func4 test failures on Linux i586 in 3.45.2

(1) By Reinhard Max (reimax) on 2024-03-14 16:51:06 [link] [source]

I see these two tests fail on openSUSE Tumbleweed for i586:

[  567s] ! func4-2.46 expected: [{}]
[  567s] ! func4-2.46 got:      [9007199254740992.0]
[  567s] ! func4-5.23 expected: [{}]
[  567s] ! func4-5.23 got:      [9007199254740992]

These failures do not happen on version 3.44.2 and I haven't checked 3.45.[01] yet.

(2) By Richard Hipp (drh) on 2024-03-14 19:26:07 in reply to 1 [link] [source]

What is the best way for me to repro the problem, not having access to a i568 machine? Is there a compiler option I can use on my x64 machines that cause gcc to generate code that emulates (precisely) i568 floating-point behavior?

(3) By Reinhard Max (reimax) on 2024-03-15 10:26:23 in reply to 2 [link] [source]

These test failures can be reproduced on 32 bit VMs or in 32 bit userland chroot environments running on 64 bit kernels. I've set up a VM for you and will send you the login details separately.

See also Checkin [bfd8d910] fails on i686 which seems to suffer from a similar issue.

(4) By Richard Hipp (drh) on 2024-03-15 13:20:34 in reply to 1 [link] [source]

Fix check-in at d0fbe779bc2460e1. This is a change to the TCL-based test case only. No changes to any C code.

(5) By Reinhard Max (reimax) on 2024-03-15 14:56:35 in reply to 4 [source]

Fix confirmed. Thanks!