[K42-discussion] destruction of global kernel objects

[Bryan S Rosenburg]

Donour Sizemore wrote on 09/20/2006 06:01:07 PM:
> Reading between the lines, it seems like you're saying that these 
> classes shouldn't have destructors and I agree. GCC doesn't.

No.  I'm saying that if XBaseObj has to have a virtual destructor, the 
stub-generator had better take it into account when it figures out what 
method numbers correspond to what actual methods.  I'm guessing that the 
reason you're seeing the destructor invoked is that some 
protected-procedure-call client is passing a method number that it thinks 
corresponds to method foo() in the target XObj and instead corresponds to 
the destructor.

I assume you've done a full clean and rebuild.  Doing things that change 
method numbers in stub-generated code SHOULD get picked up by the 
Makefiles, but I don't trust the dependencies.  Make sure all the 
stub-generated code is re-generated.

Assuming that's not the problem, you will have to look at the intermediate 
files the stub-generator spits out.  One of them is just a skeleton class 
with all the exported methods.  It's compiled only for the sake of 
determining method numbers.  I'm not sure whether that generated class 
inherits from XBaseObj or not.  If it doesn't, you'll have to make the 
generated class match your XBaseObj in order to get the method numbers to 
be correct.

- Bryan
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://ozlabs.org/pipermail/k42-discussion/attachments/20060920/4819b3b6/attachment.htm