Catching Exceptions in Haskell
Hi all, I'm struggling to figure out how to catch an exception in Haskell. Eventually, what I want to do is to use the Control.Macho library (from Hackage) to parse binaries on Mac OS X. Naturally, I would like my program to gracefully handle corrupted files or files that aren't binaries. So I need to catch exceptions if the parsing with Data.Macho fails... This is what I've got so far: http://hpaste.org/fastcgi/hpaste.fcgi/view?id=1877#a1877 Digression: Note that if I leave out the `foo e` call, then I get scratch.hs:21:9: Ambiguous type variable `e' in the constraint: `Exception e' arising from a use of `try' at scratch.hs:21:9-11 Probable fix: add a type signature that fixes these type variable(s) I find this somewhat counter-intuitive; this is also not discussed in the book "Real-World Haskell" (at least I haven't found it), which is what I'm using to learn Haskell. What is the best way to get around this? Back to my problem. The program produces the following output: $ runghc scratch.hs Testing Pure... Caught a bad exception: abort via error Testing Macho... No exception occured, the result is: scratch.hs: too few bytes. Failed reading at byte position 4 This counter-intuitive in at least two ways: 1) How can it fail at position 4 in an empty string? That looks like a sloppy error message too me; why doesn't it report the actually failing position? 2) Why can I catch an exception raised with an error in my own code, but not the one raised by Data.Binary.Get? Looking at http://hackage.haskell.org/packages/archive/binary/0.5/doc/html/src/Data-Bin... , an exception within Data.Binary.Get is also raised with error. What is the difference? Why am I seeing what I'm seeing, and how should I be doing it in the first place? Thanks in advance, Bjoern
Ok, so I was able to extract a simpler program with the same symptoms. http://hpaste.org/fastcgi/hpaste.fcgi/view?id=1900#a1900 Apparently the exception is only triggered when m is used in line 31. So I guess that evaluate in line 24 is not causing strict evaluation of the Data.Binary.Get monad, even though it is working for the error and assert tests. Why is that? Is that expected? If so, what am I doing wrong? Thanks, Bjoern
The 3rd exception is raised when (show m) is computed, since no exception occurred when evaluating (testGet L.empty) itself. When you call "evaluate a", it only forces "one level" of evaluation of a. E.g. if you would run
evaluate (1:error "2":[]) >>= print . head
this would print 1, and not give an error. You can maybe understand this better when you change
(res :: Either SomeException String) <- try $ evaluate (f x)
into
(res :: Either SomeException String) <- try $ return (f x)
Now the 2nd exception won't be raised either: evaluate digged one level into assert, and that causes an exception, but return will not do that, it will just return the unevaluated thunk that wraps the assert computation, and this thunk will get evaluated when you print it You can get the behavior what you expect by using some functions from Control.Parallel.Strategies. Change (res :: Either SomeException String) <- try $ evaluate (f x) into (res :: Either SomeException String) <- try $ evaluate (f x `using` rnf) Now all 3 will throw an exception, as you expected. Why? (a `using` rnf) reduces its argument (a) to "head normal form", which basically means it evaluating all sub-expressions at all levels. So
evaluate (1:error "2":[] `using` rnf) >>= print . head
will now give an error.
Cheers,
Peter Verswyvelen
On Sun, Mar 1, 2009 at 1:48 AM, Bjoern Brandenburg
Ok, so I was able to extract a simpler program with the same symptoms.
http://hpaste.org/fastcgi/hpaste.fcgi/view?id=1900#a1900
Apparently the exception is only triggered when m is used in line 31. So I guess that evaluate in line 24 is not causing strict evaluation of the Data.Binary.Get monad, even though it is working for the error and assert tests.
Why is that? Is that expected? If so, what am I doing wrong?
Thanks, Bjoern _______________________________________________ Beginners mailing list Beginners@haskell.org http://www.haskell.org/mailman/listinfo/beginners
participants (2)
-
Bjoern Brandenburg -
Peter Verswyvelen