ANN: SafeSemaphore, Proposal: replace QSem/QSemN

[Chris Kuklewicz]

There is a leak in base.  A leak in the semaphores.  Drip, drip, drip.

I would like to announce a new package SafeSemaphore, at
> http://hackage.haskell.org/package/SafeSemaphore
My intent is for this to supersede the implementation of semaphore in
Control.Concurrent.QSem and Control.Concurrent.QSemN currently in the
base package.

This is needed because there is a quantity loss bug in QSem and QSemN.
Any thread blocked while waiting might be killed.  In this case a later
signal will pass quantity to the dead thread.  This quantity is never
received and is thus lost to the program.  This is logged in GHC's bug
tracker at  http://hackage.haskell.org/trac/ghc/ticket/3160

The MSem and MSemN modules in SafeSemaphore fix this problem and are
extended replacement for QSem/QSemN with what I think is an improved and
extended API.  Under the hood a MSem/MSemN only ever allocates three
MVars, instead of one MVar for every blocking waiter that QSem/QSemN
allocate.

The related Haskell wiki page at
> http://haskell.org/haskellwiki/SafeConcurrent
shows the included TestKillSem.hs file and its output.  This shows how
the QSem/QSemN bug can be triggered and shows that MSem/MSemN is immune
to this leak.

Cheers,
  Chris Kuklewicz