Skip to content

v1.0.0

Compare
Choose a tag to compare
@neilcsmith-net neilcsmith-net released this 05 Apr 17:53
· 90 commits to master since this release

This is the first full release of gst1-java-core v1. The bulk of the work leading up to this release has been finalising, stabilising and documenting the API outside of the lowlevel package, completing the work of porting the original 0.10 bindings.

Key changes

Changes in beta 1

  • All classes have been audited, non-functional legacy methods removed or replaced, documentation updated and links added to upstream GStreamer documentation where appropriate. Any methods exposing lowlevel classes have been removed or rewritten.
  • Minimum requirements have been updated to Java 8 and JNA 5.2.
  • Version support has been added to Gst.init(). The minimum GStreamer requirement remains 1.8, although this is now enforced. If you require features that require a later version of GStreamer, you must use the version of Gst.init(...) that takes a Version or these mappings will be disabled. eg. Gst.init(new Version(1, 14)). There is also an annotation @Gst.Since(minor = 14) for use on classes, methods and constants that require later versions - this annotation is included in the JavaDoc.
  • Non-functional legacy classes have been removed, including all of the controller package and various interface classes.
  • GObject, GError and various GIO related classes have been moved into the GLib package. This is an initial step towards separating out GStreamer and GLib bindings into separate libraries in future.
  • NativeObject and various other lowlevel classes have been moved out of lowlevel and into the GLib package as official API.
  • Event, Message and Query, as well as related classes have been moved into their respective packages, partly to improve encapsulation.
  • WebRTC classes have been moved into their own package.
  • Enums have been rewritten around NativeEnum and NativeFlag interfaces (in GLib package). Methods accepting int flags (eg. SeekFlags) now take EnumSet and related int constants are now enum.
  • Methods accepting or returning ClockTime now take or return a long. This is for efficiency and consistency - GLib does not give all type information required to be consistent here. NB. the value is an unsigned 64-bit integer value in nanoseconds. ClockTime remains as a collection of utility methods for working with these values. It still doesn't (yet) correctly handle "negative" values.
  • New memory handling and removal of finalization. A new NativeObject.Handle type wraps all native pointers, and subclasses implement all ref, unref, free, etc. Handle subclasses should be used to call different native functions in these cases. Memory handling is a cleaner-like mechanism, but using the existing weak reference cache rather than additional phantom references. In the past, only GObject instances were cached to ensure the same NativeObject was returned for any pointer - this now applies to all NativeObject subclasses. All garbage collection tests pass with the new code, but there may be possibility for regressions caused by slightly different behaviour.
  • A new Natives utility class provides access to methods required for translating to and from Java objects and native representations where required, including translation into a GPointer (see below) or raw Pointer. Natives.objectFor() methods replicate functionality previously in NativeObject to get the Java representation of a native reference. These should be used in all cases in favour of NativeObject.Initializer except where the native call is a constructor.
  • New extension mechanism to register additional NativeObject subclasses (it was theoretically possible to achieve this before, but with major caveats and unsupported). See NativeObject.TypeRegistration and Natives.registration(..). Instances of NativeObject.TypeProvider should be made available by the standard Java ServiceLoader mechanism. Internal types use the same mechanism, reducing uses of reflection in the library.
  • GPointer and subclasses have been added to the lowlevel package. These are subclasses of JNA's PointerType and will form the basis of native mapping in future. They are already required for Handle-related functionality. They may become part of the API in future. Using anything else in lowlevel will likely break from release to release. Lowlevel is no longer included in the JavaDoc.
  • Travis CI configuration added. All future PR's must pass before integration, and should include tests where appropriate. The current CI only covers GStreamer 1.8 (baseline), but will be extended to include testing on later GStreamer as well. Tests may need to be version aware if they target features higher than 1.8.

Additional changes from beta 2

  • FIX make sure GObject subclasses can be GC'd after explicit call to dispose() - fixes #146
  • Minor updates to NativeObject.Handle API and added documentation. The atomic reference and boolean fields are now correctly private (mistakenly left protected in beta-1). ownsHandle() renamed as ownsReference().
  • Move GObject signal and Pad probe native management into Handle to ensure the callbacks are removed when GC'd as before.
    • Behavioural change - all callbacks are now disconnected on explicit calls to invalidate() and dispose(). In general make sure to keep references to objects with callbacks and dispose of them explicitly - relying on the GC for this is not recommended.
    • The Pad method for adding an event probe with an int mask has been removed from the API.
    • Event probe testing has been fixed.
  • NativeObject now implements AutoCloseable. Defaults to calling dispose().
  • FIX getting a GObject subclass from a GValue will now return a caller-owned reference (switched to g_value_dup_object). Iterating objects, eg. by calling getPads() would return and cache unowned objects whose lifecycle was not correctly tracked. This might cause regressions if code relied on the old behaviour.
  • Bus disposal is now linked to Pipeline lifecycle. Bus instances will be GC'd even when not explicitly closed. Test for this no longer ignored.
  • Added a system property to detach callback threads (workaround for #147 where JNA isn't clearing thread proxies on macOS) - use -Dglib.detachCallbackThreads=true