scala-regression-reproducer

name

scala-regression-reproducer

description

Reduce Scala compiler or standard library regressions found in community-build projects to minimal, self-contained scala-cli repros. Use when a Scala nightly breaks a project under community-build3, when Codex needs to auto-discover the matching `scripts/run.sh` project key and baseline Scala version, when the regression must be confirmed with a known-good build and a nightly build, or when a failing build log needs to be minimized into `test.scala` / `test-macro.scala` files.

Scala Regression Reproducer

Use this skill to turn a failing community-build project into a small Scala compiler reproducer with a clear good-vs-bad version boundary.

Workflow

1. Confirm the regression before reducing it.

   • Run scripts/confirm_regression.sh [project-key] [known-good-scala] [bad-scala].
   • Prefer SKIP_BUILD_SETUP=1 scripts/confirm_regression.sh ... once the project already exists in .github/workflows/buildConfig.json.
   • Omit project-key to auto-discover it from the checked-out repo remote and .github/workflows/buildConfig.json.
   • Omit known-good-scala to use publishedScalaVersion from buildConfig.json.
   • Omit bad-scala to let ../scripts/run.sh pick the latest nightly.
   • If the user already has the failing repo checked out and gives an exact scala-cli command plus cwd, reproduce that command in that directory before relying on helper artifacts.
   • For Scala 3.8.x or 3.nightly, use a current Scala CLI runner (at least 1.12.x) for direct verification; older pinned runners can obscure the real boundary.
   • If the helper or build-server path is noisy or hangs, confirm the boundary directly in the checked-out repo with scala-cli compile . -S <version> and retry with --server=false only if needed.
   • scripts/run.sh regenerates .github/workflows/buildConfig.json for the designated project unless SKIP_BUILD_SETUP=1 is set.
   • If you need that regeneration because the project entry is missing, restore .github/workflows/buildConfig.json afterward with git restore .github/workflows/buildConfig.json unless you want to keep the change.
   • scripts/run.sh clones into $PWD/repo and removes any existing repo/ directory there before each run; this helper contains that cleanup by running each case inside out/scala-regression-reproducer/<timestamp>/{good,bad}.
   • The helper stores separate artifacts under out/scala-regression-reproducer/<timestamp>/{good,bad}.

2. Inspect the bad run artifacts.

   • Read build-status.txt, build-summary.txt, and build-logs.txt from the bad/ directory first.
   • Extract the failing module, source file, exact compiler phase, and full assertion or error text.
   • If the bad run passes or the good run fails, stop and resolve the baseline mismatch before reducing.

3. Reduce aggressively, but preserve the failing compiler shape.

   • Keep the compiler trigger, not the library surface area.
   • Inline only the reflection, inline, macro, or type-level sequence that keeps the same compiler path alive.
   • Remove external dependencies last; if a library macro is the trigger, copy only the relevant local expansion logic into self-contained files.
   • For inline access regressions, first try reducing to just the access shape: a class with the protected/private member, the inline companion logic that calls it, and one top-level trigger expression.
   • Split macro definitions and use sites across files when Scala 3 staging blocks same-file macro calls.
   • If a scalac option is required, encode it as a scala-cli //> using options ... directive so the final compile command stays short.

4. Finalize the repro.

   • Write test.scala or a very small file set at the top level of the checked-out repo.
   • Prefer a truly self-contained test.scala once the failure no longer needs repo-local //> using file directives or external dependencies.
   • Re-run the smallest possible compile command on both the good and bad versions.
   • Report the exact commands and the pass/fail boundary.

Discovery Hints

• Prefer the checked-out repo remote (git remote get-url origin) to derive the community-build project-key.
• Match that remote against .github/workflows/buildConfig.json repoUrl entries.
• Use publishedScalaVersion as the first default known-good version.
• build-summary.txt gives module-level failures quickly; build-logs.txt contains the exact compiler output.
• If publishedScalaVersion is missing, pass the known-good Scala version explicitly instead of guessing.
• For Scala CLI projects, the repo-root scala-cli compile . -S <version> result is often the authoritative signal; use helper outputs mainly to discover versions, project keys, and logs.

Constraints

• Keep the final reproducer self-contained unless the user explicitly asks for a library-dependent repro.
• Preserve exact error text, relevant scalac options, and concrete version strings.
• Confirm both sides of the regression boundary before claiming success.