Skip to content

fix: read changelog file causing app to crash - #34

Merged
craigharman merged 59 commits into
mainfrom
develop
Jun 30, 2026
Merged

craigharman merged 59 commits into
mainfrom
develop

Conversation

@craigharman

Copy link
Copy Markdown
Owner

No description provided.

craigharman and others added 30 commits June 19, 2026 15:37
fix: rebase detection wasn't working

Bug fix: Rebase detection wasn't working

The currentOperation() method in CLIGitService.swift:597 used git rev-parse --git-path rebase-merge which returns a relative path (e.g. .git/rebase-merge). The subsequent FileManager.default.fileExists(atPath:) resolved that relative path against the app's working directory, not the repo root -- so it would never find the rebase marker directories. Fixed by resolving the path against root with root.appendingPathComponent(raw).path.

New feature: Continue and Skip rebase

Added across four files:

GitService.swift -- added continueRebase() and skipRebase() protocol methods
CLIGitService.swift -- implemented them as git rebase --continue and git rebase --skip
RepositoryViewModel.swift -- added continueRebase() and skipRebase() methods using the existing runIntegration pattern
RepositoryWorkspaceView.swift -- the OperationBanner now shows three buttons for rebase operations: Continue, Skip, and Abort (merge still shows just Abort)
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
The release workflow currently skips Sparkle signing entirely when SPARKLE_PRIVATE_KEY isn’t set. Because the app is configured with SUPublicEDKey/SUFeedURL, publishing an unsigned DMG means Sparkle won’t be able to apply updates (and the feed won’t be updated). It’s safer to fail the release build if the key is missing so you don’t publish a “non-updatable” release artifact.

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
The appcast update is not idempotent: it always inserts a new <item> block via content.replace('</channel>', ...), so re-running the workflow for the same version will duplicate entries in appcast.xml. Add a guard to skip insertion when the version already exists.

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
The appcast update step is idempotent (it may leave appcast.xml unchanged when the version is already present), but this commit step always stashes/pops and runs git commit. On a re-run (or if the item already exists), this can fail with “No local changes to save” / “No stash entries found” / “nothing to commit”, causing the release workflow to error unnecessarily. Add a guard to skip this step when appcast.xml has no diff before attempting to stash/commit.

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
…ull action

fix: sparkle framework was not embedded into app on build.
pushBranch falls back to hard-coding the remote as "origin" when the branch has no upstream. That can (a) fail when the repo has remotes but no origin, and (b) push to the wrong remote when ahead/behind was inferred against a same-named remote branch on a different remote (see CLIGitService.refs() remote selection). Consider selecting the remote by: upstream if present → matching remote-tracking branch suffix /<name> → first configured remote; and bail with a banner error if none are available.

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
The fix tries the arch-specific path first (*/macos-arm64_x86_64/Sparkle.framework), then falls back to a broader */Sparkle.framework search if that doesn't match. The error exit is preserved if neither search finds anything.
Two changes made:

Piped the key via stdin using --ed-key-file - instead of writing to a temp file — this is Sparkle's recommended approach for secrets and avoids the key touching disk.
Fixed argument order — --ed-key-file - now comes before the positional "${DMG}" argument, matching the expected usage: sign-update [--ed-key-file <file>] <file-path>.
The old invocation "${SIGN_UPDATE}" "${DMG}" --ed-key-file "${KEY_FILE}" put the DMG path first, which likely caused sign_update to misparse the arguments and fail silently.
@craigharman
craigharman merged commit 380e11f into main Jun 30, 2026
1 check passed
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.

2 participants