ci(pr): make InspectCode solution discovery robust under errexit

[Chris-Wolfgang]

The blocker

#267 merged the InspectCode job into main at 16:04. Because pr.yaml runs via pull_request_target, every open PR now runs InspectCode from main — and it fails at the "Run InspectCode" step with exit code 2, before jb even starts (the build step passes). #245 and every other open PR are blocked on this.

Root cause

GitHub's shell: bash runs with -e -o pipefail. The step picked a solution by parsing ls:

SLN=$(ls *.slnx 2>/dev/null | head -n1)

This repo has ETL-Abstractions.sln but no .slnx, so ls *.slnx exits 2 → pipefail propagates it → set -e aborts the whole step before the .sln fallback runs. This breaks InspectCode on every repo that has only a .sln (i.e. essentially all of them).

It slipped through because InspectCode needs: detect-projects, and on #267's own PR the protected-file guard failed that job, so InspectCode was skipped and never actually executed before #267 was bypass-merged.

Fix

Rather than band-aid the ls pipeline with || true (which would also mask genuine failures), replace the ls-parsing with nullglob + array expansion:

shopt -s nullglob
solutions=(*.slnx *.sln)
if [ ${#solutions[@]} -eq 0 ]; then
  echo "::error::No .slnx or .sln found at repo root — InspectCode needs one to run."
  exit 1
fi
SLN="${solutions[0]}"

A non-matching glob expands to nothing instead of failing, so discovery is correct under errexit/pipefail by construction — no ls exit-2 to trip on. Listing *.slnx before *.sln keeps .slnx preferred. Also puts ~/.dotnet/tools on PATH so the jb global tool resolves.

Verified in a local bash -e -o pipefail harness across all four cases: only-.sln, only-.slnx, both (.slnx wins), and neither (fails loudly, exit 1).

Merge note

This touches .github/workflows/pr.yaml (a protected file), so Detect .NET Projects will fail by design — it's the only expected failure. Needs an admin-bypass merge. Once merged, I'll re-run InspectCode on #245 (and the other open PRs) to confirm green.

The same bug exists in the canonical repo-template; I'll fold this fix into the open feed-back PR #439.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.