@nativescript/android - npm Package Versions

8.5.1 (2023-07-24)

Bug Fixes

• Avoid setting the same property on an ObjectTemplate twice (9e610c8)

• Don't access super property on implementation object (d8b8bc0)

• Don't iterate properties in GetImplementedInterfaces (9dfae65)

• intermediary fix for https://github.com/NativeScript/android/pull/1771 (32c7abb)

• Leave context after Runtime::PrepareV8Runtime() (cd1d285)

• memory leak on accessing static interface methods (88ce2d8)

• memory leak on saving code cache (6d416a1)

• Update common-runtime-tests-app (c8db3ca)

• update legacy android version in package.json (#1744) (b4ad8e5)

• Use Isolate::TryGetCurrent() (afe026a)

• Remove weak callback from __postFrameCallback cache (#1755) (ff1b979), closes #1755

Features

• add support for kotlin 1.8 (#1765) (1a928e4)

Performance Improvements

• avoid unnecessary string copying when calling static properties of interfaces (8b53d02)

BREAKING CHANGES

• __startNDKProfiler() and __stopNDKProfiler() global bindings no longer available.

The NDK profiler was not functional, since nothing in the build process defined the NDK_PROFILER_ENABLED preprocessor symbol. The start and stop functions were already no-ops.

• chore: Remove outdated comment

RunMicrotasks no longer exists, it's already been replaced in the code with PerformMicrotaskCheckpoint.

• chore: Use unique_ptr for NewBackingStore in byte buffers

V8 doc: https://docs.google.com/document/d/1sTc_jRL87Fu175Holm5SV0kajkseGl2r8ifGY76G35k/edit

The V8 usage examples show unique_ptr here; it probably doesn't matter much, but we don't need the backing store after creating the ArrayBuffer, and I believe shared_ptr is slightly more overhead than unique_ptr.

For convenience, replace the manual empty deleter for direct buffers with v8::BackingStore::EmptyDeleter.

• chore: Remove weak finalizer callback from __postFrameCallback()

Weak finalizer callbacks are going away in V8 11.x, so we have to remove this one. Luckily, a finalizer callback is not necessary - it's never needed to prevent the frame callback from being collected.

However, a weak callback is not necessary in the first place. We can just clean up the cache entry after the callback is executed, since it is only executed once.

Note that the call to erase() destructs the FrameCallbackCacheEntry instance, making entry a dangling pointer, so don't use it after the erase(). There's probably a safer way to do this, although the way that first occurred to me (pass the key to the callback instead of the entry, and then use std::unordered_map::extract()) is not available on robin_hood::unordered_map.

• fix: Make sure frame callbacks are not ignored

There was a bug where __postFrameCallback() would not always cause the callback to be called. Without initializing removed, sometimes it would have a nonzero value, so the callback would be ignored.

• chore: Clear callback caches' persistent handles in destructor

Clearing the persistent handles in the destructor makes it a bit easier to deal with the cache entry's lifetime: they are cleared whenever the entry is removed from the cache.

We do this for both the main thread callback cache and the frame callback cache.

Adding a destructor makes the cache entries non-movable. But the only place where they were moved was when inserting them into the cache anyway. We can use C++17's try_emplace() method to insert them without having to move them.

• chore: Construct FrameCallbackCacheEntry with ID

This avoids the situation of forgetting to add an ID to the cache entry.

• chore: Improve usage of unordered_map APIs in CallbackHandlers

This fixes a few places where we can avoid double lookups:

• In RunOnMainThreadFdCallback, we already have a valid iterator, so no need to look up the same key again in RemoveKey (this is the only usage of RemoveKey, so we can remove it.) (Additionally, we probably want to remove the cache entry before throwing a NativeScript exception.)

• In PostFrameCallback and RemoveFrameCallback, we should not do contains() immediately followed by find().

• chore: Fix runtime typo

• chore: Ignore main thread and frame callback return values

We don't do anything with the return value from these callbacks, so it's OK to ignore them and not convert them to a local handle.

8.5.0 (2023-06-27)

Bug Fixes

• add semicolon after console type (32259a9)
• always log console messages and uncomment live sync (c0f5514)
• Compile as C++17 (221a9c2)
• Correctly initialize context in inspector client init() (0bc0480)
• drain microtasks after devtools message (4834a2b)
• Implement console.log inspector with Runtime protocol (2c4337b)
• memcpy array data for non-direct java bytebuffers (#1740) (1c0214a)
• multi threading on MethodResolver (bc8bc52)
• possible infinite loop and memory leak in metadata reader (#1749) (c2c8aa8)
• Re-enable inspector code (f357ce6)
• refactor console.log implementation a bit (#1741) (d3c52cb)
• remove free of non-owned jni buffer (81806b3)
• Remove use of DISALLOW_COPY_AND_ASSIGN macro in inspector (6da1a6b)
• Remove use of V8InspectorPlatform (1f2b202)
• Restore DOM and Network callback handlers (35689a7)
• uncomment abifilters (f5a5434)
• update js arg conversions (0640fce)
• use min sdk 17 (2f2358c)

Features

• add console message type prefixes (9a10e2b)
• initial support for __inspectorSendEvent (233b7c3)
• native timer polyfills (#1733) (3415e5c)
• remove old WeakRef polyfill (97d7465)
• update all archs to use v8 10.3.22 (a68b057)
• use v8 static (ee8a521)

Performance Improvements

• refactor arrays marshaling (84e7ddb)
• remove instantiations and cache runtime in isolate (f23c1bb)
• update old args converter with array marshaling optimizations and use this converter again tempporarily (b5218c9)
• use vector pointers for passing parsed method signatures (aa623b7)