Red Hat Enterprise Linux 7.5 is out [LWN.net]

Red Hat Enterprise Linux 7.5 is out

Posted Apr 11, 2018 10:01 UTC (Wed) by eru (subscriber, #2753)
In reply to: Red Hat Enterprise Linux 7.5 is out by Cyberax
Parent article: Red Hat Enterprise Linux 7.5 is out

It makes sense to exclude it from RHEL8 to avoid supporting it even further into 2030-s.

I wonder if RHEL8 will be the first RHEL to potentially encounter the Y2038 bug?

to post comments

Red Hat Enterprise Linux 7.5 is out

Posted Apr 11, 2018 12:53 UTC (Wed) by cesarb (subscriber, #6266) [Link] (3 responses)

> I wonder if RHEL8 will be the first RHEL to potentially encounter the Y2038 bug?

RHEL 7.5 is 64-bit only, so I believe RHEL 8 will also be 64-bit only, which should reduce a lot of the compatibility pain of fixing Y2038 problems (time_t is already 64 bits on 64-bit Linux). But unless they extend the support period again, RHEL 8 will probably already be out of support by then, so only calculations with dates in the future could encounter potential problems.

Red Hat Enterprise Linux 7.5 is out

Posted Apr 11, 2018 13:29 UTC (Wed) by pizza (subscriber, #46) [Link]

RHEL7.5 still has 32-bit userspace libraries (at least on x86-64), and I'd be shocked (but pleased) if RHEL8 does not.

Red Hat Enterprise Linux 7.5 is out

Posted Apr 11, 2018 14:30 UTC (Wed) by vstinner (subscriber, #42675) [Link] (1 responses)

Y2038 issue is wider than just the kernel. Many binary formats (like .gz and .pcap) encode time as seconds since Epoch in 32-bit.
https://en.wikipedia.org/wiki/Year_2038_problem#Data_stru...

TCP timestamp is 32-bit but doesn't require to use the Unix Epoch as the clock reference. Maybe other network protocols use 32-bit timestamps and will get issues?

I also expect some applications casting time_t to int. I'm quite sure that we don't know well all devices and softwares impacted by this bug :-)

Red Hat Enterprise Linux 7.5 is out

Posted Apr 13, 2018 17:01 UTC (Fri) by ncm (guest, #165) [Link]

We can get a fair bit more mileage out of 32-bit second-counts by making them unsigned.

That takes us through 2105 without changing epochs. By then, it will be the machines' problem to figure out. They will have much bigger and better reasons to hate us than the Y2106 problem, but I am categorically ill equipped to guess which ones those will be.

Time zones? Nation-states? Java?