single non-numerical test failure in OpenBLAS/0.3.23-GCC-12.3.0 for x86_64/generic

[boegel]

We're seeing a single non-numerical test failure for OpenBLAS/0.3.23-GCC-12.3.0 on x86_64/generic, see #334 (comment)

                        -->   LAPACK TESTING SUMMARY  <--
SUMMARY                 nb test run     numerical error         other error
================        ===========     =================       ================
REAL                    1326267         2       (0.000%)        0       (0.000%)
DOUBLE PRECISION        1327545         1       (0.000%)        1       (0.000%)
COMPLEX                 786943          3       (0.000%)        0       (0.000%)
COMPLEX16               786918          26      (0.003%)        0       (0.000%)

--> ALL PRECISIONS      4227673         32      (0.001%)        1       (0.000%)


== 2023-09-15 22:50:01,320 openblas.py:116 INFO 4227673 LAPACK tests run - 32 failed due to numerical errors - 1 failed due to other errors

Testing DOUBLE PRECISION Nonsymmetric-Generalized-Eigenvalue-Problem-driver-EIG/xeigtstd < dgd.in > dgd.out  DDRGES3: DGGES3 returned INFO=     9.
  DGS drivers:      1 out of   1555 tests failed to pass the threshold
 passed: 8922
failing to pass the threshold: 1
Info Error: 1

This appears to be a known issue with the LAPACK test suite, see OpenMathLib/OpenBLAS#4032.

Quoting from there: "these test failures crept up after algorithm changes in Reference-LAPACK 3.10 but they appear to be more of a nuisance than an actual defect.", so it seems fine to ignore this for now.

[boegel]

The same problem is also observed with OpenBLAS 0.3.24 on POWER9, see easybuilders/easybuild-easyconfigs#18880 (comment)

[boegel]

Workaround for this problem, which is basically recommended by upstream OpenBLAS maintainer, is to just disable that one flaky test with a patch, see easybuilders/easybuild-easyconfigs#18887, which works like a charm, see #334 (comment)

So, this issue can be closed!