emacs FTBFS on s390x because of a failing unit test

[Olivier Tilloy]

As a follow up to bug #1911209, the following test is failing reliably on s390x only:

Running 6 tests (2021-01-12 15:32:18+0000, selector `(not (or (tag :expensive-test) (tag :unstable)))')
   passed 1/6 lcms-ciecam02-gold (0.000133 sec)
Test lcms-cri-cam02-ucs backtrace:
  signal(ert-test-failed (((should (eql 0.0 (lcms-cam02-ucs '(0.5 0.5
  ert-fail(((should (eql 0.0 (lcms-cam02-ucs '(0.5 0.5 0.5) '(0.5 0.5
  #f(compiled-function () #<bytecode 0xaa890bf053>)()
  ert--run-test-internal(#s(ert--test-execution-info :test #s(ert-test
  ert-run-test(#s(ert-test :name lcms-cri-cam02-ucs :documentation "Te
  ert-run-or-rerun-test(#s(ert--stats :selector (not (or (tag :expensi
  ert-run-tests((not (or (tag :expensive-test) (tag :unstable))) #f(co
  ert-run-tests-batch((not (or (tag :expensive-test) (tag :unstable)))
  ert-run-tests-batch-and-exit((not (or (tag :expensive-test) (tag :un
  eval((ert-run-tests-batch-and-exit '(not (or (tag :expensive-test) (
  command-line-1(("-L" ":/<<BUILDDIR>>/emacs-27.1+1/debian/build
  command-line()
  normal-top-level()
Test lcms-cri-cam02-ucs condition:
    (ert-test-failed
     ((should
       (eql 0.0
     (lcms-cam02-ucs ... ...)))
      :form
      (eql 0.0 3.4662020221023775e-16)
      :value nil))
   FAILED 2/6 lcms-cri-cam02-ucs (0.000129 sec)
   passed 3/6 lcms-dE-cam02-ucs-silver (0.000080 sec)
   passed 4/6 lcms-jmh->cam02-ucs-silver (0.000053 sec)
   passed 5/6 lcms-roundtrip (0.000068 sec)
   passed 6/6 lcms-whitepoint (0.000056 sec)

Ran 6 tests, 5 results as expected, 1 unexpected (2021-01-12 15:32:18+0000, 0.088409 sec)

1 unexpected results:
   FAILED lcms-cri-cam02-ucs

[Olivier Tilloy (osomon)]

That test is not new, it was present in emacs 1:26.3+1-1ubuntu2 (the version currently in the release pocket) and it used to pass. From a cursory glance, I didn't spot relevant changes in upstream code that would explain the regression, so this needs further investigation.

[Olivier Tilloy (osomon)]

This looks like a rounding error: (eql 0.0 3.4662020221023775e-16)

[Olivier Tilloy (osomon)]

I tried rebuilding emacs 1:26.3+1-1ubuntu2 (currently in the release pocket) against hirsute-proposed, and I'm seeing the same unit test failure, so it's not an emacs regression, more likely a toolchain problem.

[Olivier Tilloy (osomon)]

Rebuilding the same version (1:26.3+1-1ubuntu2) in a focal chroot on s390x succeeds, with that unit test passing.

[Olivier Tilloy (osomon)]

Rebuilding the same version (1:26.3+1-1ubuntu2) in a groovy chroot on s390x fails, with that unit test failing like it does on hirsute.
So this is not a new problem, but it's surfacing only now because the package hadn't been rebuilt since focal.

[Olivier Tilloy (osomon)]

xnox uploaded a patch that skips this test: https://launchpad.net/ubuntu/+source/emacs/1:27.1+1-3ubuntu3.