Optimizing Java using D

[John Colvin via Digitalmars-d]

On Thursday, 3 July 2014 at 15:40:33 UTC, Wanderer wrote:
> On Thursday, 3 July 2014 at 11:30:57 UTC, Alix Pexton wrote:
>> Saying that one is always more significant than the other is 
>> far too much of an oversimplification.
>
> I just thought, with the presence of structs in D, things are 
> not that simple. Structs don't use references and their 
> contents is located "right in place" if I understand this 
> correctly.
>
> In other words, if you have array of structs, each struct of 
> 100K size, and array's size is 10,000 elements, calling sort on 
> it would require shuffling 1 gigabyte of memory in the worst 
> case, stopping the entire application. (Sorting the same array 
> in Java would require shuffling only 40K of references and 
> would finish instantly, because only reference complex types 
> are allowed there.)
>
> All I can say - yes, amount of swaps can be important in this 
> case of structs, but it's terribly inefficient anyway and 
> should be avoided IMHO.

Firstly, who uses 100kB structs? That's insanely large.

Secondly, sometimes you need the data to be contiguous in memory 
after sorting, for performance or compatibility reasons.

The point where this is important is usually in a middle ground, 
things like a complex type (std.complex.Complex can be as large 
as 32 bytes IIRC) or a particle struct that holds position, 
velocity and acceleration vectors.