AbstractAlgebra and Nemo have adopted a number of conventions to help maintain a uniform codebase.

Code conventions

Function and type names

Names of types in Julia follow the convention of CamelCase where the first letter of each word is capitalised, e.g. Int64 and AbstractString.

Function/method names in Julia use all lowercase with underscores between the words, e.g. zip and jacobi_symbol.

We follow these conventions in Nemo with some exceptions:

  • When interfacing C libraries the types use the same spelling and capitalisation in Nemo as they do in C, e.g. the Flint library's fmpz_poly remains uncapitalised in Nemo.

  • Types such as gfp_poly which don't exist under that name on the C side also use the lowercase convention as they wrap an actual C type which must be split into more than one type on the Julia side. For example nmod_poly and gfp_poly on the Julia side both represent Flint nmod_poly's on the C side.

  • Types of rings and fields, modules, maps, etc. are capitalised whether they correspond to a C type or not, e.g. FqNmodFiniteField for the type of an object representing the field that fq_nmod's belong to.


  • We omit an underscore if the first word of a method is "is" or "has", e.g. iseven.

  • Underscores are omitted if the method name is already well established without an underscore in Julia itself, e.g. setindex.

  • Constructors with the same name as a type use the same spelling and capitalisation as that type, e.g. fmpz(1).

  • Functions for creating rings, fields, modules, maps, etc. (rather than the elements thereof) use CamelCase, e.g. PolynomialRing. We refer to these functions as parent constructors. Note that we do not follow the Julia convention here, e.g. PolynomialRing is a function and not a type constructor (in fact we often return a tuple consisting of a parent object and other objects such as generators with this type of function) yet we capitalise it.

  • We prefer words to not be abbreviated, e.g. denominator instead of den.

  • Exceptions always exist where the result would be offensive in any major spoken language (example omitted).

It is easy to find counterexamples to virtually all these rules. However we have been making efforts to remove the most egregious cases from our codebase over time. As perfect consistency is not possible, work on this has to at times take a back seat.

Use of ASCII characters

All code and printed output in Nemo should use ASCII characters only. This is because we have developers who are using versions of the WSL that cannot correctly display non-ASCII characters.

This extends to function and operator names, which saves people having to learn how to enter them to use the system.

Spacing and tabs

All function bodies and control blocks should be indented using spaces.

A survey of existing code shows 2, 3 or 4 space indenting commonly used in our files. Values outside this range should not be used.

When contributing to an existing file, follow the majority convention in that file. Consistency within a file is valued highly.

If you are new to Nemo development and do not already have a very strong preference, new files should be started with 3 space indenting. This maximises the likelihood that copy and paste between files will be straightforward, though modern editors ease this to some degree.

Function signatures in docstrings should have four spaces before them.

Where possible, line lengths should not exceed 80 characters.

We use a term/factor convention for spacing. This means that all (additive) terms have spaces before and after them, (multiplicative) factors usually do not.

In practice this means that +, -, =, ==, !=, <, >, <=, >= all have spaces before and after them. The operators *, /, ^ and unary minus do not.

As per English, commas are followed by a single space in expressions. This applies for example to function arguments and tuples.

We do not put spaces immediately inside or before parentheses.

Colons used for ranges do not have spaces before or after them.

Logical operators, &, |, &&, etc. usually have spaces before and after them.


Despite appearances to the contrary, we now prefer code comments explaining the algorithm as it proceeds.

The hash when used for a comment should always be followed by a space. Full sentences are preferred.

We do not generally use comments in Nemo for questions, complaints or proposals for future improvement. These are better off in a ticket on GitHub with a discussion that will be brought to the attention of all relevant parties.

Any (necessary) limitations of the implementation should be noted in docstrings.

Layout of files

In Nemo, all types are places in special files with the word "Types" in their name, e.g. FlintTypes.jl. This is because Julia must be aware of all types before they are used. Separation of types from implementations makes it easy to ensure this happens.

Abstract types should be put in the file called AbstractTypes.jl at the top level of the src directory.

Most implementation files present functions in a particular order, which is as follows:

  • A header stating what the file is for, and if needed, any copyright notices

  • Functions applying to any "types" used in the file, e.g. parent_type, elem_type, base_ring, parent, check_parent.

  • Basic manipulation, including hashes, predicates, getters/setters, functions for creating special values (e.g. one, zero and the like), deepcopy_internal. These are usually fairly short functions, often a single line.

  • Indexing (getindex, setindex), iteration, views.

  • String I/O (expressify and file access, etc.)

  • Arithmetic operations, usually in multiple sections, such as unary operations, binary operations, ad hoc binary operations (e.g. multiplication of a complex object by a scalar), comparisons, ad hoc comparisons, division, etc.

  • More complex functionality separated into sections based on functionality provided, e.g. gcd, interpolation, special functions, solving, etc.

  • Functions for mapping between different types, coercion, changing base ring, etc.

  • Unsafe operators, e.g. mul!, add!, addeq! etc.

  • Random generation

  • Promotion rules

  • Parent object call overload (e.g. for implementing R(2) where R is an object representing a ring or field, etc.)

  • Additional constructors, e.g. matrix, which might be used instead of a parent object to construct elements.

  • Parent object constructors, e.g. PolynomialRing, etc.

The exact order within the file is less important than generally following something like the above. This aids in finding functions in a file since all files are more or less set out the same way.

For an example to follow, see the src/Poly.jl and src/generic/Poly.jl files in AbstractAlgebra which form the oldest and most canonical example.

Headings for sections should be 80 characters wide and formed of hashes in the style that can be seen in each Nemo file.