I think 'shouldBeCaught' is more often than not the wrong thing.  A whitelist of exceptions you're prepared to handle makes much more sense than excluding certain operations.  Some common whitelists, e.g. filesystem exceptions or network exceptions, might be useful to have.

I like Ertugrul's suggestion a lot, however it seems a bit more invasive.  If we want to do that work, we could also restructure the wired-in exceptions so they're more hierarchical.  There are some top-level exceptions we can get by type, such as ArithException, but sadly many interesting and useful exceptions are lumped together under IOException.


On Wed, Jul 10, 2013 at 4:29 PM, Ertugrul Söylemez <es@ertes.de> wrote:
Michael Snoyman <michael@snoyman.com> wrote:

>     shouldBeCaught :: SomeException -> Bool
>
> One first stab at such a function would be to return `False` for
> AsyncException and Timeout, and `True` for everything else, but I'm
> not convinced that this is sufficient. Are there any thoughts on the
> right approach to take here?

I think there is no one right approach.  However, if you add such a
function to the exception library, it really belongs into the Exception
type class with the following type:

    shouldBeCaught :: (Exception e) => e -> Bool

However, a better approach is to have exception tags.  In most cases you
don't want to catch killThread's or timeout's exception, but you do want
to catch all error exceptions:

    data Tag = Error | Abort | TryAgain | {- ... -} | Other String
        deriving (Data, Eq, Ord, Read, Show, Typeable)

    instance IsString Tag where
        fromString t = Other t

This could then manifest in the following two functions in the Exception
type class:

    hasTag :: (Exception e) => Tag -> e -> Bool
    tagsOf :: (Exception e) => e -> [Tag]

Then exception catchers (functions that risk swallowing important
exceptions) could filter by type and tag.


Greets,
Ertugrul

--
Not to be or to be and (not to be or to be and (not to be or to be and
(not to be or to be and ... that is the list monad.

_______________________________________________
Haskell-Cafe mailing list
Haskell-Cafe@haskell.org
http://www.haskell.org/mailman/listinfo/haskell-cafe