Does the dynamic array pre-allocate any memory?

I'm curious if the ~= on an array would be time sensitive. Would it make sense to allocate memory myself and then reset the length?

class A
{
uint [] array;
:
  this()
  {
     array = new uint [10];
     array.length = 0;
  }
}

Rob Adelberg Wrote:

> Does the dynamic array pre-allocate any memory?
> 
> I'm curious if the ~= on an array would be time sensitive. Would it make sense to allocate memory myself and then reset the length?
> 
> class A
> {
> uint [] array;
> :
>   this()
>   {
>      array = new uint [10];

or, "array.length = 10;"  ?

>      array.length = 0;
>   }
> }

Rob Adelberg Wrote:
> Does the dynamic array pre-allocate any memory?

It doesn't preallocate, and currently array appending in D is a slow or very slow operation.
If you need to speed it up, you can use a struct that performs this operation in a faster way, that's available in D2 in the Phobos, and in D1 can be found around (I have a version too for D1). Something like an Appender or ArrayBuilder.

Bye,
bearophile

On Thu, 07 Jan 2010 23:32:21 -0500, Rob Adelberg <adelmetz@mindspring.com> wrote:

> Does the dynamic array pre-allocate any memory?

No.  It allocates on request.  Appending does pre-allocate more memory then asked, the logic being if you are repeatedly appending, you want to optimize the allocations.

>
> I'm curious if the ~= on an array would be time sensitive. Would it make sense
> to allocate memory myself and then reset the length?

If by time sensitive, you mean it is a performance killer?  It is to some degree :)  There is a patch in the works for D2 that will make it more efficient, but if you require the best efficiency, I'd suggest using an array builder object like bearophile alluded to.  That way, your appends don't need to look up GC info.

>
> class A
> {
> uint [] array;
> :
>   this()
>   {
>      array = new uint [10];
>      array.length = 0;
>   }
> }


This is a common trick, but after the patch it will be incorrect on D2 (the syntax will compile, but your preallocation will be wasted).  There will be a new function to replace this.

-Steve