Optimization fun

[Steven Schveighoffer via Digitalmars-d]

On 11/7/14 1:05 AM, H. S. Teoh via Digitalmars-d wrote:
>
> A little more investigation revealed the culprit: a queue object
> implemented with subarrays, and subarrays were being used as stacks by
> appending with ~= and popping by decrementing .length. A dangerous
> combination that causes excessive reallocations! After inserting
> assumeSafeAppend() in the right place, the memory consumption rate
> dropped to more reasonable levels. It's still visibly faster than when
> GC is enabled, so probably there are other places with excessive
> allocations, but this particular change stood out because after adding
> assumeSafeAppend() to the code, the solver's performance *even with GC
> enabled* improved by about 40%-50%. So this explains why calling
> GC.disable() had such a pronounced effect before: the GC is too busy
> cleaning up after unnecessarily allocation-hungry code!

In D1 this would have worked fine. A "queue array" wrapper may be a good 
candidate for Phobos that works just like a normal array but calls 
assume safe append whenever decrementing length.

-Steve