Bug 0583 - likely emergency issue

Marcus Denker denker at iam.unibe.ch
Fri Dec 3 10:23:28 UTC 2004


Am 03.12.2004 um 10:59 schrieb stéphane ducasse:

> Hi tim
>
>>> In a 6461 image, it is possible to load VMMaker from SM.
>>>
>>> In a 6475 updated image it causes a vm-fatal crash.
>> My thanks to Marcus for an incredibly rapid solution. It apears to 
>> work
>> for me.
>>
>> Stef, what is your opinion? You provided the orginal change that seems
>> to be the culprit but I assume you thought there was good reason for
>> it, so what was the intent here? Does Marcus' fix cause you other
>> problems we should pay attention to ?
>
> I'm travelling right now. So this is difficult for me to say what 
> happened.
> The refactoring I did were to avoid to crash the VM but Behavior new 
> new :)
> So I made sure that Behavior new gets a method dictionary and decent 
> format
> But I will sync tuesday or over ichat with marcus.
>

It should be fine: the test that Stef provided keeps on running.

And the second version of my fix (on mantis) shorcuts  "new"  in 
ClassDescription (sole
subclass of Behavior) so now we should be really save, as the new on 
behavior is never called.

But even before, all users that did a Behavior class>>new didn't mind 
it to be initialized to some default
values instead of nil. e.g. the first thing the classBuilder does after 
"Metaclass new" is to initialize new class correctly.

So, this is safe to go in, I'd say. An alternative would be to revert 
the change an instead add it in 3.9alpha (which I
should have done in the first place).

    Marcus



More information about the Squeak-dev mailing list