Green Card and Exceptions

Matthew Donadio m.p.donadio@ieee.org
Fri, 13 Jun 2003 12:36:06 -0400


Alastair Reid wrote:
> The %fail statements (described in the last few paragraphs of
> http://www.haskell.org/greencard/downloads/greencard-latest/type-sig.html)
> consist of two C expressions.  For example:
> 
>     %fail {f == NULL} {errstring(errno)}
> 
> The first is a test for failure.
> The second is an expression which returns a C string.
> 
> If the test expression fails, the string expression is evaluated and used to
> generate a UserError.

OK, I think I misread the manual.  Sect 7.6 talks about functions with
side effects, so I assumed that the function had to have type (IO a) to
use %fail. 
 
> It would also be nice to be able to generate a different error instead of
> UserError.  We'd need to specify the type and the exception constructor so a
> plausible syntax would be:
> 
>   %fail {f == NULL} (UserError (string {errstring(errno)}))
> 
> [Detail: should it be a Haskell98  IOError constructor or a non-standard but
> widely implemented exception constructor?   Should it be a function or a
> constuctor?]

In the case of what I am doing, I'm not sure if IOError really make
sense philosophically.  The failures I need are underflow, overflow,
loss of precision, etc.  Since IOError is a type synonym for
IOException, then perhaps accepting an Exception constructor is
appropriate.

To keep compatibility with old libraries it may be wise to keep %fail as
is, and have a new directive %throw that accepts an Exception
constructor, and uses either throw or throwIO.

On the other hand, now that I know that I can use %fail with pure
functions, I can make that work.
 
> As GreenCard maintainer, I've got to ask:
> - How many users of GreenCard are still out there?

New GreenCard user.  In my case, I need access to C land for typedefs
and macros.  I could write my own stubs, but GreenCard saves me this
step.

> - Are you developing new libraries or just maintaining the ones you've got?

New library.

> - Is there a demand for new features?

A more generic %fail mechanism?

Thanks for the response.

-- 
Matthew Donadio (m.p.donadio@ieee.org)