gsl-2.7.1-4 fails to build due to a test suite failure: ======================================== gsl 2.7.1: spmatrix/test-suite.log ======================================== # TOTAL: 1 # PASS: 0 # SKIP: 0 # XFAIL: 0 # FAIL: 1 # XPASS: 0 # ERROR: 0 .. contents:: :depth: 2 FAIL: test ========== FAIL: gsl_spmatrix_complex_float_scale_columns[53,107](COO) real (6.17895793914794922 observed vs 6.17895841598510742 expected) [21704] FAIL: gsl_spmatrix_complex_float_scale_columns[53,107](COO) imag (144.412811279296875 observed vs 144.412826538085938 expected) [21709] FAIL: gsl_spmatrix_complex_float_scale_columns[53,107](COO) real (-26.8551254272460938 observed vs -26.8551292419433594 expected) [21710] FAIL: gsl_spmatrix_complex_float_scale_columns[53,107](COO) imag (331.7882080078125 observed vs 331.788177490234375 expected) [21713] FAIL: gsl_spmatrix_complex_float_scale_columns[53,107](COO) real (233.1878662109375 observed vs 233.187850952148438 expected) [21714] FAIL: gsl_spmatrix_complex_float_scale_columns[53,107](COO) real (-64.1133270263671875 observed vs -64.1133346557617188 expected) [21716] FAIL: gsl_spmatrix_complex_float_scale_columns[53,107](COO) real (7.33260345458984375 observed vs 7.33260154724121094 expected) [21720] FAIL: gsl_spmatrix_complex_float_scale_columns[53,107](COO) imag (172.47174072265625 observed vs 172.471725463867188 expected) [21721] And so on. There is an upstream bug for this already, but it does not show any activity: <https://savannah.gnu.org/bugs/?60335> I'm not sure if the discrepancy is harmless. Reproducible: Always
This is actually not a new issue. The spec file already has: # disable FMA %ifarch aarch64 ppc64 ppc64le s390 s390x export CFLAGS="%{optflags} -ffp-contract=off" %endif GCC's -ffp-contract=fast default seems increasingly suspect to me.
On the other hand, injecting -ffp-contract=off into the build (by removing those conditionals) does not fix the issue on x86-64. Sad.
*** Bug 2218349 has been marked as a duplicate of this bug. ***
(In reply to Florian Weimer from comment #2) > On the other hand, injecting -ffp-contract=off into the build (by removing > those conditionals) does not fix the issue on x86-64. Sad. Looks like it does in fact: ELN scratch build succeeded: https://koji.fedoraproject.org/koji/taskinfo?taskID=102715329 F39 scratch build shows no regressions: https://koji.fedoraproject.org/koji/taskinfo?taskID=102715331
This bug appears to have been reported against 'rawhide' during the Fedora Linux 39 development cycle. Changing version to 39.