Incredibly Slow Dictionary Lookup when Debugging

[mattcbro]

If there is a large enough array inside a dictionary it appears to slow the debugger to a crawl. Actually this example never really finishes in any sort of reasonable time when stepping through the debugger. This code uses the Juno interface, but I believe the debugger interface has the same problem.

This is happening under Julia 1.1.1 .

""" extract the raw data from dictionary """
function valdata(val)
    # next statement takes an infinite amount of time to step through
    xraw = val["rdat"] + 1.0im .* val["idat"]
    return(xraw)
end

ar = randn(3000,4000)
ai = randn(3000,4000)
vdat = Dict()
vdat["rdat"] = ar
vdat["idat"] = ai
vdat["num"] = pi
vdat["text"] = " This is text"

xcomp = Juno.@enter valdata(vdat)

[pfitzseb]

This isn't related to the dictionary lookup, you're just running into the inherently quadratic runtime of adding two arrays:

julia> @time @interpret rand(30,40) + rand(30,40)
  1.287965 seconds (3.45 M allocations: 142.426 MiB, 4.74% gc time)

julia> @time @interpret rand(300,400) + rand(300,400)
 35.905406 seconds (227.50 M allocations: 8.209 GiB, 4.43% gc time)

[mattcbro]

It's quadratic only in one of the dimensions of the 2D array I suppose. However the bigger question is why the heck is the debugger so slow adding 2 arrays when Julia itself does the same almost instantly? Can't one compile a line of code and then run it during debug?

[KristofferC]

See JuliaDebug/JuliaInterpreter.jl#206 and JuliaDebug/JuliaInterpreter.jl#204 for performance discussion.

[KristofferC]

Closing this since it isn't really actionable from this repo.

[mattcbro]

Which repo should this be filed in then? It does make debugging impossible for most of my code.

[KristofferC]

https://github.com/JuliaDebug/JuliaInterpreter.jl but unless there is some new information here, it might get closed as a dup of one of the other performance issues.