On 12 December 2011 17:10, Chris Muller asqueaker@gmail.com wrote:
The halt is there on purpose because it is a long-standing bug with Avi's original WriteBarrier that, when it occurs, will need to be closely looked at at that point in the code. It could only possibly happen when changing the class-definition of a class with instances behind WB, so it probably wouldn't happen in prod (presumably, your tests would catch it anyway, right? :).
It did happen in development, and I expect it to happen often, because I do change class definitions while instances of a class behind WB exists. That is exactly what I like in smalltalk - programming the system in runtime, and I am not going to stop doing that. To kill all the instances I would have to clean Seaside seasons and run GC. So it's not a minor thing for me. So what do I do about it? You say it's a long-standing bug with original Avi's code - do you know what was the bug about? - do you know if the problem exists with the new WB? - maybe I can look into it, even though so far I didn't notice any problems, I just removed the "halt" - I can just switch the WB off like it used to be in Magma 1.2
For PointerFinder, I didn't look at your patch, but wouldn't you need to put the correct code in the -Pharo platform package? I'm fine to move the PointerFinder reference to the -Squeak package if necessary. People of Pharo, please let me know!
I see. It's up to you. The above bothers me more. Just know that PointerFinder is deprecated and I guess is going to be removed.