You can clone with
HTTPS or Subversion.
Replace some uses of atomicModifyIORef with STM
This seems to cure the problem of the thread-delay benchmark behaving
pathologically due to contention on IORef blackholes at times.
Committer: Bryan O'Sullivan <firstname.lastname@example.org>
A simple static HTTP server.
Make it easier to compare the old and the new I/O manager in benchmarks
Don't enqueue a zero timeout. Just execute the callback immediately.
Also, make sure that the thread-delay benchmark doesn't accidentally use a
Take the number of threads from the command line
Added a new benchmark for threadDelay
Scales much better with the new I/O manager. 20,000 threadDelay calls
take 0.577s (vs. 22.558s using the old I/O manager).