Performance Linter UI

Dynamic performance linting can easily be done with Traceur.jl and the @trace macro, which prints any warnings directly to the REPL:

julia> @trace naive_sum([1.0, 2.0])
(Base.indexed_next)(::Tuple{Int64,Bool}, ::Int64, ::Int64) at tuple.jl:54
  returns Tuple{Union{Bool, Int64},Int64}
(naive_sum)(::Array{Float64,1}) at /home/basti/Documents/test.jl:211
  s is assigned as Int64 at line 211
  s is assigned as Float64 at line 213
  dynamic dispatch to s + x at line 213
  returns Union{Float64, Int64}
3.0

This works fine, but obviously isn’t ideal for an IDE such as Juno. Fortunately, Traceur.jl provides an API that outputs those warnings in a structured way:

julia> Traceur.warnings(() -> naive_sum([1.0, 2.0]))
5-element Array{Traceur.Warning,1}:
 Traceur.Warning(Traceur.DynamicCall{Base.#indexed_next,Tuple{Tuple{Int64,Bool},Int64,Int64}}(Base.indexed_next, ((0, false), 1, 1)), -1, "returns Tuple{Union{Bool, Int64},Int64}")
 Traceur.Warning(Traceur.DynamicCall{#naive_sum,Tuple{Array{Float64,1}}}(naive_sum, ([1.0],)), 211, "s is assigned as Int64")
 Traceur.Warning(Traceur.DynamicCall{#naive_sum,Tuple{Array{Float64,1}}}(naive_sum, ([1.0],)), 213, "s is assigned as Float64")
 Traceur.Warning(Traceur.DynamicCall{#naive_sum,Tuple{Array{Float64,1}}}(naive_sum, ([1.0],)), 213, "dynamic dispatch to s + x")
 Traceur.Warning(Traceur.DynamicCall{#naive_sum,Tuple{Array{Float64,1}}}(naive_sum, ([1.0],)), -1, "returns Union{Float64, Int64}")

Displaying these properly in Juno will be what I spend the first couple of weeks of this years GSoC on.

Atom does have some packages for more traditional linting, namely linter for the logic and linter-ui-default for the UI elements. I’m not entirely sure yet if I’ll end up using them for the final design, but for quickly iterating they’re very well suited.

Concept

Integrating the linter package is super easy on the Atom side. And here’s the code for the Julia side – surprisingly simple as well:

Wrapping Traceur.jl’s output

With a bit of glue code for converting Traceur.Warnings into the Messages taken by linter we get a pretty UI:

There are still some issues to figure out, but this seems pretty neat for a first pass.

EMUlator

Concept

Wrapping Traceur.jl’s output