On Jun 12, 2006, at 5:38 PM, Klaus D. Witzel wrote:

> Ah. And I was believing the description in True>>#ifTrue:
> -------
> ifTrue: alternativeBlock
> 	"Answer the value of alternativeBlock. Execution does not actually
> 	reach here because the expression is compiled in-line."
>
> 	^alternativeBlock value
> -------
>
> So the bug is in the documentation? This is not an easy one: who  
> would doubt the implementation of such an essential behavior?

I think I see what you're saying. I did a little digging ...

Most of the time, it's only "as if" #ifTrue:, and hence #value, gets  
sent. Interestingly, the whole block gets compiled inline in these  
cases:

	foo

		4 > 3 ifTrue: [self halt]

compiles to

	17 <21> pushConstant: 4
	18 <22> pushConstant: 3
	19 <B3> send: >
	20 <9A> jumpFalse: 24
	21 <70> self
	22 <D0> send: halt
	23 <87> pop
	24 <78> returnSelf

(This output is from the "bytecodes" choice on the source dropdown of  
the browser.)

However, #ifTrue: is actually sent if the situation isn't as obvious,  
as when the block is referred to using a variable:

	foo

		| block |
		block _ [self halt].
	
		4 > 3 ifTrue: block

compiles to

	21 <89> pushThisContext:
	22 <75> pushConstant: 0
	23 <C8> send: blockCopy:
	24 <A4 03> jumpTo: 29
	26 <70> self
	27 <D0> send: halt
	28 <7D> blockReturn
	29 <68> popIntoTemp: 0
	30 <22> pushConstant: 4
	31 <23> pushConstant: 3
	32 <B3> send: >
	33 <10> pushTemp: 0
	34 <E1> send: ifTrue:
	35 <87> pop
	36 <78> returnSelf

I don't know if the VM recognizes sends of #ifTrue: and does any  
runtime optimization on them or not.

The behavior should be the same in either case, at any rate - it  
should act as though #value were sent - so I might be missing what  
you're saying about a bug.

Benjamin Schroeder