Add shellcheck, SwiftFormat and SwiftLint - #410

Closes #380 and #382; the CI half of #381. Second of the maintenance-automation batch.

Formatting had already drifted

22 of 41 files failed swiftformat --lint before any of today's work. The .swiftformat config existed with nothing enforcing it. I checked whether I had caused it — the answer is one file (FaultReportingTests.swift, mine) out of 22; the other 21 predate this session.

Applying the config touches 21 files: 155 insertions, 91 deletions. Mechanical and reproducible with swiftformat ..

HTMLTemplates.swift is excluded, deliberately

It wanted 1,200 indent changes alone. The file is 2-space Allman while the config wants 4-space K&R, so reformatting moves the closing """ delimiters of its multiline strings — and in Swift the closing delimiter's indentation determines how much leading whitespace is stripped from every line. That would change the bytes of every generated report.

It also carries a DO NOT EDIT! This file is autogenerated header, so hand-formatting it is wrong regardless. Excluded from both tools.

Verified the reformat is behaviour-preserving

Generated a report before and after and compared. That comparison found something unrelated and worth knowing: the same binary produces different output on consecutive runs.

run A: selectDevice('6CBE1E1A-E712-42AD-BAA8-A179C1823B56')
run B: selectDevice('E2105793-8D30-4944-AF86-2BBC4017C91F')

Five call sites mint a fresh UUID() per run for DOM handles (TestSummary, TargetDevice, Test ×2, Iteration). Harmless for rendering, but reports are not reproducible — which matters for diffing across runs or content-addressed caching. Filed separately; not changed here.

Without that check I would have blamed the formatting for the differing output.

SwiftLint

Scoped correctly it reports 65 warnings / 24 errors; 31 of those are in the excluded HTMLTemplates.swift. (Unscoped it reports 7,285 warnings because it lints .build/checkouts — i.e. your dependencies.)

The config starts green, non-strict:

• Layout rules SwiftFormat already owns are disabled, so the two tools cannot disagree about the same lines.
• line_length errors above what a formatter can fix — SwiftFormat wraps at 100 but cannot wrap string literals.
• todo is off; TODOs are tracked as issues here, not lint failures.

A gate that is switched on already failing gets ignored. This one can be tightened deliberately.

Two genuine violations are fixed rather than suppressed: a manual add-and-assign that should be += (UnattachedFiles.swift), and a test helper whose leading underscore tripped the naming rule (renamed _testAttachmentsExist → assertAttachmentsExist, which says "not a test" more clearly anyway).

shellcheck

prepareTestResults.sh already passes. So this job guards against regression rather than fixing anything.

I should correct something I said when proposing this batch: I claimed shellcheck would have caught two bugs I shipped today. It would not have. The lexicographic sort was Python inside a heredoc, and the set -e + read interaction isn't a standard shellcheck rule.

Also

swift test writes index.html/report.junit into the fixture directory. Those were untracked and produced a swift build warning on every subsequent build (#389). Now gitignored.

Verification

• swiftformat --lint . → clean
• swiftlint → 0 errors, 26 warnings
• swift test → 22 tests, 1 skipped, 0 failures

🤖 Generated with Claude Code

Summary by CodeRabbit

• Chores

  • Added automated code-quality checks for shell scripts and Swift sources on pushes, pull requests, and manual runs.
  • Added exclusions for generated test-report files and generated templates.

• Tests

  • Strengthened result-count and attachment assertions.
  • Improved test maintenance by removing unnecessary error declarations and updating test helpers.

• Refactor

  • Standardized formatting, comments, naming, and file metadata throughout the project without changing runtime behavior.
  • Updated the sample app to use the current Swift application entry-point syntax.