Skip to content

Prepare v0.2.0 release - #4

Closed
bojieli wants to merge 5 commits into
mainfrom
release/0.2.0
Closed

bojieli wants to merge 5 commits into
mainfrom
release/0.2.0

Conversation

@bojieli

@bojieli bojieli commented Aug 19, 2026

Copy link
Copy Markdown
Owner

Summary

  • repair the Android build after the AGP 9 dependency update
  • refresh the Windows app lock file after the ONNX Runtime update
  • date and correct the v0.2.0 changelog

Verification

  • swift test --filter DoNotTypeCoreTests (481 passed, 4 unavailable local audio fixtures skipped)
  • dotnet test windows/DoNotType.Core.Tests/DoNotType.Core.Tests.csproj -c Release --no-restore (410 passed)
  • dotnet build windows/DoNotType.slnx -c Release --no-restore
  • dotnet format windows/DoNotType.slnx --no-restore --verify-no-changes
  • cd android && ./gradlew test
  • cd android && ./gradlew lint assembleDebug --no-daemon
  • shellcheck -x scripts/*.sh eval/*.sh
  • actionlint -no-color

The existing uncommitted Android dictation-screen work is intentionally outside this branch; it is incomplete and remains preserved in the main worktree.

Date the accumulated release notes, leave a fresh Unreleased section, and replace the obsolete claim that the packaged Windows app has never been launched with the narrower hardware verification boundary.
Speech-recognition clients surface a blank transcript as ProviderError.emptyOutput so real dictations cannot disappear silently. The silence harness exists specifically to expect that result and must classify it the same way ProviderProbe does, rather than reporting every correct recognizer response as an error.
@bojieli

bojieli commented Aug 20, 2026

Copy link
Copy Markdown
Owner Author

Superseded by main. Three of the five commits here are now on main by cherry-pick — fix(eval): count empty silence transcripts as empty (947e261), fix(release): skip unavailable private attestations (aab2bda), and docs(release): prepare the 0.2.0 changelog (8673406, re-dated to the 20th since the tag was never cut).

The two left behind were both aimed at the same release failure, and neither was its cause:

  • fix(release): verify the Swift CLI version format changed the assertion to test "$cli_version" = "$VERSION", which cannot pass either: the stamper appends (<commit>, <date>) to the literal, so the built CLI never prints a bare version.
  • fix(release): verify the macOS signing team replaced the Authority=Developer ID Application: grep on the theory that hosted runners omit that line. The failing run's own log disproves it — Authority=Developer ID Application: Li Bojie (5BR9M56H9W) is right there in the codesign dump, immediately before exit code 1. Dropping that grep would weaken the check that a release did not silently lose its Developer ID identity, without fixing anything.

The actual cause was the line after the signature greps: dnt --version | grep -q "^dnt $VERSION (". Swift ArgumentParser prints configuration.version verbatim, so macOS printed a bare 0.1.0 while Windows printed dnt 0.1.0 — the assertion could never pass on macOS. Fixed on main by making both CLIs print the one format (51e8f22) rather than by relaxing the assertion, with CI now checking the stamped literals and both CLIs' real output so this cannot reach a tag again (39d6ef5).

@bojieli bojieli closed this Aug 20, 2026
@bojieli
bojieli deleted the release/0.2.0 branch August 20, 2026 01:32
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant