Separate a package's configured update source from its provenance, and fix channel selection - #42
Merged
Conversation
…the proper 'stable' channel
Contributor
Test results — macos-latest ✅1114 · ubuntu-latest ✅1114 · windows-latest ✅1111Summary
🎉 No failed tests in this run. Github Test Reporter by CTRF 💚 🔄 This comment has been updated |
…m its provenance `currentSource` served as both the source the next update should use and the one the installed copy came from, so a source change was indistinguishable from an update that had already applied it. `configuredSource` now holds the intent while `currentSource` records the install, stamped with the feed URL it came from so provenance survives a later change to the declared feed or a userFeed override. Existing configs migrate to schema v0.9.0. Channel selection reads from those same records: an explicitly requested channel wins over the recorded one, a channel the feed no longer offers falls back to the default unless the source is pinned, and a feed marking several channels as the default is reported instead of resolved by table order. merge-feed leaves exactly one channel flagged, so publishing can't produce that state in the first place.
…g nothing A comparator group is split on whitespace, so `>= 1.2.3` became a bare `>=` and a bare version; the operator token parsed as a wildcard and reduced the group to an interval no version falls in, returning false for every input with no error to show for it. node-semver accepts that spacing, so an operator is joined to the version following it before the group is tokenized. Hyphen ranges are matched before that step and keep their spaces.
A package installed for the first time recorded no update source in the config file — neither the source configured for it nor where the installed copy came from — so its next update check had nothing to go on. Its record binds the shared config file through setFile, which swapped the view's handler but left its hive in the old handler's tree. A save on the new handler never finds it there, so it wrote nothing and reported success. setFile now moves the hive across. It also deferred the file read, as it had since 2015 when it only recorded a path and left the constructor to load. A handler it creates today starts on an empty tree, so a view bound that way sits on an empty hive that the next save writes over the file's entry. It now reads the file unless the caller passes noLoad, and mergeHive fails a save whose view has no hive in the tree at all.
The refresh writes its output only where it changed something, and serve-updates loaded that output unconditionally, so a feed already in sync with the working copy died on a file that was never written. It now serves the source feed in that case, which is what the refreshed copy would have contained.
… to serve multiple channels simultaneously
…same hive A save merged the hive the handler happened to hold, so every key another writer had changed in it since that handler last read the file went back to the older value. Two handlers over one file is not exotic: a self-update reloads the module and leaves the old and new instances with a tree each, and two Aegisub instances have one apiece by definition. Each handler now keeps the file's contents as of its last load or save and writes the difference against it, so a field it never touched keeps whatever the file holds. A hive dropped from the tree is still dropped from the file, but only by a handler that actually held it.
A record writes its config when loading finds it out of sync, and that check is the generic import of scriptFields. `version` left that list when it became a computed property persisted as a string, so a release that changed nothing else was never noticed and the recorded version stayed behind. DependencyControl's own entry is where it shows, since the instance that finishes its update is the old one and no later load corrected it. Compare the recorded version against the declared one directly, as the packed-int field in scriptFields used to.
…until reload FeedLoader copied the fetch policy (blockPrivateHosts, feed size and timeout caps) into fields at construction, so disabling private-host blocking mid-session left feed fetches refusing loopback URLs for the rest of the session. The policy is now read from the config each time a feed is built, as package-file downloads already did.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Splits a package's configured update source from the record of where its installed copy actually came from, and makes channel selection deterministic — the two halves of a CI mishap that published DependencyControl's internal
mainchannel as a second default and left installs tracking a channel that never receives releases.configuredSourcevscurrentSource— the former holds the source and channel the next update should use, the latter records the completed install, stamped with the feed URL so provenance survives a later change to the declared feed or auserFeedoverride. Configs migrate to schema v0.9.0 automatically, and installs stranded onmainmove tostable.merge-feedleaves exactly one channel flagged, so a publish can't create that state again.ConfigView.setFile— bound a view to a file while leaving its hive in the previous handler's tree, so a save wrote nothing and reported success; a freshly installed package therefore recorded no source at all. It now moves the hive across and reads the file unless the caller opts out.SemanticVersion— a range written>= 1.2.3, with a space after the operator, parsed to an interval no version satisfies and returned false for everything without an error.The published feed still needs
mainrepublished once at v0.9.0, so installs stuck on it pick up the migration before that channel is retired.🤖 Generated with Claude Code