I added a entry in the bugtracker for this so we don't forget:

http://code.google.com/p/pharo/issues/detail?id=2819


On Aug 16, 2010, at 8:53 PM, Eliot Miranda wrote:

BTW here's the VisualWorks 7.7 method.  For me it's a little too clever.

whileFalse: aBlock 
"Evaluate the argument, aBlock, as long as the value of the receiver is false."

^self value
ifFalse:
[aBlock value.
[self value] whileFalse: [aBlock value]]

"This method is inlined if both the receiver and the argument are literal
blocks. In all other cases, the code above is run. Note that the code
above is defined recursively. However, to avoid actually building an
activation record each time this method is invoked recursively, we have
used the '[...] whileFalse: [..]' form in the last line, rather than the more
concise 'self whileFalse: aBlock'. Using literal blocks for both the receiver
and the argument allows the compiler to inline #whileFalse:, which (in the
absence of type inferencing) could not be done if we were to use
'self whileFalse: aBlock'."
 
but it's accurate and does the job of turning
    | a b |
    a := [i <= j].
    b := [i := i + 1].
    a whileTrue: b
into an iterative execution.

So at the very least we should improve the comment in whileTrue: and whileFalse: to point to the compiler optimization in MessageNode and IMO the VisualWorks implementation is the best compromise between clarity and honesty given the lack of tail-recursion elimination.

--
Marcus Denker  -- http://www.marcusdenker.de
INRIA Lille -- Nord Europe. Team RMoD.