ci: add Start/End times to the log

ci: more osx test from stage2 to stage-3 so that stage-2 isn't blocked by [slow] osx tests.
Also mswin-opt-impi has been stable on ps4 so disable allow_failure

Merge branch 'balay/glci-werror' into 'master'

gitlab-ci: use -Werror with more tests

See merge request petsc/petsc!2075

gitlab-ci: use -Werror with more tests

disable for opensolaris, win, pgi tests

Merge branch 'barry/2019-09-10/jed/gitlab-ci' into 'master'

Barry/2019 09 10/jed/gitlab ci

See merge request petsc/petsc!2042

Merge branch 'scott/test-improve' into 'master'

Various improvements to test harness

See merge request petsc/petsc!2024

OpenBLAS should now work incorrectly independent of the machine it is built on

Uses the undocumented CORE variable to set GENERIC

Commit-type: portability-fix, testing-fix

Merge branch 'barry/2019-08-22/fix-fblaslapack-for-nag' into 'master'

Barry/2019 08 22/fix fblaslapack for nag

See merge request petsc/petsc!2004

Merge branch 'master' into 'barry/2019-08-22/fix-fblaslapack-for-nag'

# Conflicts:
# lib/petsc/conf/test

gitlab-ci: save configure.log on the build machine

Various improvements

1. Test logging now separates tap output and just the error output.
What is streamed is now controlled with the -o flag to the shell
scripts which now has two options curr

Various improvements

1. Test logging now separates tap output and just the error output.
What is streamed is now controlled with the -o flag to the shell
scripts which now has two options currently: 'interactive' and
'err_only'. The V=0 flag and OUTPUT=1 flags to gmakefile.test
turn on the err_only option
The OUTPUT=1 is default for gitlab

2. gmakegentest.py has a (-c,--check-output) option which does nothing
but check if the output files specified by tests are in the output
directories, and return a non-zero error code if they are not there
check_output target is in outer makefile and add `make check_output` to
stage zero tests to prevent running stage 2 and 3 with missing output
files

3. report_tests.py was not giving the correct command to invoke just the
files that failed. This is because it was coded assuming a
particular naming scheme for the tests: an assumption that was both
incorrect, and led to an ill-posed problem for fixing: If a test
label has underscores, it was not easy to fix. The solution was to
change to telling users to globsearch. In fixing, globsearch had to
be generalized to allow for multiple patterns: e.g.,
make -f gmakefile test globsearch='dm_impls_plex_tutorials-ex5_* ts_tutorials-ex11_*'

4. Bug fix for subtests with loop variables
(Reported Jakub Kruzik <jakub.kruzik@vsb.cz> on 6/26/19)

5. This does a minor fix in this utility script which prints out datafiles.

Commit-type: testing-fix, feature

Out check_output rule to outer makefile

Commit-type: feature
Reported-by: gitlab-ci

show more ...

Gitlab-CI should tell me what CONFIG_OPTS is

Commit-type: feature

Merge branch 'jed/gitlab-ci'

fixes for c89

fix c89 stage-1 test - CONFIG_OPTS variable with quotes and spaces is not processed correctly by CI

CI: always check for latest datafiles

remove opensolaris from stage2

switch stage2 mswin test to a separate mswin runner

Merge branch 'jed/gitlab-ci'

fix: IPL64 -> ILP64

attempt to fix linux-pkgs-opt errors by binding this test to a single machine [compute004]
Also move openblas tests from cg to this box [so that cg, es are mostly working on valgrind tests]

CI: remove osx-prefix [it doesn't add value - and overloads OSX machines]

CI: add a new test for nagfor

Merge branch 'master' into this one.

Merge branch 'jed/gitlab-ci'

openblas: tests should use OPENBLAS_NUM_THREADS=1