Finali[zs]ers

[Manuel M T Chakravarty]

"Simon Marlow" <simonmar at microsoft.com> wrote,

> I've added a finalizer rationale document to the repository, under
> haskell-report/ffi (i.e. the same place as the FFI spec).  If you have
> CVS commit privs, please feel free to modify it, or otherwise suggest
> changes on this list.

Thanks a lot for the summary.  To be honest, by now I am
convinced that the Right Thing is to revert back to Haskell
finalizers.  Reasons:

* 2 out of 3 systems regard Haskell finalizers as a
  non-issue to implement and there exists a prototype
  implementation for the 3rd system.

* Everybody except Alastair seems in favour of reverting
  back to Haskell finalizers.

* Alastair's main counter argument appears to be 

  - potential to introduce race conditions
  - potential for deadlock

  We need to place a side condition on what a Haskell
  finalizer may do (ie, not update IORef's) in the same way
  as we would have to place side conditions on what a C
  finalizer may do (ie, not call back into Haskell).

So, I am going to revert the spec back to Haskell finalizers
for the next RC.  I agree with Alastair that we need to add
to the spec at the same time a precise statement about what
Haskell finalizers are allowed to do (or not do).

Manuel