typelevel/cats-effect v2.1.0 on GitHub

This is the second major release in the Cats Effect 2.x lineage. It is fully binary compatible with all 2.0.x releases, and should be source compatible in all but a few rare edge cases. It is also the first Cats Effect release which is not published for Scala 2.11. Cross-builds are available for 2.12 and 2.13 only.

Notable New Features

New SyncEffect typeclass: analogous to Effect, but for non-asynchronous datatypes such as SyncIO
Concurrent#background function, similar to start, but with resource handling wrapped around the Fiber to ensure it is properly scoped and cannot leak
Resource is now covariant in both its F[_] and A parameter types. This makes it considerably more convenient to use in conjunction with subtype encodings. A common example:
```
if (test)
  Resource.liftF(IO.pure(Some(42)))
else
  Resource.liftF(IO.pure(None))
```
As of this change, the above will now correctly infer Resource[IO, Option[Int]], as you would expect. Note that this change is source incompatible under some very specific circumstances (especially when using import cats.implicits._). For more discussion on this, see the pull request.
There is now a Parallel instance for Resource (for any F where Parallel[F]). This makes it possible to safely run resource acquisition in parallel with the par-combinators (e.g. parZip).
Scala.js 1.0 support. The Scala.js ecosystem is approaching its 1.0 milestone. Cats Effect is now cross-published for both 0.6.x and 1.0.x. The 0.6.x cross-build will be removed at some point after 1.0.x becomes relatively widespread.

Notable Bug Fixes

Eliminated the (unintentional) async boundaries generated by IO#bracket, which in turn corrupted SyncIO's guarantees regarding purely synchronous execution
Actions sequenced within Blocker will now properly exit when fatal errors are raised, mirroring the behavior of actions sequenced within the default ContextShift
Under some circumstances, the use action within an IO#bracket could be evaluated even if the acquire is canceled. Obviously, acquisition cannot be interrupted, but when cancelation occurs during acquisition, the intended semantics are that use is simply ignored and the release is run immediately. That wasn't happening, and has now been corrected.

User-Facing Pull Requests

#638 - Don't create async boundaries for SyncIO#bracket (@LukaJCB)
#682 - Remove ambiguity in Clock derivation (@bplommer)
#697 - Add a derivation for Timer[Resource[F, ?]] whenever Timer[F] is in scope (@barambani)
#690 - Add Blocker.blockOnK (@nigredo-tori)
#694 - Wrap default Blocker EC with exitOnFatal (@danxmoran)
#713 - Drop support for Scala 2.11.x (@LukaJCB)
#622 - Added SyncEffect typeclass (@LukaJCB)
#640, #719 - start as Resource - background (@kubukoz)
#722 - Indexed blocker thread names JVM (@Slakah)
#729 - Add Align instances for IO, SyncIO and IO.Par (@LukaJCB)
#706 - Bracket: Do not evaluate "use" if "acquire" is canceled (@Avasil)
#731 - Make Resource covariant in F and A (@7mind)
#742, #766 - Parallel resource (@SystemFw)
#753 - Support ScalaJS 1.0.x (@cquiroz)
#674, #685, #704, #665, #730, #761 - Documentation improvements/corrections/awesomeness (@kluen, @mlrv, @bwignall, @xuwei-k san, @qingwei91, @arosien)

Other contributors: @iRevive, @takayahilton, @jhnsmth, @travisbrown

Special thanks to each and every one of you!