| Message ID | pull.1253.v2.git.1655125988.gitgitgadget@gmail.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | ci(GitHub): mark up compile errors, too | expand |
"Johannes Schindelin via GitGitGadget" <gitgitgadget@gmail.com> writes: > * The curly bracket construct was replaced by a proper subshell, to avoid > jumbled output and a race where the exit.status file could be read before > it was written. I do prefer () when making a subshell in a case like this (e.g. upstream of a pipe), so I am happy with this version, but the above is curious. I am not sure how "jumbled output" is possible, let alone "reading exit.status before it is written". The output goes to sed to be processed either way, nobody else other than "$@" produces such an output from there, and sed would not exit until it finishes reading from the upstream so res=$(cat exit.status) won't kick in before the upstream exits. Anyway, thanks, will queue. > Johannes Schindelin (2): > ci(github): use grouping also in the `win-build` job > ci(github): also mark up compile errors > > ci/lib.sh | 10 ++++++++-- > ci/make-test-artifacts.sh | 2 +- > 2 files changed, 9 insertions(+), 3 deletions(-) > > > base-commit: 3069f2a6f4c38e7e599067d2e4a8e31b4f53e2d3 > Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-1253%2Fdscho%2Fci-mark-up-compile-failures-v2 > Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-1253/dscho/ci-mark-up-compile-failures-v2 > Pull-Request: https://github.com/gitgitgadget/git/pull/1253 > > Range-diff vs v1: > > 1: 5212c5ec474 = 1: 5212c5ec474 ci(github): use grouping also in the `win-build` job > 2: 19d6e34f038 ! 2: 34daf06bb71 ci(github): also mark up compile errors > @@ ci/lib.sh: else > - "$@" > - res=$? > + # work around `dash` not supporting `set -o pipefail` > -+ { > ++ ( > + "$@" 2>&1 > + echo $? >exit.status > -+ } | > -+ sed 's/^\(\([^ ]*\):\([0-9]*\):\([0-9]*:\) \)\(error\|warning\): /::\5 file=\2 line=\3::\1/' > ++ ) | > ++ sed 's/^\(\([^ ]*\):\([0-9]*\):\([0-9]*:\) \)\(error\|warning\): /::\5 file=\2,line=\3::\1/' > + res=$(cat exit.status) > + rm exit.status > end_group
On Mon, Jun 13 2022, Johannes Schindelin via GitGitGadget wrote: > Just like we mark up test failures, it makes sense to mark up compile > errors, too. > > In a sense, it makes even more sense with compile errors than with test > failures because we can link directly to the corresponding source code in > the former case (if said code has been touched by the Pull Request, that > is). The only downside is that this link currently is kind of misleading if > the Pull Request did not even touch the offending source code (such as was > the case when a GCC upgrade in Git for Windows' SDK all of a sudden pointed > out problems in the source code that had existed for a long time already). > We will see how the GitHub Actions engineers will develop this feature > further. > > This patch series is based on js/ci-github-workflow-markup. Which also > serves as an example how this looks like if the offending source code was > not touched by the Pull Request: > https://github.com/dscho/git/actions/runs/2477526645 because it still > triggers the above-referenced GCC build failure. > > Changes since v1: > > * Using a comma in the workflow command now, as described in the official > documentation ;-) (Thank you, Ævar) You're welcome! > * The curly bracket construct was replaced by a proper subshell, to avoid > jumbled output and a race where the exit.status file could be read before > it was written. > > Johannes Schindelin (2): > ci(github): use grouping also in the `win-build` job > ci(github): also mark up compile errors It's still genuinely unclear to me what exactly the expected before/after result is, and I wish the 2/2 commit would discuss it. So, in v1 we had this: https://github.com/gitgitgadget/git/actions/runs/2461737185 Where the *summary* for the CI said e.g. "syslog.c line=53#L1", so that was the "needs a comma" bug, now it says syslog.c#L53 instead: https://github.com/dscho/git/actions/runs/2477526645 (your link above). So that's good. But re my earlier comment where I asked/wondered if fixing that would link to the source file at line 53 it still seems to just link to the diff. Is that a bug? The desired result? If the commit was modifying syslog.c would the link work? Clearly an end result where we link to the source file/lines at the rev we're testing is much more useful. I found this discussion: https://github.community/t/are-github-actions-notice-warning-error-annotations-broken/225674 Which has a link to an example run at: https://github.com/IronTooch-ColdStorage/Github-AnnotationTest/actions/runs/1782265048 So isn't this for creating "annotations" for just the regions that would be involved in your diff? I.e. it shows a notice for the line(s) involved in the diff itself, but presumably nothing else? If that's the case I think it would be much more useful to just e.g. wrap $(CC) in some "tee"-like command to spew its output somewhere, and then have a "step" where we extract the warnings/errors emitted, and emit URLs you could click on, unless there's some way to make the GitHub UX emit the same information. I.e. it'll be quite hit & miss whether the annotation will show up in the diff, the compiler will often warn about a line some distance away from the change made, e.g. if a variable is made unused. Unless the intent is only to aggregate them on the summary page, but then why do we need to link to the "line" at all, which will at best work unreliably, and at worst be actively misleading. In any case, needing to do less reading of the tea leaves would be nice, i.e. if the commit message explain what the desired change is exactly, and how it should be handling these cases. Thanks.
Just like we mark up test failures, it makes sense to mark up compile errors, too. In a sense, it makes even more sense with compile errors than with test failures because we can link directly to the corresponding source code in the former case (if said code has been touched by the Pull Request, that is). The only downside is that this link currently is kind of misleading if the Pull Request did not even touch the offending source code (such as was the case when a GCC upgrade in Git for Windows' SDK all of a sudden pointed out problems in the source code that had existed for a long time already). We will see how the GitHub Actions engineers will develop this feature further. This patch series is based on js/ci-github-workflow-markup. Which also serves as an example how this looks like if the offending source code was not touched by the Pull Request: https://github.com/dscho/git/actions/runs/2477526645 because it still triggers the above-referenced GCC build failure. Changes since v1: * Using a comma in the workflow command now, as described in the official documentation ;-) (Thank you, Ævar) * The curly bracket construct was replaced by a proper subshell, to avoid jumbled output and a race where the exit.status file could be read before it was written. Johannes Schindelin (2): ci(github): use grouping also in the `win-build` job ci(github): also mark up compile errors ci/lib.sh | 10 ++++++++-- ci/make-test-artifacts.sh | 2 +- 2 files changed, 9 insertions(+), 3 deletions(-) base-commit: 3069f2a6f4c38e7e599067d2e4a8e31b4f53e2d3 Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-1253%2Fdscho%2Fci-mark-up-compile-failures-v2 Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-1253/dscho/ci-mark-up-compile-failures-v2 Pull-Request: https://github.com/gitgitgadget/git/pull/1253 Range-diff vs v1: 1: 5212c5ec474 = 1: 5212c5ec474 ci(github): use grouping also in the `win-build` job 2: 19d6e34f038 ! 2: 34daf06bb71 ci(github): also mark up compile errors @@ ci/lib.sh: else - "$@" - res=$? + # work around `dash` not supporting `set -o pipefail` -+ { ++ ( + "$@" 2>&1 + echo $? >exit.status -+ } | -+ sed 's/^\(\([^ ]*\):\([0-9]*\):\([0-9]*:\) \)\(error\|warning\): /::\5 file=\2 line=\3::\1/' ++ ) | ++ sed 's/^\(\([^ ]*\):\([0-9]*\):\([0-9]*:\) \)\(error\|warning\): /::\5 file=\2,line=\3::\1/' + res=$(cat exit.status) + rm exit.status end_group