YASGuide - Yet Another Style Guide For Julia

This guide is currently a work in progress. In the future, it should feature more examples and justification/explanations for various decisions.

It is more strict than the Julia Manual's Style Guide, and diverges in some places. For stylistic concerns not addressed in this guide, the manual's guide is a good fallback.

This guide is inspired by the JuMP Style Guide and adopts a few of its guidelines.

Naming Guidelines

All type names should be CamelCase.
All struct names should be CamelCase.
All module names should be CamelCase.
All function names should be snake_case (all lowercase).
All variable names should be snake_case (all lowercase).
All constant names should be SNAKE_CASE (all uppercase).
All abstract type names should begin with Abstract.
All type variable names should be a single capital letter, preferably related to the value being typed.
Whole words are usually better than abbreviations or single letters.
Single letters can be okay when naming a mathematical entity, i.e. an entity whose purpose or non-mathematical "meaning" is likely only known by downstream callers. For example, a and b would be appropriate names when implementing *(a::AbstractMatrix, b::AbstractMatrix), since the "meaning" of those arguments (beyond their mathematical meaning as matrices, which is already described by the type) is only known by the caller.

Comment Guidelines

TODO to mark todo comments and XXX to mark comments about currently broken code
Quote code in comments using backticks (e.g. `variable_name`).
Use section headers to split code files into logical chunks. Section header formatting is as follows:
```
#####
##### Section Name
#####
```

Line/Alignment/Spacing Guidelines

Note that lines containing only comments are not considered empty lines.

All files should have newline character at the end of the file.
Use spaces around binary operators (e.g. x+y is bad, x + y is good). The exceptions are:
- Within type parameter lists, where no spaces should surround <: (only within type parameter lists, though - outside of a type parameter list, spaces should surround <: when it is used as a binary operator), and
- Around = for keyword arguments and NamedTuple arguments. This provides easy visual differentiation between keyword/NamedTuple arguments and calls that span multiple lines and variable assignments.
Use a single space after commas and semicolons. The exception is within type parameter lists, where no spaces should be added after commas.
Do not use extra spaces for unary operators, parentheses, or brackets.
Use 4 spaces for indentation; no tabs, ever.
Never follow an empty line with another empty line.
For single-line method definitions, use the form f(args...) = ....
For method definitions spanning multiple lines, use the function ... end definition form.
The first and last lines of an indentation block should never be empty.
Top-level assignments or definitions should generally be separated by a single empty line.
Function definitions using the function form should always be preceded by an empty line and followed by an empty line.
If defining more than one single-line method for the same function, they may be grouped together without empty lines between the definitions.
Try to limit line length to 80 characters per line. This rule should not be enforced arbitrarily, but is rather a soft guideline.
Definitions of types/structs without fields and functions without implementations ("stubs") should written in single-line form (e.g. struct Foo <: AbstractFoo end, function foo end).
If an argument list extends to multiple lines, align new lines to the parenthesis/bracket/etc. that started the argument list.
keyword...end statements form "code blocks" if the delimiting terms (keyword and end) are on separate lines. The contents of such code blocks should be indented. The end delimiter of a code block should either be on the same line as the keyword delimiter, or on its own line following the code block content. In other words, this is bad:
```
function f(x)
    return g(x, 1) end
```
while this is okay:
```
function f(x)
    return g(x, 1)
end
```

Other Syntax Guidelines

Do not explicitly name type variables in method signatures when unary <: or >: would suffice.
Type variable bindings should always be enclosed within {} brackets when using where syntax, e.g. Vector{Vector{T} where T} is good, Vector{Vector{T}} where {T} is good, Vector{Vector{T}} where T is bad.
A decimal should never be the last character in numeric literals (e.g. 1. is bad, 1.0 is good).
Use in instead of = when specifying for statement iteration variables (e.g. for i = 1:5 is bad, for i in 1:5 is good).
If a field type annotation in a struct definition would default to Any, explicitly annotate the field with ::Any.
In argument lists, always separate positional arguments from keyword arguments using a semicolon (;).
In method signatures, only provide names to arguments that are actually referenced in the method body (e.g. f(x::Number, y) = y is bad, f(::Number, y) = y is good).
Do not use the "pipe" operator (|>). The primary use case for |> is to break up nested function calls. For this scenario, one should instead favor assigning intermediary results to well-named variables. This makes code easier to read, debug, and piecewise evaluate in the REPL. Another reason to avoid |> is that it clutters stacktraces.
Qualified macrocalls should directly prefix the macro name - not the qualifying namespace - with @ (e.g. @Module.macro_name is bad, Module.@macro_name is good).

Programming Guidelines

Dispatch type constraints should be used chiefly for dispatch, not for artificially restricting method signatures to "known" types, or merely for documentation.
Dispatching on type parameters should be avoided unless necessary, especially if the method author does not "own" the method and/or the type being parameterized (in which case, this kind of dispatch could be an instance of "type piracy"). This is because naive dispatch on type parameters often does not (and cannot) describe values the way the programmer intends, unless the programmer explicitly accounts for invariance. For example, AbstractArray{<:MyType} does not describe "any value of type <:AbstractArray containing elements of type <:MyType"; it only describes a subset of such values (e.g. typeof(Any[1]) <: AbstractArray{<:Number} is false). Likewise, AbstractArray{>:MyType} describes "any value of type <:AbstractArray that could contain elements of type <:MyType", which is likely a superset of what the programmer intended to describe (e.g. AbstractArray{Any} <: AbstractArray{>:Number} is true). This guideline can generally be relaxed when the method author "owns" both the method and the type being parameterized.
The return keyword should always be omitted from return statements within short-form method definitions (f(...) = ...). The return keyword should never be omitted from return statements within any other context (function ... end, macro ... end, etc.).
If a function does not have a clearly appropriate return value, then explicitly return nothing.
If a function definition is not at top-level, then it should be an anonymous function rather than a named function. This lessens confusion about whether the function is properly overloadable.
When overloading a function from another module, the function name should be qualified with its module (e.g. imported_function(...) = ... is bad, ParentModule.imported_function(...) = ... is good).
Modules should only be loaded with using, never import. This helps prevent methods from being accidentally extended without qualification (whereas import Module: f allows f to be extended without qualification).
It's friendly to selectively export module bindings, but unfriendly to export everything. A module should export unambiguously named bindings that are part of the module's intended API. It's also okay to not export (but document) functions that are intended to be called in qualified form, e.g. CSV.read.
Do not use @assert statements for any error handling/input checking/etc. that should occur even in optimized builds. Per @assert's documentation: "An assert might be disabled at various optimization levels."