atof1 test fails on i686 arch, when --enable-rtree option enabled (Fedora Rawhide)

(1) By zmiklank on 2022-03-07 14:36:09 [link] [source]

Hello,

currently, I am trying to package sqlite 3.38.0 to Fedora Rawhide.

The process is mostly finished, except on the failing test, see $subj.

The build of a standard package in Fedora is done for multiple archs: x86_68, i686, aarch64, ppc64le, s390x.

Except for the i686, everything is fine. For the i686 the atof1 test fails, in incorrect float->string->float and string->float conversions.

Here is the test log (it will be available only for two weeks, IMO): https://kojipkgs.fedoraproject.org//work/tasks/60/83780060/build.log

Even though the log is from rpmbuild, I removed all our patches and flags, and did only:

./configure --enable-rtree
make
make test

,for the reproduction to be the closest to upstream code, as possible.

We agreed, that it will be OK to just remove this one test for the i686, but I wanted to let you (upstream) know before publishing the rpm, to clarify.

Do you think, that it is necessary to fix this, or is it OK, to ship the package even with this bug(?)?

Let me please know, if you need more information.

BR, Zuzana

(2) By Richard Hipp (drh) on 2022-03-07 18:47:14 in reply to 1 [link] [source]

I don't have access to i686 hardware for testing. So please tell me: If you run "tclsh" and type the command "parray tcl_platform", what do you get?

I get this:

tcl_platform(byteOrder)     = littleEndian
tcl_platform(engine)        = Tcl
tcl_platform(machine)       = x86_64
tcl_platform(os)            = Linux
tcl_platform(osVersion)     = 4.4.0-210-generic
tcl_platform(pathSeparator) = :
tcl_platform(platform)      = unix
tcl_platform(pointerSize)   = 8
tcl_platform(threaded)      = 1
tcl_platform(user)          = drh
tcl_platform(wordSize)      = 8

Please follow-up with a post that shows the output for your i686 machine and then I'll see if I can't devise some kind of (blind) fix.

(3) By zmiklank on 2022-03-08 10:43:11 in reply to 2 [source]

Thanks,

I do not have access to the i686 machine either, I locally tested the build in mock [1]. However, I suppose, that if the build is sufficient to be done in mock, then also an output of this command would be correct in the mock environment:

<mock-chroot> sh-5.1# tclsh
% parray tcl_platform
tcl_platform(byteOrder)     = littleEndian
tcl_platform(engine)        = Tcl
tcl_platform(machine)       = i686
tcl_platform(os)            = Linux
tcl_platform(osVersion)     = 5.16.12-200.fc35.x86_64
tcl_platform(pathSeparator) = :
tcl_platform(platform)      = unix
tcl_platform(pointerSize)   = 4
tcl_platform(threaded)      = 1
tcl_platform(user)          = root
tcl_platform(wordSize)      = 4

Hope this helps.

[1] https://github.com/rpm-software-management/mock

(4) By Sam James (thesamesam) on 2022-04-02 04:26:29 in reply to 1 [link] [source]

See also this report which is quite similar.