Re: Feature suggestion

[Michael Ferenduros <mike.ferenduros@...>]

Yep, I can definitely see how it'd be impractical to implement.

I guess the alternative would be to come at it from the opposite  
direction by letting mutators to replace the thing they're mutating. Eg:

metatable.__newindex = function( thing, index, value )
	local result = thing:copy()
	rawset( thing, index, value )
	return result
end

v.x = 100		-- => "v = getmetatable(v).__newindex( v, 'x', 5 ) or v"


Dunno how implementable this is though.

Mike F



On 22 Jul 2008, at 19:15, Mark Hamburg wrote:

> It definitely takes slightly different thinking to deal with  
> mutability v immutability. One way to do this is to have immutable  
> types plus a calculator object which supports an API for working on  
> mutable values. When you ask the calculator for a value, it would  
> make an immutable copy. There are probably also useful patterns from  
> the Java community that would apply here since it has the same  
> issues with all complex types being pass-by-reference.
>
> The performance problem with your approach is that it would require  
> attempting to invoke the metamethods on every parameter ever passed  
> and every value ever assigned.
>
> Mark