Skip to content

Julia v1.0 transition

Jump to bottom
Emil Ringh edited this page Sep 18, 2018 · 3 revisions

Revise

Revise is used to not have to restart Julia every time code changes are made. To install Revise, hit ] in the Julia REPL to enter Pkg REPL-mode, then type:

(v1.0) pkg> add Revise

Then add the following to ~/.julia/config/startup.jl (which you may have to create if it doesn't exist already) to enable Revise on startup, and restart Julia:

try
    @eval using Revise
    # Turn on Revise's automatic-evaluation behavior
    Revise.async_steal_repl_backend()
catch err
    @warn "Could not load Revise."
end

For more details, read about Revise here.

Setting up the project

Start/navigate Julia to the NonlinearEigenproblems directory. Then hit ] in the Julia REPL to enter Pkg REPL-mode, then type:

(v1.0) pkg> activate .

Hit backspace to return to the REPL, then type:

julia> using NonlinearEigenproblems

Fixing ArgumentError: Package...

The error

ERROR: LoadError: ArgumentError: Package Arpack [7d9fca2a-8960-54d3-9f78-7d1dccf2cb97] is required but does not seem to be installed:                                                     
 - Run `Pkg.instantiate()` to install all recorded dependencies.
...

Run this command

julia> using Pkg
julia> Pkg.update("Arpack")

(I know it's deprecated, so maybe there is an better way.)

Add new dependency

Start/navigate Julia to the NonlinearEigenproblems directory. Then hit ] in the Julia REPL to enter Pkg REPL-mode, then type:

(v1.0) pkg> activate .

(NonlinearEigenproblems) pkg> add NAME-OF-PACKAGE

This will update the file Project.toml, and may also update Manifest.toml.

Running all unit tests from command line

Go to the NonlinearEigenproblems directory. Then run:

julia --check-bounds=yes --color=yes -e 'using Pkg; Pkg.activate("."); Pkg.test()'

General notes

  • There are likely errors remaining in currently untested code paths where opnorm is called with a vector. Assuming that the call is always performed with vectors (and not matrices), these can simply be changed to changed to norm.

  • eye and speye are removed. Try to use the I form whenever possible. Avoid using I as a variable name.

  • Be cautious about transposed matrices when used with sparse matrices. For example:

A = sprandn(100000,100000,0.000001);
x = randn(100000,1);
@time x'*A*x

In Julia 0.6.4, this code runs in 0.0016 seconds. In Julia 0.7, it runs in 13.4 seconds. The reason is that x' is no longer a Matrix{Float64}, but an Adjoint{Float64,Matrix{Float64}}. This can be fixed by materializing the adjoint matrix:

@time copy(x')*A*x
  0.000983 seconds (10 allocations: 1.526 MiB)

  • Try to avoid global variables, even in tests. Wrap variable declarations in tests in a @testset or a function. If you need a global variable for some reason, try to give it a unique name, e.g. test_broyden_nep.

  • Stack trace line numbers may be incorrect. Not sure what this is caused by, some cases might be due to inlining, others perhaps by having multiple statements on the same line? Sometimes hard to track.

Loading packages globally at startup

If you want to hack around in REPL, and not having to load all "basic packages" manually, you could try to add this code to ~/.julia/config/startup.jl. It takes some time to start Julia since it loads all the packages. However, depending on how you use it it might be nice to have. Or at least to have a short-command for doing it. It gives more of a MATLAB-ish feeling to is since you do not have to manually load the basics every time.

for pkg = [:SparseArrays, :LinearAlgebra, :Random, :IterativeSolvers, :Printf, :Statistics, :InteractiveUtils] #:Plots
  try
      @eval using ($pkg)
  catch err
      @warn "Could not load $pkg."
  end
end
Clone this wiki locally