Developer Style Guide

In general we aim to follow the Julia Style Guide but there are some exceptions due to our specific needs and a different background.

The content of this page are merely guidelines. There may be good reasons to deviate from them in some cases; in that case just do so.

General styleguide

Use Julia conventions where applicable and when they don't contradict our own rules above.
Unless really really necessary, don't add new dependencies. Every new dependency complicates the development workflow, in that we will need to stay compatible with this package.
If already existing types in OSCAR are almost what you need, consider improving them instead of writing your own. While it might be tempting to create a new polynomial ring type for the new application because some feature is missing, it causes a lot of work and compatibility issues: Will the new type support
- normal functions (gcd, factor),
- quotient fields,
- modules and residue rings,
- conversion to and from other already existing types?
Whenever functions return the same mathematical object, but in different mathematical categories, the first argument should be the desired return type. One example is projective_space(NormalToricVariety, *) vs projective_space(ProjectiveScheme, *). However, if the return type is different, even if the result describes the same mathematical object, it should be indicated in the function name, for example automorphism_group vs automorphism_group_generators vs automorphism_list.
Whenever functions expect a ring, field, algebra, etc. as input they should be passed as the first argument, for example, polynomial_ring(QQ, :x).
Follow the mathematics. If your function needs a list of points, you should create a point-type (or use the one already there) and then use this. For user-facing functions, please do not use re-purposed lists, arrays, matrices...
Input sanity checks should be enabled by default, they can then be disabled internally if they are known to be true, and manually by users.
All user-facing functions that expect some kind of indeterminant name etc. (like polynomial_ring(QQ, <indeterminant_name>)) should accept a VarName = Union{Symbol, Char, String}, and convert it to a symbol for internal handling. Library and test code should (if possible) call such functions with Symbol arguments, as this is the most efficient way.

The usual Julia naming conventions apply to OSCAR, too (that said, for various reasons our code still violates quite some of them; but in general we strive to reduce these). Here is a summary of the naming convention followed in OSCAR:

Use CamelCase for types and snake_case for everything else. (Internal functions do not have to follow these rules.) Types (and their constructor) tend to be in CamelCase. However, please also provide the constructor (or a constructor) in snake_case. As a user one usually does not know if something is a constructor or a function.
For filenames we recommend using snake_case.jl.
Noteworthy difference to Julia base is that we do not have exceptions for is* or has* as prefix. It is is_foo instead of isfoo and has_bar instead of hasbar. The main reason is to avoid awkward constructions like isvery_ample, while also being consistent. For compatibility with standard Julia, while staying consistent internally, we also provide aliases (using AbstractAlgebra.@alias) for various standard Julia functions, e.g. is_one as alias for isone.
A function returning the number of some things should be named number_of_things, alternatively it can be named n_things with an alias to number_of_things. The preferred style should be consistent throughout the corresponding part. For some very common things, like the number of generators, we additionally provide a shorter alias, e.g. ngens for number_of_generators. These aliases should be very short and without underscores.
For generic concepts choose generic names, based on general algebraic concepts, preferably not special names from your area of speciality.
Avoid direct access to members of our objects. This means, do not use something like A.foo, instead use a suitable getter get_foo(A), and if there is none, please write one or request that one be written. Internal member names are free to change at any time, but functions can be deprecated properly.
In Julia we have multiple dispatch, so we do not need functions like point_from_matrix as the "from" part is clear by the type of the argument. It should be called points(T::Matrix) in some variation. Similarly for matrix_to_points. Of course it is fine to use them internally, where useful.

Code formatting

Before making some suggestions for code formatting rules, a warning: we deliberately are lax about enforcing these rules, as often contributors (esp. new ones) already struggle enough without being forced to also adhere to specific formatting rules. As long as code is sufficiently readable, we may accept it.

But in the same vein (i.e., to minimize frictions for others working on OSCAR), we ask everyone to generally refrain from reformatting large chunks of code (even if it is to make it adhere to the rules described below), unless this is carefully coordinated with all stakeholders of the affected code.

Also, ideally don't mix code reformatting with other changes, as this makes it harder to understand what is going on. At the very least, use different commits for the reformatting changes and the actual changes. But in general you shouldn't reformat code unrelated to the changes you are making.

Editor configuration

Please check if your editor can be configured to honor our .editorconfig file, see https://editorconfig.org for more information about this.

JuliaFormatter

There is a .JuliaFormatter.toml in our git repository. To format your files, first add JuliaFormatter.jl in Julia and then use

using JuliaFormatter
format_file("path/to/file/file.jl")

Unicode

As most modern programming languages, Julia allows the use of Unicode, e.g., α, in the REPL as well as in source code. As this reduces accessibility to various groups of users and developers, the use of Unicode should be kept to a minimum. Here is a general principle:

Do not use Unicode characters inside functions. See Unicode printing for the exception concerning printing.

Whitespace

Do not use tabs.
Do not put spaces "inside" parenthesis.
Do put spaces after commas.

Good example:

f(x, y) = x + 1
print(f(1, 2))

Bad example:

f( x,y ) = x + 1
print( f ( 1,2 ) )

Loops and other control structures

for loops should use in not =
don't put spaces around the : in a range

Good example:

for i in 1:3
  println(i)
end

Bad example:

for i = 1 : 3
  println(i)
end

Code structure

do not nest loops and if clauses too deeply; if you are using 5 or more levels, then in general that's a hint that you should refactor; e.g.
- by moving parts of the code into a separate function
- by replacing guard constructs like
```
for i in A
  if flag
    ...
  end
end
```
  by
```
for i in A
  if !flag
    continue
  end
  ...
end
```
  or
```
for i in A
  flag || continue
  ...
end
```
- by merging loops: you can replace
```
for i in A
  for j in B
    ...
  end
end
```
  by
```
for i in A, j in B
  ...
end
```
Functions should not have too many arguments. If you need a bunch arguments, chances are that introducing a new type makes it more readable.
Functions should not be too long; very long functions are in general harder to understand; it is also more difficult to see all the code at once. Consider splitting the function into multiple ones, if it is sensibly possible.
Every export statement must be confined to a single line; the intention is to make it easy to use tools like git grep to find exports. In general it is recommended export exactly one identifier per export statement. Exceptions may be made for certain tightly related identifiers, e.g. is_finite, set_is_finite and has_is_finite could be put on a single line. In general if multiple export statements appear in sequence, they must be sorted alphabetically.

However, as always, rules sometimes should be broken.

Optional arguments for parents of return values

Several objects in OSCAR have parents, e.g. polynomials, group elements, ... Whenever a function creates such objects from an input which does not involve the output's parent, we strongly recommend that the user should have the possibility to pass on this parent as a keyword argument under the name parent. Beyond that you can make more entry points for such parents available for the user's convenience.

Let's see an example. Say, you want to implement the characteristic polynomial of a matrix. You could do it as follows:

function characteristic_polynomial(A::MatrixElem)
  kk = base_ring(A)
  P, x = kk[:x]
  AP = change_base_ring(P, A)
  return det(AP - x*one(AP))
end

You can see that the polynomial ring P, i.e. the parent of the output, is newly created in the body of the function. In particular, calling this function two times on two different matrices A and B might produce incompatible polynomials p = det(A - x*one(A)) and q = det(B - x*one(B)) with different parents. Calling p + q will result in an error.

To solve this, we should have implemented the function differently:

# Implementation of the recommended keyword argument signature:
function characteristic_polynomial(
    A::MatrixElem;
    parent::AbstractAlgebra.Ring=polynomial_ring(base_ring(A), :t)[1]
  )
  AP = change_base_ring(parent, A)
  x = first(gens(ring))
  return det(AP - x*one(AP))
end

# Optional second signature to also allow for the specification of the 
# output's parent as the first argument:
function characteristic_polynomial(
    P::PolyRing,
    A::MatrixElem
  )
  coefficient_ring(P) === base_ring(A) || error("coefficient rings incompatible")
  return characteristic_polynomial(A, parent=P)
end

In fact this now allows for two different entry points for the parent ring P of the output: First as the required parent keyword argument and second as the first argument of a method of characteristic_polynomial with an extended signature. Note that within the scope of the first method's body the OSCAR function parent is necessarily overwritten by the name of the keyword argument. Hence to call the actual parent of any other object, you must then use Oscar.parent. E.g. to get the MatrixSpace of the matrix A, write Oscar.parent(A).

Documentation

In general we try to follow the list of recommendations in the Documentation section of the Julia manual.
Via the MathJax integration it is possible to use LaTeX code, and this is the preferred way to denote the mathematical symbols in the docstrings.

Printing in Oscar

The 2 + 1 print modes of Oscar

Oscar has two user print modes detailed and one line and one internal print mode terse. The latter is for use during recursion, e.g. to print the base_ring(X) when in one line mode. It exists to make sure that one line stays compact and human readable.

Top-level REPL printing of an object will use detailed mode by default

julia> X
detailed

Inside nested structures, e.g. inside a Vector, the one line mode is used.

julia> [X,X]
3-element Vector{TypeofX{T}}
one line
one line
one line

An Example for the 2 + 1 print modes

# detailed
General linear group of degree 24
  over Finite field of degree 7 over GF(29)

# one line
General linear group of degree 24 over GF(29^7)

# terse
General linear group

The print modes are specified as follows

Detailed printing

the output must make sense as a standalone without context to non-specialists
the number of output lines should fit in the terminal
if the object is simple enough use only one line
use indentation and (usually) one line to print substructures

One line printing

the output must print in one line
should make sense as a standalone without context
variable names/generators/relations should not be printed only their number.
Only the first word is capitalized e.g. Polynomial ring
one should use terse for nested printing in compact
nested calls to one line (if you think them really necessary) should be at the end, so that one can read sequentially. Calls to terse can be anywhere.
commas must be enclosed in brackets so that printing tuples stays unambiguous

Terse printing

a user readable version of the main (mathematical) type.
a single term or a symbol/letter mimicking mathematical notation
should usually only depend on the type and not of the type parameters or of the concrete instance - exceptions of this rule are possible e.g. for GF(2)
no nested printing. In particular variable names and base_ring must not be displayed. This ensures that one line and terse stay compact even for complicated things. If you want nested printing use one line or detailed.

For further information and examples we refer you to our section Details on printing in Oscar.

Deprecating functions

Sometimes it is necessary to rename a function or otherwise change it. To allow for backwards compatibility, please then introduce a new line in the file src/deprecations.jl. If the interface did not change, it is enough to write:

# Deprecated after CURRENT_RELEASE_VERSION
@deprecate old_function new_function

It is possible to transform the arguments too, if the syntax has changed. If this process needs an auxiliary function, which otherwise is unnecessary, please add it above:

# Deprecated after CURRENT_RELEASE_VERSION
function transform_args_for_new_function(args)
    # Do something
    return new_args
end
@deprecate old_function(arg1::Type1, arg2::Type2, ...) new_function(transform_args_for_new_function(args))

In simple cases (like changing the order of arguments), you don't need an auxiliary function:

@deprecate old_function(arg1::Type1, arg2::Type2) new_function(arg2, arg1)

The comment about the version number is only necessary if you are the first one adding to deprecations.jl after a release, otherwise please add to the existing block.

If you renamed a type and want to deprecate the old one, please add a line like

Base.@deprecate_type OldType NewType

This makes it still possible to use OldType in signatures and type annotations, but it will throw a deprecation warning (if they are enabled).

Note

Please make sure to change to the new function everywhere in the existing OSCAR code base. Even if you think, you were the only one using the function, run a quick grep to make sure. When you are done, deprecations.jl should be the only place mentioning old_function. To make sure, you can start Julia with --depwarn=yes or even --depwarn=error and then run the tests.

Approved abbreviations

Types for rings/groups/ideals/modules/... end with Ring/Group/Ideal/Module/...
Types for elements should have the same name as the type of the parent with Elem added;
- Exception: MatrixSpace elements end with Matrix.
We abbreviate certain parts of type names, according to a fixed set of substitutions; further abbreviations should be carefully decided upon.
Every abbreviation must be unique; e.g. Abs stands for Absolute, and so must not be used for e.g. Abstract.
List of approved abbreviations
- absolute -> Abs
  - abstract -> Abstract
- decorated -> Dec
- group -> Group
- ideal -> Ideal
- localized -> Loc
- matrix -> Matrix
- module -> Module
- multivariate polynomial -> MPoly
- polynomial -> Poly
- quotient -> Quo
- relative -> Rel
- ring ->Ring
- subquotient -> Subquo
If a type comes in sparse and dense variants, then call the dense type T and the sparse one SparseT.