Re: broken kernel

Today I am in a paper writing mode, so let me try to explain this a little bit more than usual: The lock in the SmallWiki kernel makes sure that only one command is executed at once. It makes sure that the model doesn't get into an invalid state due to concurrent write accesses. However, having locks especially during the development phase, can be pretty boring: say if you are using the debugger to step through a critical section, restart, jump out of it, ... you might run into troubles because its semaphores are not properly reset. This is neither an implementation error of SmallWiki nor one of the the lock-implementation in Smalltalk, but a common and well known issue in powerful languages such as Smalltalk or Scheme, where the exception handling is done at the language and not at the VM level. Java, C#, C++, PHP, ... don't show this issue since their execution stack is not a first class object and the exception handling is out of the scope of the programmer, after an unhanded exception your application needs to be restarted anyway. A greatly simplified implementation of a lock might look the following piece of code. The critical piece is the ensure block, that ensures that the lock is opened after doing the critical thing in aBlock. So whatever happens inside aBlock (finishing the evaluation of the block, doing a non-local return from within the block, raising an exception, etc.) first the ensure-block is evaluated before continuing with the default executing. So if you have a self halt in there the ensure-block will be evaluated because the debugger is opened in a exception handler at the very top of the stack. However with the debugger we are able to jump back into aBlock and continue the execution, but now within the unlocked context. Of course when exiting again from aBlock the ensure-block will be evaluated again and the lock might get into a bad state. lock: aBlock self waitUntilUnlocked; lock. aBlock ensure: [ self unlock ]. Especially when using Seaside you have to be aware and take care about the use of ensure-blocks: say if you are using continuations to jump into aBlock or out of aBlock the semantics of the #ensure: are basically undefined! Should it be evaluated multiple times? Only the first time? Only the last time? How do we know when we leave aBlock for the last time? Actually I suggest, not to mix continuations and ensure-blocks. No, it is a known problem, without good solutions. Search Google for "continuation unwind" and you will find a lot of material about this. SW2Kernel kernels do: [ :each | each instVarAt: 'lock' put: each defaultLock ]. Cheers, Lukas -- Lukas Renggli http://www.lukas-renggli.ch