[Peter Graves <peter@ar...>]

On Sun, 30 Oct 2005 at 13:40:05 +0100, Andras Simon wrote:
> There are a few things that I don't understand about this. Why don't
> conditions that are subclasses of JAVA-EXCEPTION inherit the cause
> slot?

They do:

    CL-USER(1): (define-condition throwable (java-exception) ())
    THROWABLE
    CL-USER(2): (signal 'throwable)
    Debugger invoked on condition of type THROWABLE:
      #<THROWABLE {374956}>
    Restarts:
      0: TOP-LEVEL Return to top level.
    [1] CL-USER(3): (inspect *debug-condition*)
       0 LAYOUT -----------> #<SYSTEM:LAYOUT {1FC1B56}>
       1 CAUSE ------------> #<UNBOUND>
       2 FORMAT-CONTROL ---> #<UNBOUND>
       3 FORMAT-ARGUMENTS -> #<UNBOUND>
    [2i] CL-USER(4): 

> And where does the message disappear? It is passed to signal, as
> in
>
>   signal(condition, new SimpleString(getMessage(t)))

Yes, but that just passes the message to the real SIGNAL, defined in
signal.lisp:

    public static final LispObject signal(LispObject condition, LispObject message)
        throws ConditionThrowable
    {
        return Symbol.SIGNAL.execute(condition, Keyword.MESSAGE, message);
    }

I don't think the :message keyword is used anywhere else, so this
doesn't make much sense, and that signal() function should go away.

The underlying problem is that conditions don't have a MESSAGE slot.

The Java class Condition has a message field:

    protected String message = null;

But that's very old, and dates back to a time when ABCL's conditions
weren't CLOS objects. That field should probably go away too, but there
are vestiges of it in a number of places.

The right way to do things nowadays is to use the FORMAT-CONTROL slot
(via the :format-control initarg), which all instances of subclasses of
CONDITION have.

That said, there's clearly a bug in this area:

    CL-USER(21): (define-condition test-error (program-error) ())
    TEST-ERROR
    CL-USER(23): (error 'test-error :format-control "this is a test")
    Debugger invoked on condition of type TEST-ERROR:

    Debugger invoked on condition of type TYPE-ERROR:
      The value #<UNBOUND> is not of type LIST.
    Restarts:
      0: TOP-LEVEL Return to top level.

I'll try to fix that later today.

> Is any of this related to my messing with exceptions?

I don't think so, since the problem occurs with conditions derived from
PROGRAM-ERROR too.

> Ideally, what I think should happen is that any condition that is a
> subtype of JAVA-EXCEPTION has a CAUSE and a MESSAGE slot (with readers
> and all), and they get filled in when the condition is signaled.

The CAUSE slot should be there, and JAVA:JAVA-EXCEPTION-CAUSE is the
reader for it.

As mentioned, there is no MESSAGE slot.

> How could I arrange this?

I'll let you know after I figure out what the bug is.

Stay tuned... :)

-Peter