Now, Self-hosted LiveSync has finally begun to be split into the Self-hosted LiveSync plugin for Obsidian, and a properly abstracted version of it.
This may not offer much benefit to Obsidian plugin users, or might even cause a slight inconvenience, but I believe it will certainly help improve testability and make the ecosystem better.
However, I do not see the point in putting something with little benefit into beta, so I am handling this on the alpha branch. I would actually preferred to create an R&D branch, but I was not keen on the ampersand, and I feel it will eventually become a proper beta anyway.
### Refactored
- Separated `ObsidianLiveSyncPlugin` into `ObsidianLiveSyncPlugin` and `LiveSyncBaseCore`.
- Now `LiveSyncCore` indicates the type specified version of `LiveSyncBaseCore`.
- Referencing `plugin.xxx` has been rewritten to referencing the corresponding service or `core.xxx`.
### Internal API changes
- Storage Access APIs are now yielding Promises. This is to allow more limited storage platforms to be supported.
### R&D
- Browser-version of Self-hosted LiveSync is now in development. This is not intended for public use now, but I will eventually make it available for testing.
- We can see the code in `src/apps/webapp` for the browser version.
- No longer unexpected deletion-propagation occurs when the parent directory is not empty (#813).
### Revert reversions
- Reverted the reversion of ModuleCheckRemoteSize. Now it is back to the service feature.
- No longer deleted files are not clickable in the Global History pane.
- Diff view now uses more specific classes (#803).
- A message of configuration mismatching slightly added for better understanding.
- Now it says `When replication is initiated manually via the command palette or ribbon, a dialogue box will open to address this.` to make it clear that the user can fix the issue by themselves.
### Refactored
- `ModuleRedFlag` has been refactored to `serviceFeatures/redFlag` and also tested.
- `ModuleInitializerFile` has been refactored to `lib/serviceFeatures/offlineScanner` and also tested.
- Fixed the inexplicable behaviour when retrieving chunks from the network.
- Chunk manager has been layered to responsible its own areas and duties. e.g., `DatabaseWriteLayer`, `DatabaseReadLayer`, `NetworkLayer`, `CacheLayer`,and `ArrivalWaitLayer`.
- All layers have test now!
- `LayeredChunkManager` has been implemented to manage these layers. Also tested.
- `EntryManager` has been mostly rewritten, and also tested.
- Now we can configure `Never warn` for remote storage size notification, again.
- Unexpected errors no longer occurred when the plug-in was unloaded.
- Hidden File Sync now respects selectors.
- Registering protocol-handlers now works safely without causing unexpected errors.
### Refactored
- LiveSyncManagers has now explicit dependencies.
- LiveSyncLocalDB is now responsible for LiveSyncManagers, not accepting the managers as dependencies.
- This is to avoid circular dependencies and clarify the ownership of the managers.
- ChangeManager has been refactored. This had a potential issue, so something had been fixed, possibly.
- Some tests have been ported from Deno's test runner to Vitest to accumulate coverage.
- `ModuleTargetFilter`, which was responsible for checking if a file is a target file, has been ported to a serviceFeature.
- And also tests have been added. The middleware-style-power.
- `ModuleObsidianAPI` has been removed and implemented in `APIService` and `RemoteService`.
- Now `APIService` is responsible for the network-online-status, not `databaseService.managers.networkManager`.
- Hidden file synchronisation now works!
- Now Hidden file synchronisation respects `.ignore` files.
- Replicator initialisation during rebuilding now works correctly.
### Refactored
- Some methods naming have been changed for better clarity, i.e., `_isTargetFileByLocalDB` is now `_isTargetAcceptedByLocalDB`.
### Fixed
- Unlocking the remote database after rebuilding has been fixed.
### Refactored
- Now `StorageEventManagerBase` is separated from `StorageEventManagerObsidian` following their concerns.
- Now `FileAccessBase` is separated from `FileAccessObsidian` following their concerns.
During concurrent initialisation (UPDATE STORAGE runs up to 10 ops in
parallel), getAbstractFileByPath can return null for .md files whose
vault index entry hasn't been populated yet, even though the file
already exists on disk. This causes vault.create() to throw "File
already exists."
The same root cause (stale in-memory index) was already identified for
non-.md files (see comment above) and handled via adapterWrite. Extend
that workaround to .md files by catching the "File already exists"
error and falling back to adapterWrite, consistent with the existing
approach.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
