[philippsuess]

💡 Motivation and Context

test-ios-simulator ran three independent make targets one after another on a single runner, taking 24–38 minutes on recent main runs (e.g. run 40654985085: 22 + 6 + 8 min of test steps, 38 min total). Each target already does its own full xcodebuild (different scheme, configuration or compile flags), so nothing is shared between them.

This splits them into a matrix of three parallel jobs, so wall time becomes the longest job instead of the sum (about 17–22 min instead of 31–38):

• test-ios-simulator (unit tests) → make testOniOSSimulator
• test-ios-simulator (survey UI) → make testSurveyUI
• test-ios-simulator (presentation masks) → make testPresentationMasks

Notes for reviewers:

• Required check name is preserved. The repo ruleset requires a check named test-ios-simulator. A new test-ios-simulator job now aggregates the matrix (needs + if: always(), fails unless the matrix result is success), so no ruleset change is needed.
• Log artifacts are split per job (ios-test-logs-unit, -survey-ui, -presentation-masks) since artifact names must be unique. Nothing else references ios-test-logs.
• The flaky (retried) test report only runs in the unit tests job, the only one that retries.
• Markdown-only PRs behave as before: every job echoes and passes.

💚 How did you test it?

CI only. Maintainers: workflow runs on this PR and on the test PRs below need your approval. The test PRs are draft, DO NOT MERGE, and are closed once validated.

• Green path (this PR): the three test-ios-simulator (…) jobs and test-ios-simulator are green; wall time about 17–22 min
• Failure path (test(ci): DO NOT MERGE - failure path for #982 #983): test-ios-simulator (survey UI) and test-ios-simulator are red; the other two matrix jobs still finish
• Markdown-only path (test(ci): DO NOT MERGE - markdown-only path for #982 #984): detect-markdown-only is stubbed to true; all jobs skip their work and pass; test-ios-simulator is green

📝 Checklist

• All commits are signed (see Commit signing)
• I reviewed the submitted code.
• I added tests to verify the changes.
• I updated the docs if needed.
• No breaking change or entry added to the changelog.

🤖 Agent context

Autonomy: Human-driven (agent-assisted)

Written with Claude Code. I analyzed the slow run's per-step timings, compared the workflow with how getsentry/sentry-cocoa structures its test jobs (matrix of parallel jobs), and chose the matrix split as the biggest low-risk win. Faster runners, simulator boot overlap and compile caching were considered and left out of this PR.

🤖 Generated with Claude Code

[marandaneto]

The code looks good; no qualifying findings. Verdict: correct.

[marandaneto]

makes sense