On Wed, Jun 08 2022, Johannes Schindelin via GitGitGadget wrote: > From: Johannes Schindelin <johannes.schindelin@xxxxxx> > > Git now shows better information in the GitHub workflow runs when a test > case failed. This commit message should be more on-point. "Git" isn't showing anything, and it's unclear that this is a regression fix for fc5a070f591 (Merge branch 'js/ci-github-workflow-markup', 2022-06-07). > However, when a test case was implemented incorrectly and therefore > does not even run, nothing is shown. The *report* came about because of an incorrectly implemented test (of mine). But the real issue is that your recent change to the CI output is implemented in such a way as to hide entire classes of errors that we'd previously show. The test being broken is besides the point, of course we'll have broken tests. When we do so we should be showing output relevant to the failure. The selling point of your js/ci-github-workflow-markup topic was that we *could* accurately summarize our failures as to make browsing the raw logs unnecessary. Quoting your https://lore.kernel.org/git/pull.1117.git.1643050574.gitgitgadget@xxxxxxxxx/: I know, and I do not expect any new contributor, not even most seasoned contributors to know, that I have to patiently collapse the "Run ci/run-build-and-tests.sh" job's log, and instead expand the "Run ci/print-test-failures.sh" job log (which did not fail and hence does not draw any attention to it). I know, and again: I do not expect many others to know this, that I then have to click into the "Search logs" box (not the regular web browser's search via Ctrl+F!) and type in "not ok" to find the log of the failed test case (and this might still be a "known broken" one that is marked via test_expect_failure and once again needs to be ignored). To be excessively clear: This is not a great experience! As before we agree that it's not the ideal UX experience. But I don't see how this isn't worse, especially for the proverbial new contributor. I.e. when js/ci-github-workflow-markup landed we'd gotten rid of the ci/print-test-failures.sh output entirely, because the topic argued that we *could* accurately summarize these failures, and that any remaining issues requiring that step were so obscure as to be left to the artifacts downloading step. But per my report at https://lore.kernel.org/git/220603.86fskmxd43.gmgdl@xxxxxxxxxxxxxxxxxxx/ (which b.t.w, would be useful to link to in the commit message) this will fail in cases where we'll die early in test-lib.sh itself, e.g. due to a syntax error. Now, I haven't fully dug into what those cases are, but a regression fix really should take the time to do so. E.g. if we fail in a prereq that the test needs (with exit and/or BUG) do we similarly fail, how about e.g. chainlint, or is it just if "eval" itself fails? But getting back to the point, because of that we'll now be asking those using the CI output to view the default GitHub Markup'd failure, and in addition knowing enough about its implementation to know if/when they should be opening the "print test failures" step as well, in case it over-elided some output. As an example, the output is now with this change, and a change on top to re-trigger the reported failure: https://github.com/avar/git/runs/6808880749?check_suite_focus=true#step:4:1757 I.e. per the report above we only show a cryptic "Error: Process completed with exit code 1.", but we do have a "print test failures" step now. And here's the version with the "GitHub markup" changes reverted: https://github.com/avar/git/runs/6808882949?check_suite_focus=true#step:4:1719 Which to be fair isn't much better in itself, but the greater context is that *usually* this step would be sufficient to view a failure, but some of the time it isn't, and you need to know enough to know when that's the case. That's new. > [...]Let's bring back the step that prints the full logs of the failed > tests[...] Except we still have another bug in js/ci-github-workflow-markup where bug in it altered the raw logs themselves. To be fair in this case you can see the failure itself. Here's what we used to do: https://github.com/avar/git/runs/6808882949?check_suite_focus=true#step:5:579 And what we do now: https://github.com/avar/git/runs/6808880749?check_suite_focus=true#step:5:445 I.e. the former prints "expecting success of 3105.36 'setup: HEAD_short_* variables': " followed by the test source itself, and then the "set -x" output, whereas the new output only has the "set -x" output. Both of them get to: t3105-ls-tree-output.sh: 20: local: --abbrev: bad variable name So it doesn't matter much in this particular case, but we'll have other cases where browsing the raw logs is really useful, and having the now-missing context of the test source is even more useful. > , and to improve the user experience, print out an informational > message for readers so that they do not have to know/remember where to > see the full logs. I don't think we should call these "full logs" while the bug above remains unsolved. Perhaps "more verbose logs emitted when running with --github-workflow-markup, omit the flag to get the full *.out logs"? > diff --git a/.github/workflows/main.yml b/.github/workflows/main.yml > index 3fa88b78b6d..cd1f52692a5 100644 > --- a/.github/workflows/main.yml > +++ b/.github/workflows/main.yml > @@ -119,6 +119,10 @@ jobs: > - name: test > shell: bash > run: . /etc/profile && ci/run-test-slice.sh ${{matrix.nr}} 10 > + - name: print test failures When ci/print-test-failures.sh was last in this file before 08dccc8fc1f (ci: make it easier to find failed tests' logs in the GitHub workflow, 2022-05-21) there was no "name" field, that's an unrelated change that shouldn't be part of a narrow regression fix. > + if: failure() && env.FAILED_TEST_ARTIFACTS != '' We likewise just had "if failure()" then, is the distinction different in all these cases? > + shell: bash ...and you've made every single one of them run with "bash" instead of the default shell, which is another "change while at it" that isn't discussed.
