At 03:04 PM 12/9/98 -0500, Adam P. Jenkins wrote:
>> The problem is independent of duplication. Say you have a method that says
>>
>> foo
>> 	| b |
>> 	b := 'abc'.
>> 	Transcript space; show: b.
>> 	b at: 2 put: $d
>>
>> What prints the second time you send #foo? Because the literal (copied or
>> not) is stored with the compiled method and not the activation, the second
>> time aroud b is 'adc' even though it's right after the assignment
>> of 'abc'.
>
>Yuck!  I assume from what you say below that this problem wouldn't occur
>with VisualWorks.  Is that true?

No, the problem occurs. With VisualWorks 'abc' prints the first time and
'adc' prints the second time because the strings are generated at
compile-time and are mutable.

>> For VisualWorks, we got rid of the sharing but not the mutability.
>
>This seems like the best solution to me.  I can't imagine that sharing is
>useful very often.  How often are there really identical string or array
>literals in the same method?  And in general a literal is going to be small,
>since you have to type it in, so the memory savings by sharing are also
>small.

Who says code has to be typed in? VisualWorks, for example, has a number of
code-generating subsystems, and I wouldn't be surprised if there were many
in Squeak land as well. In particular, our window painter produces
representations of window layout that are encoded as literal arrays.

>The simplest method is to just create a new object every time a
>literal is encountered.

And, for example in the case of our window builder, suffer a large
performance penalty every time you open a window. And that penalty would be
paid by everyone just so that the few people who accidentally store into a
literal array or string would have a better time of it. That's not the
tradeoff we decided to make, and I suspect that this would also be a worse
choice for Squeak.

If this upsets your sensibilities ("Yuck" is a clue), I think you're more
likely to find a better path by making immutability cheap in Squeak, than
by removing compile-time literal construction.

>If a programmer wants them to be shared, they can
>just do
>
>a := 'hello'.
>b := a.
>c := a.

Unlike compile-time object creation,I don't know of cases where sharing
would be an important issue, although it may be my lack of imagination.

glenn