Skip to main content
Version: Next (unreleased)

Upgrade notes

Notes for anyone upgrading a workflow, or an instructor repository, from an earlier GrillMyCode release. The first section covers how releases reach a workflow. The notes follow it, newest changes first.

Choosing when to upgrade​

The uses: line of a workflow names the release it runs.

uses: line ends inRunsChanges when
@v0The newest v0 releaseAny v0 release is published
@v0.30The newest v0.30 releaseA fix to v0.30 is published

@v0 is the default. It is what the Workflow Wizard and every example write. Fixes and new features arrive on their own, and a change that needs action is listed in the notes below.

Name a minor version to keep an assignment unchanged, so that every student is assessed the same way from the first push to the last. Find the newest version on the releases page and leave off its last number: release v0.30.2 is named as @v0.30. What it costs:

  • No new features, and in time no fixes. A minor version normally stops receiving fixes once the next one is released.
  • These pages describe the newest release. An input added after the version you named is ignored by it.
  • Moving up is yours to do. Change the line in the template repository. Students who have already accepted get the change as they would a new workflow: see Students who have already accepted.

Two limits:

  • An exact version is not held. @v0.30.2 runs the newest v0 release, as @v0 does. Name the minor version instead.
  • Use v0.30 or later. An earlier minor version may run the newest v0 release.

File patterns are more forgiving, and broad overrides no longer bring back protected files​

Most workflows need no change. These apply to additional_exclude_patterns and exclude_pattern_overrides; see Pattern syntax.

  • Patterns that matched nothing now work. data/, /data/**, ./data/** and *.{js,ts} are read as their author meant, and a plain name such as data also covers the folder of that name. Patterns may be separated by line breaks as well as commas. If a workflow carries one of these, more files are excluded or re-included than before.
  • Case is ignored in your patterns and in the patterns always excluded, so README.MD and Data/ are now matched. Auto-detected template patterns are unchanged.
  • A broad override no longer re-includes environment files, lock files or dependency folders. frontend/** used to bring back frontend/.env and frontend/node_modules/. Name them to assess them: frontend/.env, *.lock, node_modules/. See Protected files.
  • A leading ! or # now produces a warning. What the pattern does is unchanged.

The issue carries hidden data, and only GrillMyCode's own issues are replaced​

Nothing to change in a workflow. Each assessment issue now holds its questions' numbers, files and lines as hidden data, which the VS Code extension reads.

  • Issues posted by an earlier release are still recognized, and the next run updates them as before.
  • An issue a person wrote with the same title and the assessment label used to be overwritten, or deleted as a duplicate. It is now left alone. See When the questions are regenerated.
  • GrillMyCode Companion 0.2.0 and earlier read the new issues as they did the old ones.

Quizzes are built from data/questions.json​

Each run now writes a questions.json in a data/ subfolder of the student's folder, and the quiz workflow builds the quiz from it alone. raw-ai-output.md moved into data/ too.

  • Folders assessed by an earlier release have no data/questions.json, so the quiz workflow skips them. Their existing quiz files are left as they are. To rebuild a student's quiz, run GrillMyCode for that student again. The old raw-ai-output.md at the top of the folder isn't removed; delete it if you like.
  • To correct a question in the quiz, edit it in data/questions.json. Editing questions.md doesn't change the quiz.

The first run after an upgrade regenerates every student's quiz​

Expected, once. The quiz workflow decides what to rebuild by comparing a content hash stored inside each .imscc and .csv. A repository that has just received an updated workflow has no current hashes on file, so a single run rebuilds the package for every student, serialized by the workflow's concurrency group. For a class of thirty that is a long run, not a broken one — subsequent pushes go back to rebuilding only the student who pushed.