(integrating ClassRenameFixTest was: Re: Collecting actions for RC2)

stephane ducasse stephane.ducasse at gmail.com
Fri Oct 13 17:17:35 UTC 2006

HI chris

I'm going over RC2. What is the status for this problem?

> Yeah, because of the need for the oldName.  Before the fix, the  
> system would not set the class's (new) name until after  
> renameClass:as: was called, and that method did all the clean up  
> (such as removing the class-name key from Smalltalk) based on the  
> (still old) class name.  At the bottom of renameClass:as: it then  
> signals the SystemChangeNotifier.  Everything is "cleaned up"  
> except, OOPS, the class' name itself is still the old name.  This  
> is the inconsistent state that.
> To fix the problem, I set the class' name *before* calling the new  
> method, renameClass:from:, and passing in the old name from which  
> we can do all the Smalltalk cleanups before finally signaling the  
> SystemChangeNotifier while all clean and consistent.
> Hope that made sense..
> I don't like the mostly-duplicated code either, but its more  
> important to me for it to work correctly and then we can figure out  
> how to get rid of renameClass:as: later..
>> 2- I am a bit worried about your remark about the obsolete class and
>> the bad event handler...
>> Do you still have an image with this problem ?
> 3.9-7061 has the problem.  I think its just a event-handler that  
> didn't get cleaned up from one of the tests due to unusual  
> circumstances.  3.9 underwent a lot of changes, bringing in Traits,  
> new Compiler.  Who knows what all Stef did in the wee hours, maybe  
> exiting with debugger open and then ensure's didn't get called or  
> something, who knows..  I don't think its anything to worry about;  
> the important things are:
>   - we can (and should) easily clean it up in the existing 7061 image
>   - whatever caused it was probably a very unusual situation  
> related to the construction of the 3.9 image
>   - the test cleans up after itself, we can't get the problem to  
> happen again
>   - the problem is an artifact of only the test, even if we didn't  
> clean it, it doesn't affect anything else, not even the bug fix
> Thanks,
>   Chris

More information about the Squeak-dev mailing list