_COMPAT_32BIT_TIME, which adds syscalls for 32 bit time_t. If this option is not enabled, much badness happens if you run 32 bit binaries, for instance, Steam and many 32 bit native linux games fail to run. (I ran into this problem a few weeks ago)

32 bit linux applications can use 64 bit time_t if they were configured to do so and glibc and the kernel are new enough. But enough closed source binaries don't support 64 bit time_t that we'll probably have problems in 2038.

https://cateee.net/lkddb/web-lkddb/COMPAT_32BIT_TIME.html

[cesarb]

> even on 32bit linux systems, time_t is still 64bit

No, on 32-bit Linux systems, time_t is 32 bits (except perhaps on newer ports like 32-bit RISC-V).

[WkndTriathlete]

https://stackoverflow.com/a/60709400/571787 states otherwise with recent Linux kernels + glibcs.

[cesarb]

Note that the glibc release mentioned there (2.32) has been released only two months ago (https://lwn.net/Articles/828210/), and its release notes don't mention time_t now being 64 bits. The glibc manual page linked to by that answer (https://www.gnu.org/software/libc/manual/html_node/64_002dbi...) says "at this point, 64-bit time support in dual-time configurations is work-in-progress, so for these configurations, the public API only makes the 32-bit time support available", which probably means in practice that 64-bit time_t is still not available for common 32-bit configurations (unless you want to lose binary compatibility with all existing software). I haven't been following this Y2038 work, but it's probably been postponed for a later glibc release (perhaps even the next one).

[mackal]

Which MySQL still isn't safe for :P (2038 that is, for where they use it)

[SoSoRoCoCo]

An oldie but a goodie:

https://people.cs.nctu.edu.tw/~tsaiwn/sisc/runtime_error_200...