[Development] Help requested: 32-bit and the year 2038

[Thiago Macieira]

On Tuesday, 31 January 2023 15:00:07 PST Thiago Macieira wrote:
> As most of you are aware, a signed 32-bit time_t overflows in 2038. Linux
> has recently deployed "time64_t" (by certain values of "recently", as in
> 2015, see [1])

> I don't know if this is an emulation bug. It's likely.

Ah, it looks like the kernel side of this didn't get merged until Linux 5.1 
(2019), with commit 48166e6ea47d23984f0b481ca199250e1ce0730a. A quick look at 
QEMU sources shows it did get support for those system calls in 2020. That 
means that the QEMU in the Ubuntu 20.04 in the CI is probably too old to 
emulate the system call.

While I will be asking for an update in the infra there, that brings the 
question up: is it acceptable to require Linux 5.1 or later for 32-bit 
deployments of Qt? If not for 6.6, when would it be?

If it's not going to be in the near future, I *can* modify the patch to detect 
that the new system call is not implemented and then fall back to the old one. 
That would mean that every contended mutex or semaphore will incur two system 
calls instead of 1 on Linux 5.0 or older (which I find acceptable; just upgrade 
to regain performance).

Let me know what the community prefers. I don't have an opinion.

-- 
Thiago Macieira - thiago.macieira (AT) intel.com
  Cloud Software Architect - Intel DCAI Cloud Engineering
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 5167 bytes
Desc: not available
URL: <http://lists.qt-project.org/pipermail/development/attachments/20230131/16e5ed39/attachment.bin>