Re: [Haskell-cafe] ANN: asynchronous-exceptions

Hi João, * João Cristóvão [2014-02-05 15:53:34+0000]

Hi Roman,

One other question: in Michael's article, he also mentions the possibility of: "Consider that, for some strange reason, we decided to asynchronously throw an IOException to a worker thread"

Meaning that you can throwTo any exception asynchronouly, even if it is not a SomeAsyncException. At least from what I gathered from (GHC Head) throwTo code, this does not seem to be a restriction.

Thus, your module will work for any well behaved code that does the right thing, and respects the SomeAsyncException superclass, but not for other spurious asynchronous exceptions, right?

Precisely.

I'm just trying to frame the possible use cases of your library. Thanks Joao

2014-02-05 Roman Cheplyaka :

...

It works as expected.

With GHC 7.8 it doesn't print the timeout exception.

With earlier GHC it does (again, as expected, because the timeout exception isn't marked as asynchronous). In practice one should rarely want to use System.Timeout anyway (because of the overflow issue), and I'm going to patch one of the better timeout packages (such as unbounded-delays) to support asynchronous-exceptions.

* Michael Snoyman [2014-02-05 16:56:31+0200]

...

I don't think this package works as expected. Consider the following:

import Control.Concurrent import Control.Exception.Async import System.Timeout

main :: IO () main = do timeout 1000000 $ do threadDelay 10000000 `catchSync` \e -> do print e threadDelay 10000000 return ()

The expected behavior would be that the timeout- an async exception- would kill the thread delay, the catch would ignore the async exception, and the program would exit. In reality, catchSync treats the timeout as a synchronous exception, prints it, and delays once again. Compare this to classy-prelude's catchAny, which handles the situation correctly, via the technique I described in "Catching all exceptions."[1]

In this case, the issue is that the timeout exception type is not recognized as async, and a special case could be added to handle that exception type[2]. However, I think the overall approach of determining *how* an exception was thrown based on *what* was thrown is not tenable.

[1] https://www.fpcomplete.com/user/snoyberg/general-haskell/exceptions/catching... [2] It's a bit difficult to do so, since IIRC the type is never exported. But a hack using the Typeable instance- while ugly- is likely possible.

On Wed, Feb 5, 2014 at 1:28 PM, Roman Cheplyaka wrote:

...

The links are:

http://hackage.haskell.org/package/asynchronous-exceptions https://github.com/feuerbach/asynchronous-exceptions

* Roman Cheplyaka [2014-02-05 13:23:38+0200]

...

It is often useful to distinguish between synchronous and asynchronous exceptions. The common idiom is to run a user-supplied computation catching any synchronous exceptions but allowing asynchronous exceptions (such as user interrupt) pass through.

base 4.7 (shipped with GHC 7.8) will have SomeAsyncException type that solves this problem.

asynchronous-exceptions is a new package that serves two purposes: * provide compatibility with older `base` versions that lack the `SomeAsyncException` type * define convenient functions for catching only synchronous exceptions

Roman

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

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