Currently we have a lof of limits in the VM, some of them require in every step recompute the count for ensure that fits the expected values. I think that is too expensive and sometimes could be unfair. For example: One array of 1025 items is not allowed, but two of 1024 yes.
I don't have now the perfect solution but I think that we should improve this.
This method is very expensive, but with our current module is needed:
|
private bool CheckStackSize(bool strict, int count = 1) |
|
{ |
|
is_stackitem_count_strict &= strict; |
|
stackitem_count += count; |
|
|
|
if (stackitem_count < 0) stackitem_count = int.MaxValue; |
|
if (stackitem_count <= MaxStackSize) return true; |
|
if (is_stackitem_count_strict) return false; |
|
|
|
// Deep inspect |
|
|
|
stackitem_count = GetItemCount(InvocationStack.Select(p => p.EvaluationStack).Distinct().Concat(InvocationStack.Select(p => p.AltStack).Distinct()).SelectMany(p => p)); |
|
if (stackitem_count > MaxStackSize) return false; |
|
is_stackitem_count_strict = true; |
|
|
|
return true; |
My proposal is something like this:
class Memory
{
public int Max;
public int Current;
public StackItem CreateInteger(int value)
{
var ret=new StackItem(value);
Current+=ret.Size;
return ret;
}
public void Clean(StackItem i){ Current-=i.Size; }
}
class ApplicationEngine
{
Memory mem;
public ApplicationEngine(Memory mem){ ... }
}If we centralize the creation of the StackItems in only one point, we will have a control of the current memory without need to recompute in every steps.
Is only an idea, i really know that this great minds @neo-project/ngd-shanghai will have a good ones.
But for me the goal is to have only one limit: 16mb of RAM (for example).
Currently we have a lof of limits in the VM, some of them require in every step recompute the count for ensure that fits the expected values. I think that is too expensive and sometimes could be unfair. For example: One array of 1025 items is not allowed, but two of 1024 yes.
I don't have now the perfect solution but I think that we should improve this.
This method is very expensive, but with our current module is needed:
neo-vm/src/neo-vm/ExecutionEngine.cs
Lines 106 to 121 in 2fd56dd
My proposal is something like this:
If we centralize the creation of the StackItems in only one point, we will have a control of the current memory without need to recompute in every steps.
Is only an idea, i really know that this great minds @neo-project/ngd-shanghai will have a good ones.
But for me the goal is to have only one limit: 16mb of RAM (for example).