On Oct 16, 2004, at 5:33 PM, Avi Bryant wrote:

> On Oct 17, 2004, at 1:16 AM, Andres Valloud wrote:
>
>> sd> Now I like the idea of gilad bracha on pluggable type system: The
>> sd> idea in a nutshell is to have OPTIONAL type systems (ownership,
>> sd> ...) that you can use to browse the code but that are NOT used at
>> sd> all for checking or defining the language semantics. Following
>> sd> animorphicSmalltalk. We are waiting to get a good guy to try this
>> sd> idea on Smalltalk.
>>
>> My 2 cents: since we'd do that to get performance, and 90% of the time
>> is spent on less than 10% of the code, we should only see the optional
>> type tags in less than 10% of the code - otherwise it has no value.
>
> Why for performance?  One of the cool things about animorphic was that 
> the type annotations were completely ignored by the JIT, even though 
> they were available.  The runtime info, I assume, was better.

Taking that idea a bit further, it might be useful to have type 
information flow the other direction: generating type annotations based 
on the runtime information gathered by the VM. That would be a way to 
"initialize" the type system without having to go around manually 
annotating code.

Note also that type annotations don't have to be part of the source 
code. It wouldn't be hard to have a browser switch for showing and 
hiding them, or using rollover balloons, or color coding expressions 
based on their types, or...

Colin