[MLton] Infinite loop in the toplevel handler

[Matthew Fluet]

I was able to reproduce the bug.  It is also present on (at least)
x86-darwin.  I would seem to have something to due with the top-level
exception handler, but I haven't had a chance to investigate.

On Tue, Sep 14, 2010 at 5:45 AM, Nicolas Bertolotti <
Nicolas.Bertolotti at mathworks.fr> wrote:

>  Hello,
>
>
>
> In some cases, the toplevel handler may enter an infinite loop. This can be
> highlighted using the following sample code :
>
> structure Main =
>
> struct
>
>     fun main() =
>
>         let
>
>             val _ = MLton.Signal.setHandler(Posix.Signal.pipe,
> MLton.Signal.Handler.ignore);
>
>             fun loop(i) =
>
>                 (TextIO.output(TextIO.stdErr, (Int.toString(i)) ^ "\n");
>
>                  if i <= 0 then
>
>                      ()
>
>                  else
>
>                      (Posix.Process.sleep(Time.fromSeconds(1));
>
>                       loop(i - 1)));
>
>         in
>
>             loop(10);
>
>             OS.Process.success
>
>         end
>
>         handle _ => (TextIO.output(TextIO.stdErr, "Exception raised\n");
>
>                      OS.Process.failure)
>
>
>
>     val _ = OS.Process.exit(main());
>
> end;
>
>
>
> If you do the following (linux platform):
>
> $ mlton –output test test.sml
>
> $ ./test 2>&1 | tee log.txt
>
> And, in another console (within 10 seconds) :
>
> $ strace -p `ps -auxwww | grep test | grep -v grep | awk '{ print $2 }'` >
> toto.log 2>&1 & kill `ps -auxwww | grep tee | grep log | awk '{ print $2 }'`
>
> (in order to log a “strace” of the process in “toto.log” and kill the “tee”
> process so that the standard error of the process becomes unaccessible)
>
> the process then enters an infinite loop.
>
>
>
> The “strace” log shows the following messages:
>
> “
>
> Process 27026 attached - interrupt to quit
>
> restart_syscall(<... resuming interrupted call ...>) = 0
>
> write(2, "8\n"..., 2)                   = -1 EPIPE (Broken pipe)
>
> --- SIGPIPE (Broken pipe) @ 0 (0) ---
>
> write(2, "Exception raised\n"..., 17)   = -1 EPIPE (Broken pipe)
>
> --- SIGPIPE (Broken pipe) @ 0 (0) ---
>
> write(2, "unhandled exception: Io: outp"..., 86) = -1 EPIPE (Broken pipe)
>
> --- SIGPIPE (Broken pipe) @ 0 (0) ---
>
> write(2, "unhandled exception: Io: outp"..., 86) = -1 EPIPE (Broken pipe)
>
> --- SIGPIPE (Broken pipe) @ 0 (0) ---
>
> write(2, "unhandled exception: Io: outp"..., 86) = -1 EPIPE (Broken pipe)
>
> …
>
> “
>
>
>
> I have managed to fix this in my code (simply catch the exception when
> writing “Exception raised” in the “main” function’s exception handler).
>
> Anyway, it is still an issue that I may have on other binaries and some
> other users may also face.
>
>
>
> It appears that, when stderr is not accessible, the toplevel handler which
> attempts to write its own message generates an exception that causes the
> infinite loop.
>
>
>
> I tried to fix the issue in basis-library/mlton/exn.sml but I could not
> manage to eliminate the issue:
>
> -          The exceptions that may occur in the call to “message” seem to
> be properly handled
>
> -          If I add a global ref that I set to true the first time I enter
> the function and that prevents it from running again, it has no effect.
>
>
>
> Any thought?
>
>
>
> Thanks in advance
>
>
>
> Nicolas
>
>
>
>
>
> [image: cid:image001.gif at 01C82D2B.AFBEF470] <http://www.mathworks.fr/>**
>
>
>
> Accelerating the pace of  engineering and science<http://www.mathworks.fr/>
>
> *Nicolas Bertolotti*
> Senior Development Engineer
>
> 2 Rue de Paris
> 92196 Meudon Cedex
>
> France
>
> Nicolas.Bertolotti at mathworks.fr
>
> tel:
> fax:
>
> +33.1.41.14.88.55
>
> +33.1.55.64.06.64
>
>
>
>
>
> _______________________________________________
> MLton mailing list
> MLton at mlton.org
> http://mlton.org/mailman/listinfo/mlton
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mlton.org/pipermail/mlton/attachments/20100915/c64ffa4e/attachment.htm