There were a few logs related to bad memory access within libSession and the `toLibSession` convention which we'd originally setup made me nervous as C-friendly objects would be allocated in Swift and then assigned to C struct properties but I'm not sure how the memory would actually behave in this case.
This updated approach unfortunately involves a bunch of duplicate code within 'TypeConversion+Utilities' (and some horrible type aliases for tuples) but as a result we now have compile-time safe 'get' and 'set' functions for any C struct which conforms to `CAccessible` and `CMutable`. The other nice benefit about this change is that the new `set` functions copy memory across into the C structs so we can have more confidence that the memory ownership has shifted to the libSession side of things.
• Fixed some cases where errors weren't being displayed correctly
• Tweaked the "nodeNotFound" error to be a bit more human readable
• Tweaked the BackgroundPoller timeout to have a 5 second buffer instead of a 1 second buffer
• Moved the lastHash pruning into the GarbageCollectionJob instead of the pre-poll fetching to avoid needing to use a write query before polling
• Reworked the Pollers to make their database queries part of the polling stream (and as such, cancellable)
• Fixed an issue where the conversation 'created' timestamp wasn't correctly getting set when creating a group
• Fixed an issue where sending a message to 'Note to Self' may not show the conversation in the conversation list
• Fixed an issue where sharing a message with an attachment might not include the attachment
• Fixed an issue where list paging wouldn't work in some cases after values were inserted into the database after the currently loaded pages
• Added some handling for invalid 'joinedAt' values for groups (seems like we can have an incorrect resolution)
• Added a script to extract the app group during the build and load it in at runtime (fallback to the current one)
• Updated the 'App Store Release' scheme to 'App_Store_Release' to fix build issues on some machines
• Fixed some broken unit tests
• Updated the code to stop sending legacy PNs outside of legacy group conversations
• Updated the logger logic to clean things up and use the local date/time (with time zone info) to ease debugging user reports
• Fixed an issue where messages in a community could incorrectly accept disappearing message settings
• Fixed an issue where duplicate messages could be sent in some cases
• Fixed an issue where the conversation might not scroll to the bottom after sending an attachment
• Fixed an issue where attachment encryption was happening in a db write thread
• Fixed an index out of bounds issue with the mnemonic
• Attempt to fix a crash due to failing to hash SignalAttachment
• Attempt to fix a crash due to list data change inconsistencies
• Ensuring we are shutting down the network when resetting app data
• Updated the broken unit tests
• Fixed an issue where voice messages wouldn't stop playing when deleting a message or leaving a conversation
• Fixed an issue where the notification extension was handling errors and finishing while within a database transaction
• Fixed an issue where the ShareExtension may not have been able to send LinkPreviews
• Fixed an issue where displaying or logging the paths weren't maintaining the correct order
• Tweaked some of the logging behaviours
• Fixed a bad memory crash which could occur when returning from the background
• Fixed an issue where config messages were incorrectly getting double wrapped in protobuf wrappers
• Fixed an issue where the DisappearingMessagesJob was getting run for every individual message process instead of just once after processing all messages from a poll
• Tweaked the polling logic to split the user, groups and community pollers into separate queues
• Tweaked the logic to restart the open group poller immediately after joining a community (to avoid an edge-case where there could be a long delay before the initial poll is made)
• Tweaked the logic for checking if PNs are susceptible to failure to only dispatch to the main thread once
• Cleaned up some warnings
• Updated the app versioning to be at the project level instead of target level
• Added the ability to define a unique hash for a job for deduplication
Properly fixed the busted migration issue
Updated to the latest version of Quick and Nimble (unit testing libraries)
Updated the tests based on the above
Added logic to broadcast the community message request acceptance to SOGS so we can communicate it to message request senders
Fixed an issue where database setting changes wouldn't trigger a live update on a settings screen
Fixed an issue where some setting toggles wouldn't animate the state change
Fixed a rarw force-unwrap crash
Cleaned up the Dependencies so that tests can run synchronously without having to custom set queues as much
Sorted out the crypto and network dependencies to avoid needing weird dependency inheritance
Fixed the flaky tests so they are no longer flaky
Fixed some unexpected JobRunner behaviours
Updated the CI config to use a local build directory for derivedData (now works with build tweaks)
Added parallelisation for the unit tests
Added the SSH_KEY env variable for the 'Upload artifacts' step
Disabled xcbeautify for the prod build (as I expect it's suppressing the build error)
Disabled support for starting a conversation with a 'blinded25' id (would be invalid at this stage)
Fixed a broken unit test
Added the XCBeautify pod (so the CI doesn't need to separately install something)
Updated the CI build script to use XCBeautify
Fixed some broken unit tests
Reworked the way libSession-util builds to use a static library and be less hacky
Updated to the latest version of WebRTC-lib (no longer includes bitcode)
Removed the 'skip_web_rtc_re_rsync' patch as it's no longer an issue since the bitcode was removed so the framework is much smaller
Updated the SessionApp.presentConversation function from using the DBWrite thread if it didn't need to
Updated the PagedDatabaseObserver to process database commits async on a serial queue to avoid holding up the DBWrite thread
Moved another Atomic mutation outside of a DBWrite closure
Refactored the PagedDatabaseObserver 'databaseDidCommit' logic to be much more straightforward
Tweaked a couple of flaky unit tests
Added a patch to the Podfile to avoid rsync'ing and signing WebRTC-lib for simulator builds shaving off 10+ seconds of build time per target due to the sheer size of the WebRTC debug framework
Added some basic unit tests to validate the current search behaviour
Fixed some buggy search behaviours
Removed the 'immediatelyOnMain' extensions as they would break in some cases (eg. upstream errors with multiple 'receive(on:)' calls) resulting in logic running on unexpected threads
Updated the ReplaySubject to add subscribers in an Atomic just to be safe
Updated the code to remove the invalid open group when the user receives an error after joining
Fixed a bug with editing closed group members
Fixed broken unit tests
Removed the 'readPublisherFlatMap/writePublisherFlatMap' functions as they easily resulted in behaviours which held up database threads
Tweaked the logic around starting the open group pollers to avoid an unlikely atomic lock blocks
Updated some logic to avoid accessing database read threads for longer than needed
Updated the OpenGroupManager to only update the 'seqNo' value for valid messages
Cleaned up some double Atomic wrapped instances which had some weird access behaviours
Fixed an issue where a database read thread could have been started within a database write thread
Fixed an issue where the ReplaySubject might not emit values in some cases
Fixed a busted version comparison
Fixed an issue where the config dump population wasn't setting the 'created' timestamp for contacts
Fixed an issue where the 'SyncPushTokensJob' could run logic on the wrong thread
Fixed a bug where the 'scroll to bottom' button wouldn't initial be visible in some cases
Fixed a bug where the 'scroll to bottom' button would fade out when there were subsequent pages
Fixed a bug where an open group image might not get downloaded in some cases
Fixed an issue where we would incorrectly append a wildcard character to the end of a search term that ended in a quotation mark
Finished refactoring the OpenGroupAPI to use PreparedSendData
Morgan Pretty
Ryan ZHAO