Re: fhpd vs RuntimeExceptions

[Mark Wielaard <mark@klomp.org>]

Hi Andrew,

On Thu, 2007-11-15 at 14:05 -0500, Andrew Cagney wrote:
> That unfortunatly isn't sufficient; the old code in CLI.java was 
> differentiating between and NPE and other exceptions - dumping the stack 
> when an NPE occured.  I'll restore this behavior; so we're at least back 
> to a usable status quo (and from a HPD user prospective in a better 
> position - these back-traces plain suck)

OK, but I don't understand this heuristic with NullPointerException. You
introduced a "nasty()" method (cute name) that gobbles up the Exception
except if it is a NullPointerException or has an empty message. But that
doesn't seem to cover other catastrophic failures like
ClassCastException, ArrayOutOfBoundsException, ArithmeticException,
IllegalArgumentException or NumberFormatException, etc. that might or
might not have empty messages, but that are real core bugs if they occur
and "bubble up" to the CLI.

I think your argument that from a user perspective printing fatal
exceptions just plain sucks is right. But if there are really fatal
exceptions in the core code that then get hidden by the CLI layer we
(and we are also our own users) have a hard time fixing those. At least
our users will be unable to properly report what went wrong. So the real
solution is to just not plain suck, by not having bugs in our core
code :)

> I'll also add a prototype version of CLI.printError() that we discussed; 
> and have just EvalCommands use it; we can then refine it as needed.

That seems fine. I recommend you also just rip out the whole Message
queuing structure and handle that in the same way with a simple
printMessage(). Having two different mechanisms for creating user
feedback in the CLI is probably confusing.

Cheers,

Mark