Repository navigation
No CI runs the Swift build or tests on pull requests #44
Description
Nothing runs swift build / swift test on PRs today (only homebrew.yml exists), so regressions merge unverified even though CONTRIBUTING.md asks for passing tests. The review also found several ways the build/release scripts fail loudly-late or silently-quiet:
build-llama.shsilently builds without the no-logits patch when it doesn't apply (after a TAG bump), then caches that unpatched build foreverbuild.shhardcodes one Sparkle xcframework slice path; a blank.sign-idyieldscodesign --sign ""release.shvalidates the Sparkle key only after multi-minute notarization; uses deprecated--deep; keeps no notary logs; a DMG built withoutAPPCAST=is silently left off the update feedappcast.pybuilds XML with f-strings (unparseable on]]>notes) and hardcodes the upstream download URL- testbed scripts abort on empty globs;
say -v Samanthaandcp -cassume a stock APFS Mac homebrew.ymlcommits with-am, sweeping unrelated tap files into release commits- worker
text()fails the whole request on one bad input, contradicting the protocol docstring
Fixed by #37.
Activity
ARahim3 commentedon Oct 21, 2026
Thanks @dowjames. CI on pull requests is in now (a workflow that builds and tests the engine on macOS and Linux and runs the shell's tests). From the script review I'm taking three: the llama.cpp build must fail if the patch doesn't apply, the release script checks the update key before notarizing, and the appcast writer escapes the notes. Closing this one.