Re: [Chicken-users] 1095 (prerelease) + some eggs

[felix]

Joerg F. Wittenberger wrote:
> Sorry
>
> to follow up myself.  Stupid editing mistake.  Here the actual source
> code I handed the compiler.
>
> (define (askemos:run-tcp-server
>          host port connection-handler
>          maximum-semaphore name)
>   (define request-queue (tcp-listen port))
>   (define (exception-handler ex)
>     (receive
>      (title msg args rest) (exception->fields ex)
>      (logerr "run-tcp-server (host ~a port ~a) ~a ~a ~a\n"
>              host port title msg args)))
>   (let loop ()
>     (handle-exceptions
>      ex (exception-handler ex)
>      (receive
>       (in-port out-port) (tcp-accept request-queue)
>       (receive
>        (local remote) (tcp-addresses in-port out-port)

                                               ^
`tcp-addresses' takes only a single argument. Normally, your
exception-handler should logerr something. Are you compiling in
safe mode?

>        (if maximum-semaphore (semaphore-wait maximum-semaphore))
>        (thread-start!
>         (make-thread
>          (lambda ()
>            (handle-exceptions
>             ex (exception-handler ex)
>             (connection-handler in-port out-port remote)
>             (if maximum-semaphore (semaphore-signal maximum-semaphore))
>             (close-input-port in-port)
>             (with-exception-handler
>              ex (begin (close-output-port out-port) (loop)))))

Argh! Very wrong and very dangerous: the first arg to `with-exception-handler' 
must be
the handler itself and the second arg has to be a thunk. And then it may not 
simply
call `loop': that's in a different thread!
Don't use `with-exception-handler'. Use `handle-exceptions', it's safer.

>            name))))
>        (loop))))

One question: SRFI-18 says that an uncaught exception is stored
in the end-exception field of a thread and that the thread should
terminate. If the parent now join!s,
that uncaught exception will be raised. The problem is that errors
in child-threads are not reported to the user, which makes debugging
challenging, but not exactly fun.

Should the default exception handler still invoke `error'? No, because that 
would
throw back to toplevel (without storing something in the end-
exception field).

So what should it do?

1) Terminate silently, and store the end-exception
2) If not the primordial thread: Print a message to stderr and store it,
   otherwise invoke `error'
3) something else


cheers,
felix