Fix typing in Binary Heap

[ChrisRackauckas]

The valtree in the binary heap is not strictly typed. In some codes I have, this causes a 2x performance reduction in the entire code (<.1% is data structure related) due to checking if a binary heap is empty. The simple fix is to make it Vector{T} instead of the non-concerete Array{T}.

Reference for performance problems: SciML/StochasticDiffEq.jl#13

[codecov-io]

Current coverage is 95.85% (diff: 100%)

Merging #263 into master will not change coverage

@@             master       #263   diff @@
==========================================
  Files            30         30          
  Lines          2220       2220          
  Methods           0          0          
  Messages          0          0          
  Branches          0          0          
==========================================
  Hits           2128       2128          
  Misses           92         92          
  Partials          0          0          

Powered by Codecov. Last update 52b74de...000fcbc

[kmsquire]

Thanks!

[ChrisRackauckas]

Could a patch be tagged with this change? This'll have a large effect on some of the main DifferentialEquations.jl methods. It's blocking a release which inlines a few more things which previously needed a function barrier to stop the propagation of this type instability.

[ChrisRackauckas]

Though I understand if you want to wait until after you migrate to JuliaContainers

[kmsquire]

I'll try to tag this today. The move to JuliaContainers shouldn't affect tagging, as a redirect will be in place.

[ChrisRackauckas]

Thanks!

[kmsquire]

@ChrisRackauckas, wondering if this is urgent or can wait a few days to get some additional changes in?

Reasoning isn't JuliaContainers (or whatever org that becomes), but more that there are changes in SortedDict constructors that warrant a minor version bump, as well as some upcoming changes in SortedSets and SortedMultiDicts (with pull requests) that warrant a minor version bump, and I would rather only bump the minor version once.

If this is urgent, I can just this change on a branch without the SortedDict changes and tag (cc: @tkelman), and then tag all of the sorted container changes at once (this commit will still be in the history, albeit with a different SHA).

Let me know your thoughts.

[ChrisRackauckas]

It's not urgent. If it takes a week or two I wouldn't mind. I mentioned it because the tag frequency in this repo looks unpredictable, and that is a small enough change that I was hoping to make sure that it wouldn't just sit two months waiting for a larger change to happen. But since it sounds like you have something else on your plate that you're churning through, I can wait.

[kmsquire]

If it starts to take a while (say longer than a week), bump this and I'll put it on a branch and tag it. But my hope is that the other changes are in over the next few days.

…

On Sun, Jan 29, 2017 at 12:38 PM Christopher Rackauckas < ***@***.***> wrote: It's not urgent. If it takes a week or two I wouldn't mind. I mentioned it because the tag frequency in this repo looks unpredictable, and that is a small enough change that I was hoping to make sure that it wouldn't just sit two months waiting for a larger change to happen. But since it sounds like you have something else on your plate that you're churning through, I can wait. — You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub <#263 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AANoElTbL7rrAfXpVffIAPxT8j96lhi9ks5rXPjhgaJpZM4Lwbvf> .

[ChrisRackauckas]

Thanks!