6.7 KiB
Nim Enhancement Proposal #1 - Standard Library Style Guide
Abstract
Although Nim, through its flexible AST and case-sensitivity settings, supports a variety of code and formatting styles, it is nevertheless beneficial that certain community efforts, such as the standard library, should follow a consistent set of style guidelines when suitable. This enhancement proposal aims to list a series of guidelines that the standard library should follow. Note that these are guidelines only. The nature of Nimrod being as flexible as it is, there will be parts of this style guide that don't make sense in certain contexts. Furthermore, just as Python's style guide changes over time, this style guide will too.
Style Guidelines
Spacing and Whitespace Conventions
-
Lines should be no longer than 80 characters. Limiting the amount of information present on each line makes for more readable code - the reader has smaller chunks to process.
-
2 spaces should be used for indentation of blocks; tabstops are not allowed (the compiler enforces this). Using spaces means that the appearance of code is more consistant across editors. Unlike spaces, tabstop width varies across editors, and not all editors provide means of changing this width.
-
Although use of whitespace for stylistic reasons other than the ones endorsed by this guide are allowed, careful thought should be put into such practices. Not all editors support automatic alignment of code sections, and re-aligning long sections of code by hand can quickly become tedius.
# This is bad, as the next time someone comes # to edit this code block, they # must re-align all the assignments again: type WordBool* = int16 CalType* = int ... # 5 lines later CalId* = int LongLong* = int64 LongLongPtr* = ptr LongLong
Naming Conventions
Note: While the rules outlined below are the current naming conventions, these conventions have not always been in place. Previously, the naming conventions for identifiers followed the Pascal tradition of prefixes which indicated the base type of the identifer - PFoo for pointer and reference types, TFoo for value types, EFoo for exceptions, etc. Though this has since changed, there are many places in the standard library which still use this convention. Such style remains in place purely for legacy reasons, and will be changed in the future.
-
Type identifiers should be in CamelCase. All other identifiers should be in pascalCase with the exception of constants which may use CamelCase but are not required to.
const aConstant = 42 const FOO_BAR = 4.2 var aVariable = "Meep" type FooBar = object
-
When naming types that come in value, pointer, and reference varieties, use a regular name for the variety that is to be used the most, and add a "Obj", "Ref", or "Ptr" suffix for the other varieties. If there is no single variety that will be used the most, add the suffixes to all versions.
type handle = int64 # Will be used most often handleRef = ref handle # Will be used less often
-
Exception and Error types should have the "Error" suffix.
type UnluckyError = object of E_Base
-
Unless marked with the
{.pure.}
pragma, members of enums should have an identifying prefix, such as an abbreviation of the enum's name. Since non-pure enum members can be referenced withouttype PathComponent = enum pcDir pcLinkToDir pcFile pcLinkToFile
Non-pure enum values should use pascalCase whereas pure enum values should use CamelCase.
Coding Conventions
-
The 'return' statement should only be used when it's control-flow properties are required. Use a procedures implicit 'result' variable instead. This improves readability.
-
Prefer to return
[]
and""
instead ofnil
, or throw an exception if that is appropriate. -
Use a proc when possible, only using the more powerful facilities of macros, templates, iterators, and converters when necessary.
-
Use the 'let' statement (not the var statement) when declaring variables that do not change within their scope. Using the let statement ensures that variables remain immutable, and gives those who read the code a better idea of the code's purpose.
Conventions for multi-line statements and expressions
-
Any tuple type declarations that are longer than one line should use the regular object type layout instead. This enhances the readability of the tuple declaration by splitting its members information across multiple lines.
type ShortTuple = tuple[a: int, b: string] ReallyLongTuple = tuple wordyTupleMemberOne: string wordyTupleMemberTwo: int wordyTupleMemberThree: double
-
Similarly, any procedure type declarations that are longer than one line should be formatted in the style of a regular type.
type EventCallback = proc ( timeRecieved: TTime errorCode: int event: Event )
-
Multi-line procedure declarations/argument lists should continue on the same column as the opening brace. This style is different from that of procedure type declarations in order to distinguish between the heading of a procedure and its body. If the procedure name is too long to make this style convenient, then one of the styles for multi-line procedure calls (or consider renaming your procedure).
proc lotsOfArguments(argOne: string, argTwo: int, argThree:float argFour: proc(), argFive:bool): int
-
Multi-line procedure calls should either have one argument per line (like multi-line type declarations) or continue on the same column as the opening parenthesis (like multi-line procedure declarations). It is suggested that the former style be used for procedure calls with complex argument structures, and the latter style for procedure calls with simpler argument structures.
# Each argument on a new line, like type declarations # Best suited for 'complex' procedure calls. readDirectoryChangesW( directoryHandle.THandle, buffer.start, bufferSize.int32, watchSubdir.WinBool, filterFlags, cast[ptr dword](nil), cast[POverlapped](ol), cast[LPOverlappedCompletionRoutine](nil) ) # Multiple arguments on new lines, aligned to the opening parenthesis # Best suited for 'simple' procedure calls startProcess(nimrodExecutable, currentDirectory, compilerArguments environment, processOptions)