Thanks, all. Now I'm understood differences.
On Fri, Apr 10, 2009 at 20:09, Ralph Johnson johnson@cs.uiuc.edu wrote:
In the original Smalltalk, blocks were almost but not quite closures. The ANSII standard says they are closures. As far as I know, all the Smalltalks except Squeak make blocks be closures. So, changing the way Squeak implements blocks will bring it into compliance with the standard and more like other implementations of Smalltalk
I first ran into the problem of blocks not being closures when I tried to write a program like
(1 to: 100) do: [:each | [self process: each] fork]
This will fork 100 processes. If the first spawned process (the one running [self process: 1]) starts to run right away, it will behave like I expect. But suppose the main thread creates all 100 processes before it gives up control. Then the value of "each" will be 100 and so all 100 processes will execute [self process: 100], which is not at all what I expect. If blocks were closures then each execution of the block would have its own context to store its argument, i.e. there would be 100 versions of "each".
You can see a similar problem if you try to implement factorial the way that Scheme programmers do it.
|factBlock| factBlock := [:x | x <= 1 ifTrue: [1] ifFalse: [x * (factBlock value: x-1)]]. factBlock value: 100
Squeak complains that you are trying to evaluate a block recursively. Most Smalltalk programmers would never try this, so they don't care about Squeak's limitations. But people coming to Squeak from a functional programming language are annoyed, because they expect that they can do this with blocks, but blocks are not complete closures.
-Ralph Johnson _______________________________________________ Beginners mailing list Beginners@lists.squeakfoundation.org http://lists.squeakfoundation.org/mailman/listinfo/beginners