fix: make check-stubs distinguish test failures from tooling errors

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

scripts/check-stubs.sh

@@ -1,24 +1,51 @@
 #!/usr/bin/env bash
 # Usage: scripts/check-stubs.sh <js|scala|fsharp>
-# Runs each module M01..M11 that has tests against the stubs in src/. Fails if any module's tests pass.
+# Runs each module M01..M11 that has tests against the stubs in src/. Fails if any module's tests pass,
+# and also fails (exit 2) on tooling/build errors, so a broken toolchain can never look like "stubs fail".
 set -uo pipefail
 track="${1:?usage: check-stubs.sh <js|scala|fsharp>}"
 repo="$(cd "$(dirname "$0")/.." && pwd)"
 dir="$repo/$track"
+export PATH="$HOME/.dotnet:$HOME/.local/bin:$PATH"
+
 case "$track" in
-  js)     has() { compgen -G "$dir/test/m$1-*.test.js" >/dev/null; } ;;
+  js)     tool=node;      has() { compgen -G "$dir/test/m$1-*.test.js" >/dev/null; } ;;
-  scala)  has() { compgen -G "$dir/test/M$1*.test.scala" >/dev/null; } ;;
+  scala)  tool="${SCALA_CLI:-scala-cli}"; has() { compgen -G "$dir/test/M$1*.test.scala" >/dev/null; } ;;
-  fsharp) has() { compgen -G "$dir/tests/M$1*.fs" >/dev/null; } ;;
+  fsharp) tool=dotnet;    has() { compgen -G "$dir/tests/M$1*.fs" >/dev/null; } ;;
   *) echo "unknown track: $track" >&2; exit 2 ;;
 esac
+if ! command -v "$tool" >/dev/null 2>&1; then
+  echo "check-stubs: required tool '$tool' not found on PATH" >&2; exit 2
+fi
+[ -x "$dir/test.sh" ] || { echo "check-stubs: $dir/test.sh missing or not executable" >&2; exit 2; }
+
+# Succeeds if the (ANSI-stripped) output in $1 shows tests really ran and at least one failed
+# (stub throws / ??? / failwith count as failures), as opposed to a compile or tooling error.
+tests_failed() {
+  case "$track" in
+    js)     ! grep -qE 'SyntaxError|ERR_MODULE_NOT_FOUND|does not provide an export|ReferenceError' <<<"$1" \
+              && grep -qE '^ℹ tests [1-9]' <<<"$1" && grep -qE '^ℹ fail [1-9]' <<<"$1" ;;
+    scala)  ! grep -qE 'Compilation failed|\[error\]|Error compiling|Downloading failed' <<<"$1" \
+              && grep -qE '==> X' <<<"$1" ;;
+    fsharp) ! grep -qE 'error (FS|MSB|NU)[0-9]+|[Bb]uild failed|Build FAILED' <<<"$1" \
+              && grep -qE '[1-9][0-9]* (failed|errored)' <<<"$1" ;;
+  esac
+}
+
 status=0
 for n in 01 02 03 04 05 06 07 08 09 10 11; do
   m="M$n"
   if ! has "$n"; then echo "$m: no tests"; continue; fi
-  if (cd "$dir" && ./test.sh "$m") >/dev/null 2>&1; then
+  out="$(cd "$dir" && ./test.sh "$m" 2>&1)"; rc=$?
+  out="$(sed $'s/\x1b\\[[0-9;?]*[A-Za-z]//g' <<<"$out")"
+  if [ $rc -eq 0 ]; then
     echo "$m: FAIL (tests pass against stubs)"; status=1
-  else
+  elif tests_failed "$out"; then
     echo "$m: ok (fails against stubs)"
+  else
+    echo "$m: ERROR (tooling/build error, exit $rc; not a test failure)" >&2
+    echo "$out" >&2
+    exit 2
   fi
 done
 exit $status