I expected the memory to ramp up in the first couple iterations and eventually reach a stable point, but for some reason, windows task manager shows it increases on every iteration. Compiling with -m64 doesn't seem to change this issue.

http://pastebin.com/G5JXR9AA

On Sunday, 16 June 2013 at 02:12:02 UTC, Geancarlo Rocha wrote:
> I expected the memory to ramp up in the first couple iterations and eventually reach a stable point, but for some reason, windows task manager shows it increases on every iteration. Compiling with -m64 doesn't seem to change this issue.
>
> http://pastebin.com/G5JXR9AA

I think it is safe to assume the current GC is unreliable, booo!

On 06/16/2013 07:20 PM, Geancarlo Rocha wrote:
> On Sunday, 16 June 2013 at 02:12:02 UTC, Geancarlo Rocha wrote:
>> I expected the memory to ramp up in the first couple iterations and
>> eventually reach a stable point, but for some reason, windows task
>> manager shows it increases on every iteration. Compiling with -m64
>> doesn't seem to change this issue.
>>
>> http://pastebin.com/G5JXR9AA
>
> I think it is safe to assume

This way of reasoning is flawed.

> the current GC is unreliable, booo!

No it is not. You are not even allocating GC-controlled memory during the loop.

This is a std.container.BinaryHeap issue. It never shrinks the underlying container.

http://d.puremagic.com/issues/

On Sunday, 16 June 2013 at 18:02:26 UTC, Timon Gehr wrote:
> On 06/16/2013 07:20 PM, Geancarlo Rocha wrote:
>> On Sunday, 16 June 2013 at 02:12:02 UTC, Geancarlo Rocha wrote:
>>> I expected the memory to ramp up in the first couple iterations and
>>> eventually reach a stable point, but for some reason, windows task
>>> manager shows it increases on every iteration. Compiling with -m64
>>> doesn't seem to change this issue.
>>>
>>> http://pastebin.com/G5JXR9AA
>>
>> I think it is safe to assume
>
> This way of reasoning is flawed.

Agreed, but hey...it caught someone's attention. Mission accomplished :)

>
>> the current GC is unreliable, booo!
>
> No it is not. You are not even allocating GC-controlled memory during the loop.
>
> This is a std.container.BinaryHeap issue. It never shrinks the underlying container.
>
> http://d.puremagic.com/issues/

If that is the case, I don't think that's something a beginner would figure out unless he went to look the source code.
Coming from C#, even the idea that you have to pass a "store" to the BinaryHeap is weird to me. I also don't understand this:

void toy()
{
    Array!(int) store;
    auto minheap = new BinaryHeap!(Array!(int), "a > b")(store);
    auto rng = new Random(unpredictableSeed);

    assert( store.length == 0 );
    writeln("inserting...");

    for( uint i = 0 ; i < 100_000 ; i++)
    {
        minheap.insert(1);
        assert( store.length == 0 );
    }
    writeln("done.");
    assert( store.length == 0 );

    writeln("removing");
    while(!minheap.empty)
    {
        minheap.removeFront();
        assert( store.length == 0 );
    }
    writeln("done.");
    assert( store.length == 0 );
}

I thought the BinaryHeap would modify store's buffer AND it's properties, but it doesn't seem to be the case. How is store being used? Again this is the sort of thing that makes a beginner's learning curve steeper.


Thanks for your attention

On 06/16/2013 08:37 PM, Geancarlo Rocha wrote:
> ...
>>
>>> the current GC is unreliable, booo!
>>
>> No it is not. You are not even allocating GC-controlled memory during
>> the loop.
>>
>> This is a std.container.BinaryHeap issue. It never shrinks the
>> underlying container.
>>
>> http://d.puremagic.com/issues/
>
> If that is the case, I don't think that's something a beginner would
> figure out unless he went to look the source code.

That's what I have done. I consider the behaviour entirely unreasonable. You may report it using the link above.

> Coming from C#, even the idea that you have to pass a "store" to the
> BinaryHeap is weird to me.

You mean coming from the System.Collections namespace of the .NET framework. BinaryHeap is a library artefact.

> I also don't understand this:
>
> void toy()
> {
>      Array!(int) store;
>      auto minheap = new BinaryHeap!(Array!(int), "a > b")(store);
>      auto rng = new Random(unpredictableSeed);
>
>      assert( store.length == 0 );
>      writeln("inserting...");
>
>      for( uint i = 0 ; i < 100_000 ; i++)
>      {
>          minheap.insert(1);
>          assert( store.length == 0 );
>      }
>      writeln("done.");
>      assert( store.length == 0 );
>
>      writeln("removing");
>      while(!minheap.empty)
>      {
>          minheap.removeFront();
>          assert( store.length == 0 );
>      }
>      writeln("done.");
>      assert( store.length == 0 );
> }
>
> I thought the BinaryHeap would modify store's buffer AND it's
> properties, but it doesn't seem to be the case. How is store being used?

I'd guess a copy of store is used, since Array is a value type (using a ref-counted payload).

> Again this is the sort of thing that makes a beginner's learning curve
> steeper.
>

I wouldn't spend too much time on std.container. A sane collections library currently is not part of the standard distribution.

>
> Thanks for your attention