Instead of rethrowing the JavaScript error emitted from a callback function, we should instead pass it to `node::FatalException`.
Callbacks are employed when there is no other JavaScript frame on the stack prior so rethrowing the JS error as a C++ exception is not going to propagate the error back to JavaScript. `node::FatalException` will raise the `uncaughtError` event on the `process` object , print the error and stacktrace, and alert the debugger is there is one attached. This would make our async callbacks behave consistently with Node’s own async callbacks such as `setTimeout` when encountering an error.
* Make it possible to assign a List or Results to a List property.
The change made in #1069 to improve error messages when values of
incorrect types were assigned to a given property was being too
narrow in the types it accepted for list properties, allowing only
arrays. Lists and Results are now accepted once more.
* Update CHANGELOG.md
* Add support for linkingObjects
* Test linkingObjects
* Borrow names helper from list tests
* include computed properties when serializing the schema for the RN debugger
* add API docs
* review comments
* Expose admin users to JS (#1100)
The JS binding used to conflate `SyncUser::is_admin()` with the user being created by calling `Realm.Sync.User.adminToken()`, but now that we expose a user’s role on the server under `is_admin()` this supposition is no longer correct.
#1097 attempted to fix one such case, but fixing it only uncovered another: in `UserClass<T>::all_users()`. I’ve gone through all the callsites of `SyncUser::is_admin()` to make sure they don’t assume an admin token user.
* [1.8.3] Bump version
* add linkingObjects method to Realm.Object
* changelog
The JS binding used to conflate `SyncUser::is_admin()` with the user being created by calling `Realm.Sync.User.adminToken()`, but now that we expose a user’s role on the server under `is_admin()` this supposition is no longer correct.
#1097 attempted to fix one such case, but fixing it only uncovered another: in `UserClass<T>::all_users()`. I’ve gone through all the callsites of `SyncUser::is_admin()` to make sure they don’t assume an admin token user.
* Add and adopt `Value::is_binary` / `Value::to_binary` / `Value::from_binary`.
These methods allow conversions between `BinaryData` and the equivalent
JavaScript types without using `NativeAccessor`. Instead,
`NativeAccessor` now itself delegates to these methods.
* Have `NativeAccessor::value_for_property` and `RealmObjectClass::set_property`
verify that values are valid for the property in question.
If not, we throw an exception that includes the name and type of the
property in question.
`NativeAccessor` is changed to always hold a reference to a `Realm` and
an `ObjectSchema` in order to make this validation possible.
* Fix the Windows build.
* Remove an unused, incorrect forward declaration of a template class
named `Realm` that caused ambiguity with object store's `Realm` class.
* Disambiguate between `realm::js::PropertyType` and
`realm::PropertyType`.
* Update CHANGELOG.md
FIXME: We'll need to update to sync v1.9.2 once it is available as
v1.9.1 does not include core v2.8.1. This may result in the missing
symbols on Linux problem only being fixed when sync is disabled.
implementation.
This matches what core uses for integers, and avoids ambiguity that
otherwise results when `int64_t` is declared as `long` rather than `long long`.
Tweak the signature on some of the accessor methods to take
`JSEngine::Value`s by value rather than by const reference. Some Node
APIs appear to assume that only non-const objects will be used.
* Remove the windows install check
* Node.js Win32 support
* Bring back the `REALM_HAVE_CONFIG` definition
* Download core for Windows when building
* Implement cross-platform node platform.cpp with libuv
* wip
* Make jasmine run quicker
https://github.com/jasmine/jasmine/issues/1204
* Wait for worker process to close in AsyncTests
* Cross-platform paths in tests
* Normalize path separator for forward slash on Windows
* MSVC exception voodoo
* cross-platform uv_cwd
* fix linux build
* make the prepublish script cross-platform
* Disable encryption tests on windows
* ignore vendor/realm-node
* jenkinsfile work
* Only run the prepublish script for publish and pack
* Jenkinsfile work
* Include gyp files in package
* rewrite default_realm_file_directory()
* fix React Native Android build
* delete all realm artifacts in remove_realm_files_from_directory
* bring back build environment variables
* node-pre-gyp windows
* Update CHANGELOG.md
* Get a session’s error handler
* Change the signature of all methods to add the callee
* Deserialize rpc callbacks
* Expose the session error handle on its config property
* When a token refresh error occurs and there is a session error handler
propagate the error to the handler
* Add Session class and expose it to JS
* Add error property on the sync config object for the event handler
* tests
* refactor access token refresh
* chrome debugging
* Add user tests
* Add Sync member to Realm
* Add a simple test for addListener
* Introduce browser User class
* Introduce Sync.User proxy
* Fix User.all
* Add User.logout
* Remove some console.log statements
* Update rpc.cpp from tests-folder
* Use user.current from user-methods, not cpp
* remove user.current from two missing places
* Start server in realmjs and react tests
* Add user tests
* Add Sync member to Realm
* Add a simple test for addListener
* Introduce browser User class
* Introduce Sync.User proxy
* Fix User.all
* Add User.logout
* Remove some console.log statements
* Update rpc.cpp from tests-folder
* Use user.current from user-methods, not cpp
* remove user.current from two missing places
* Start server in realmjs and react tests
* Replace needle with node-fetch
* Make AuthError work in the browser
* Allow for persistent callbacks
* Expose _authenticateRealm to native
* Use clear_test_state in roc
* Streamline test inclusion
* Fix login tests
* Promisify testLogin()
* Disable AsyncTests for now
* Mode jasmine timeout to spec/unit_tests.js
* Avoid use of global for fetch module
* Only include user-tests is sync is enabled
* Skip ROS for android tests
* Add some comments about persistent callbacks
* Fix results-test
* Run async-tests in node
* Add a comment about (user object)
* Cache all users in User.current and return promises in fetch
* node_require async-tests so RN doesn't try and package them
* Add missing parenthesis
* Always invoke sync callbacks on the JavaScript thread
* Address review comments
* Make Callback::m_state be a shared, not a weak pointer
* Use the protected GlobalContext in the session bind handler
* Only dispatch on the event loop if we’re not on its thread
* Update OS
* Use the latest sync
* Update objectstore
* Use realm-sync 7.2
* Add files to project
* Use OS submodule with fix
* Fix object store includes in xcode project
* add missing object store file to binding.gyp
* Update OS with fix for the GlobalNotifier
* Add OS files to Android jni
* Use master@ObjectStore
* Add a test
* Catch TypeErrorException
* Make exception asserts check error type
* Remove commented-out testcase and check for error type
* Test for Error, not TypeError
* Update to sync v1.0.0-BETA-6.3.
This requires updating to current object store master to accommodate the
changes to sync error reporting.
* Add some include folder search paths
* Add thread_safe_reference to the RealmJS target
Under npm3 the `realm` and `nan` modules might end up as siblings so the best way to get the path to `nan` is to require it from within the `realm` module. The `nan` module prints its path so it can be captured by an external build system like in this patch.
`StringData::data()` returns the underlying buffer, which might not be null-terminated, so the resulting string from appending to `"class_"` has garbage in the end. Use the `std::string` conversion of `StringData` because it takes the size of the string into account.
Also includes the following minor changes:
* Renames `ReadOnlyPropertyValueException` to
`ReadOnlyPropertyException` since it's the property that's read-only,
not the value.
* Elminiates some unnecessary copies of arguments passed to the
exception constructors.
* Makes the exception type data members public, since otherwise there's
no point in storing them at.
This avoids the need to eagerly fetch the object schema in order to
construct a `List` or `Results`. Instead the work can be deferred until
the object schema is requested. Since `List` and `Results` never use the
object schema themselves this can avoid unnecessary work in some bindings.
`snapshot()` functions are a better fit for what realm-js needs. The new
API also makes it clearer that the liveness of a given `Results`
cannot change at arbitrary times. Changing the liveness at arbitrary
times was not safe and could result in incorrect behavior, such as a
non-live `Results` changing.
There seems to a bug inside the `strip` command (or its `ld64` library) that is being triggered by some users who get this message: `Assertion failed: (0 && "need to handle arm64 -r reloc")`
It is reported that disabling this setting side-steps the issue, and we don't need any stripping done on the static library itself.
Fixes#503
By default, React Native projects target a minimum of iOS 7, but our projects were set to iOS 8 because previously we used a dynamic framework.
When building for an iOS 7 minimum target version, the linker would complain about thread local variables not working on iOS 7. So it was necessary to add even more checks (iOS >= 8 or OS X >= 10.7) before using native thread local variables rather than our shim.
This also changes our projects to target iOS 7 so we can catch these issues quicker in the future.
This will need to be updated again with a newer version since 1.1.2 build was created on CI but is not actually on static.realm.io until a new release is made.
It uses the Android ExternalCommitHelper implementation.
The Dockerfile requires that you have a static library build of core for Linux with PIC enabled.
After a thread is destroyed, a thread id might be reused. This is true for pthread_self() and hence std:🧵:id. We were hitting this in Chrome debug mode because the "worker" thread was being destroyed and a new one immediately created when reloading. The thread id would be the same, and therefore we'd get back the SharedRealm for the previously destroyed thread (which had yet to be garbage collected.
The new implementation uses an atomically incremented, thread-local identifier, which serves our needs very well.
Fixes#473
The base ClassDefinition defines its constructor member as null, but some subclass have a static constructor function. This allows both cases to be handled without a warning.
The main motivation for this is that building error messages via string
concatenation is tedious and makes it hard to judge what the actual message
looks like when there are a lot of parts being inserted, to the extent that
I've been tempted to leave out some potentially useful information because the
code was getting unwieldy.
It also has some small functional benefits: bools are printed as true/false
rather than 1/0, and it is optimized for minimizing the compiled size.
Currently it cuts ~30 KB off librealm-object-store.dylib even with the addition
of new functionality.
constructors.
This avoids having to repeat the move constructor's logic in the move
assignment operator, and allows the copy assignment operator to compile
despite `TableViewBase`'s missing copy assignment implementation. the
copy assignment implementation can be defaulted once `TableViewBase` is
fixed.
The custom implementation was an attempt to ensure that
`m_has_used_table_view` and `m_wants_background_updates` had appropriate
intial values. Thomas pointed out that we can remove the reliance on the
initial values by ensure that `prepare_async` sets
`m_wants_background_updates`, removing the need for a custom copy
constructor.
* object-store/master:
Make Realm::is_closed() work on read-only instances
Fix for object store build issues on Android
Implement commit helper and realm notifier for Android (#69)
Build both dynamic and static libraries.
moving pegtl dependency as step 1 in install doc
update docu (install procedure)
`Realm`, and moving from a `Results` to not result in a use-after-free.
The compiler generated move-assignment operator resulted in `m_notifier`
being assigned to without first calling `CollectionNotifier::unregister`.
This left a retain cycle in place, causing the `Realm` and other objects
to leak.
`ResultsNotifier` keeps track of which `Results` it should update when a
new `TableView` becomes available. When `Results` move-assignment
operator and move-constructor transfer ownership of the
`ResultsNotifier` to a new instance they also need to update its target
so it won't attempt to update the moved-from `Results`.
The Xcode project generated by CMake doesn't create the dylib when built
as the dylib target does not contain any source files. Adding an empty
placeholder .cpp file to the target is sufficient to convince Xcode to
produce the dylib.
Yavor Georgiev
Yavor Georgiev
Mark Rowe
Marius Rackwitz
blagoev
astigsen
Adam Fish
Adam Lebsack
kristiandupont
Thomas Goyne
Radu Tutueanu
Radu Tutueanu
Ari Lazier
Radu Tutueanu
Søren Vind
Yavor Georgiev
Thomas Goyne
Scott Kyle
Kenneth Geisshirt
Mark Rowe
Chen Mulong