Hide Forgot
Description of problem: date fails to parse dates bigger than INT_MAX (2147483647), 32-bit architecture. Version-Release number of selected component (if applicable): Always How reproducible: coreutils-8.4-31.el6.i686 date (GNU coreutils) 8.4 Steps to Reproduce: 1. date -d "2038-01-19 04:14:08" "+%s" Actual results: date: invalid date `2038-01-19 04:14:08' Expected results: 2147483648 Additional info: i:i386|m:i686 root@x86-64-v06 [~]# date -d "2038-01-19 04:14:07" "+%s" 2147483647 i:i386|m:i686 root@x86-64-v06 [~]# date -d "2038-01-19 04:14:08" "+%s" date: invalid date `2038-01-19 04:14:08'
The limitation here stems from the glibc and kernel representation of time_t which on 32 bit x86 at least is long, i.e. 31 bits. Now theoretically we could work around this by defining time64_t, mktime64(), localtime64() in gnulib, and parse_datetime() etc. would use those. Those functions in turn could call out to _time64() on windows, or otherwise on signed 32 bit time_t systems, handle at least the 32 bit wrap, thus handling dates until 2106. We could split the problem by only implementing mktime64() in gnulib, and thus supporting absolute times like in the above report without needing to worry about the kernel interface required for relative times. Though would it be worth doing that, when 64 bit supports this already? What's the use case for this exactly?
Greetings, thank you for your quick response at first. We use this datetime functions in certificate testing what is broken on 32-bit archs. Timestamp is basic commonly used data type and shouldn't be faulty. Thanks and Regards!