Julia coding guidelines

Our recommendations for creating and maintaining reliable scientific software in Julia.

Good default choices

• Version control: GitHub
  • de facto standard for open source software
  • provides all necessary tools for developing software in a team
  • includes Continuous Integration pipelines for continuous testing
• Continuous integration: GitHub actions
  • comes integrated into GitHub already
  • large template library from which your workflow can be put together
• Testing framework: Test.jl
  • standard library option for unit tests, rough analogon to Python’s UnitTest library
• Documentation: Documenter.jl
  • Markdown based
  • includes doctests for code examples
  • can be seen as an analogon to Sphinx
• Profiling:
  • Profile.jl:
    • Part of the Julia standard library.
    • Can be used to profile allocations and CPU - time
    • various visualizations available via separate packages (ProfileView.jl, FlameGraphs.jl, ProfileVega.jl)
  • BenchmarkTools.jl: For more high-level, statistical information about execution time.

Tooling

Julia is a relatively young language with a smaller community than, say, Python, C++ or Matlab. As such, the selection of tools is a bit more limited. Some suggestions:

• IDE/Code Editor:
  • Visual Studio Code: Free. The de-facto standard in the community. Provides support for linting, test discovery, formatting and more. Recommended way to go.
  • Juno: Free. Build on top of Atom. Atom is no longer supported by Microsoft/GitHub, and Juno is only maintained and wont receive new features. Only consider this if you already have an Atom-based workflow and are unwilling to switch.
  • Sublime Text: Free, but will ocassionally ask you to buy it. Syntax highlighing for Julia is available, but support for other functions is less developed than in the other two.
• Code Formatting:
  • automatic formatting is available via VSCode’s Julia plugin or through Juno.
• Linting:
  • linting is available via VSCode’s Julia plugin or through Juno.

Recommended libraries

• Arrays and vectors (numpy analogon):
  • standard library core contains classes for dense, n-dimensional arrays analogous to numpy-arrays.
  • SparseArrays.jl: Part of the standard library. Supports sparse vectors and matrices.
  • Tensors.jl and TensorOperations.jl: For higher-order-Tensor math.
• Linear Algebra:
  • LinearAlgebra.jl: Standard library module for linear algebra operations like matric traces, in verse, multiplication and so on.
• Data analysis and data transformation:
  • DataFrames.jl and DataFramesMeta.jl for tabular data. Julia’s analog to Pandas known from Python. Use this for higher performance.
  • Query.jl Apply common data-analysis workflows like split-apply-combine, filter etc to almost any iterable Julia datastructure. Useful when working with data that’s available in various different formats.
• Data input/output and storage:
  • HDF5.jl: For arbitrary data that should have a hierarchical organization in a single file.
  • Arrow.jl: For arbitrary data that’s organized in a tabular fashion.
  • CSV.jl: For tabular, human readable data output.
• Visualization:
  • Plots.jl: Julia Frontend with a homogeneous interface for various well known plotting packages like matplotlib, plotly or GR.
  • Makie.jl: Feature-complete plotting package in pure Julia.
  • AlgebraOfGraphics.jl Provides support for the grammar-of-graphics paradigm based on Makie.jl .
  • VegaLite.jl: Provides access to VegaLite from Julia, particularly useful for making interactive graphics.
• Machine Learning/Deep Learning:
  • SciKitLearn.jl for having access to the python based scikit-learn library in Julia.
  • Flux.jl for deep learning applications.

Have a look at the JuliaHub package database to find other packages for your use case.

Style guides and best-practices:

Much has already been written about how to write good Julia code which needs not be repeated here. Here are some starting points:

• Official style guide
• Official performance tips