349767a7b6
Type inference seems to fall down with the rank-n type of `forkWithUnmask` + `const`: Control/Monad/Conc/Class.hs:525:27: error: • Couldn't match type ‘b5’ with ‘forall a. m a -> m a’ Expected: (forall a. m a -> m a) -> m () Actual: b5 -> m () Cannot instantiate unification variable ‘b5’ with a type involving polytypes: forall a. m a -> m a • In the first argument of ‘forkWithUnmask’, namely ‘(const ma)’ In the expression: forkWithUnmask (const ma) In an equation for ‘fork’: fork ma = forkWithUnmask (const ma) • Relevant bindings include ma :: m () (bound at Control/Monad/Conc/Class.hs:525:6) fork :: m () -> m (ThreadId m) (bound at Control/Monad/Conc/Class.hs:525:1) | 525 | fork ma = forkWithUnmask (const ma) |
||
---|---|---|
.. | ||
Control | ||
CHANGELOG.rst | ||
concurrency.cabal | ||
LICENSE | ||
README.markdown | ||
Setup.hs |
concurrency
A typeclass abstraction over much of Control.Concurrent (and some extras!). If you're looking for a general introduction to Haskell concurrency, you should check out the excellent Parallel and Concurrent Programming in Haskell, by Simon Marlow. If you are already familiar with concurrent Haskell, just change all the imports from Control.Concurrent.* to Control.Concurrent.Classy.* and fix the type errors.
A brief list of supported functionality:
- Threads: both unbound and bound.
- Getting and setting capablities.
- Yielding and delaying.
- Mutable state: STM,
MVar
, andIORef
. - Atomic compare-and-swap for
IORef
. - Exceptions.
- All of the data structures in Control.Concurrent.* and Control.Concurrent.STM.* have typeclass-abstracted equivalents.
- A reimplementation of the async package, providing a
higher-level interface over threads, allowing users to conveniently
run
MonadConc
operations asynchronously and wait for their results.
This is quite a rich set of functionality, although it is not complete. If there is something else you need, file an issue!
This used to be part of dejafu, but with the dejafu-0.4.0.0 release, it was split out into its own package.
Why this and not something else?
-
Why not base: like lifted-base, concurrency uses typeclasses to make function types more generic. This automatically eliminates calls to
lift
in many cases, resulting in clearer and simpler code. -
Why not lifted-base: fundamentally, lifted-base is still using actual threads and actual mutable variables. When using a concurrency-specific typeclass, this isn't necessarily the case. The dejafu library provides non-IO-based implementations to allow testing concurrent programs.
-
Why not IOSpec: IOSpec provides many of the operations this library does, however it uses a free monad to do so, which has extra allocation overhead. Furthermore, it does not expose enough of the internals in order to accurately test real-execution semantics, such as relaxed memory.
Contributing
Bug reports, pull requests, and comments are very welcome!
Feel free to contact me on GitHub, through IRC (#haskell on freenode), or email (mike@barrucadu.co.uk).