[HOpenGL] Re: FFI Addendum, CVS Id 1.19

Manuel M. T. Chakravarty chak at cse.unsw.edu.au
Thu Jan 10 04:37:03 EST 2002

Sven Panne <Sven.Panne at informatik.uni-muenchen.de> wrote,

> The addendum looks fine for me, well done Manuel! But as usual, I've got a few
> minor points:
>    * Section 3.4 (export declarations) says:
>         "If an evaluation triggered by an external invocation of an exported
>          Haskell value returns with an exception, the system behaviour is
>          undefined."
>      So far, so good... It continues:
>         "Thus, Haskell exceptions have to be caught within Haskell and
>          explicitly marshalled to the foreign code."
>      I understand the idea here, but what about System.exitWith and
>      System.exitFailure in callbacks? GHC implements this via throwing an
>      exception, but this is an implementation detail IMHO. The consequence
>      of this implementation choice is a fatal error (uncaught exception)
>      when exit{With,Failure} is evaluated in callbacks. So there are two
>      questions:
>         * Should we allow exit{With,Failure} in such circumstances?
>           (My opinion: yes)
>         * If the answer is no, what is the "official" way to terminate
>           cleanly from callbacks? Note that in the context of HOpenGL
>           the callback can't return a value due to the nature of an
>           external lib (GLUT). I fear that the answer could be hs_exit(),
>           but that would lead to a backwards compatibility hell...

I really think that this is GHC's problem.  What the FFI
report says is that the *user* cannot expect to get an
exception properly marshalled across a callback.  If GHC
uses exceptions internally to implement features, which H98
doesn't define to be based on exceptions, then it is GHC's
responsibility to handle these "special" exceptions in a
special way.  Overall, I guess, GHC needs to check - at
least - for exit{With,Failure} failures at callback
boundaries and handle them appropriately.

>    * The example in section 5.4.2 should use "wrapper" instead of the old f.e.d.



More information about the FFI mailing list