Go is a new language. Although it's in the C family of languages it has some unusual properties that make effective Go programs different in character from programs in C, C++, or Java. To write Go well, it's important to understand its properties and idioms.
This document gives tips for writing clear, idiomatic Go code and points out common mistakes. It augments the language specification and the tutorial, both of which you should read first.
The first step in learning to write good code is to read good code. The Go package sources are intended to serve not only as the core library but also as examples of how to use the language. Read them and follow their example.
Formatting issues are the most contentious but the least consequential. People adapt to different formatting styles, but they shouldn't be asked to. Everyone should use the same formatting; as in English, consistent punctuation and spacing make the text easier to read. Most of the local formatting style can be picked up by reading existing Go programs, but to make them explicit here are some common points.
Use tabs, not spaces, for indentation.
Let tools such as gofmt take care of lining things up.
There should be no trailing white space at the end of lines.
Go has no 80-character limit. Don't bother with fancy line wrapping just because a line is wider than a punched card. If a line is too long, indent with an extra tab.
Go does not require parentheses around the expression
following the for, if, range,
switch, and return keywords.
Go provides C-style /* */ block comments
and C++-style // line comments.
Use line comments by default,
reserving block comments for top-level package comments
and commenting out large swaths of code.
Every package should have a package comment, a block comment preceding the package clause. It should introduce the package and provide information relevant to the package as a whole.
/*
The regexp package implements a simple library for
regular expressions.
The syntax of the regular expressions accepted is:
regexp:
concatenation { '|' concatenation }
concatenation:
{ closure }
closure:
term [ '*' | '+' | '?' ]
term:
'^'
'$'
'.'
character
'[' [ '^' ] character-ranges ']'
'(' regexp ')'
*/
package regexp
Consider how the package comment contributes to the appearance
of the godoc page for the package. Don't just
echo the doc comments for the components. The package comment
can be brief.
// The path package implements utility routines for // manipulating slash-separated filename paths.
If a comment immediately precedes a top-level declaration, the Go documentation server (TODO: that's not a public URL.) uses that comment as the documentation for the constant, function, method, package, type or variable being declared. These are called doc comments. To detach a comment from a declaration, insert a blank line between them.
Every exported (capitalized) name in a program should have a doc comment, as should the package declaration itself. If a name appears multiple times due to forward declarations or appearance in multiple source files within a package, only one instance requires a doc comment, and any one will do.
Doc comments consist of complete English sentences. The first sentence should be a one-sentence summary that starts with the name being declared:
// Quote returns a double-quoted Go string literal
// representing s. The returned string s uses Go escape
// sequences (\t, \n, \xFF, \u0100) for control characters
// and non-ASCII characters.
func Quote(s string) string {
Use of complete English sentences admits a wider variety of automated presentations.
Go programs are meant to read equally well using fixed-width and variable-width fonts. Don't use fancy formattings that depend on fixed-width fonts. In particular, don't assume that a single space is the same width as every other character. If you need to make a columnated table, use tabs to separate the columns and the pretty printer will make sure the columns are lined up properly in the output.
If you need comments to separate sections in a file, use a simple block comment:
/* * Helper routines for simplifying the fetching of optional * fields of basic type. If the field is missing, they return * the zero for the type. */or
/*
Helper routines for simplifying the fetching of optional
fields of basic type. If the field is missing, they return
the zero for the type.
*/
Comments are text, not HTML; they contain no markup. Refrain from ASCII embellishment like *this* or /this/.
Go's declaration syntax allows grouping of declarations. A comment can introduce a group of related constants or variables.
// Flags to Open, wrapping those of the underlying system. // Not all flags may be implemented on a given system. const ( O_RDONLY = syscall.O_RDONLY; // Open file read-only. O_WRONLY = syscall.O_WRONLY; // Open file write-only. ... )
A grouping can also indicate relationships between items, such as the fact that a set of variables is controlled by a mutex.
// Variables protected by countLock. var ( countLock sync.Mutex; inputCount uint32; outputCount uint32; errorCount uint32; )
Go uses the case of the first letter in a name to decide whether the name is visible in other packages. Multiword names use MixedCaps or mixedCaps rather than underscores.
Package names are lower case single-word names:
there should be no need for underscore or mixedCaps.
The package name is conventionally the base name of
the source directory: the package in src/pkg/container/vector
is installed as "container/vector" but has name vector,
not container_vector and not containerVector.
The package name is only the default name used
when importing the package; it need not be unique
across all source code.
A name's length should not exceed its information content.
For a function-local variable
in scope only for a few lines, the name i conveys just
as much information as index or idx and is easier to read.
Letters are easier to distinguish than numbers; use i and j
not i1 and i2.
Exported names must convey more information
because they appear far from their origin.
Even so, longer names are not always better,
and the package name can help convey information:
the buffered Reader is bufio.Reader, not bufio.BufReader.
Similarly, once.Do is as precise and evocative as
once.DoOrWaitUntilDone, and once.Do(f) reads
better than once.DoOrWaitUntilDone(f).
Encoding small essays into function names is not Go style;
clear names with good documentation is.
One-method interfaces are conventionally named by
the method name plus the -er suffix: Reader,
Writer, Formatter.
A few method names—Read, Write, Close, Flush, String—have
canonical signatures and meanings. To avoid confusion,
don't give your method one of those names unless it
has the same signature and meaning.
Conversely, if your type implements a method with the
same meaning as a method on a well-known type,
give it the same name and signature.
Some function-local variables have canonical names too.
Just as i is idiomatic in Go for an
index variable, n is idiomatic for a count, b for a []byte,
s for a string, r for a Reader,
err for an os.Error
and so on.
Don't mix shorthands: it is especially confusing to
have two different variables i and idx,
or n and cnt.
A struct literal is an expression that creates a new instance each time it is evaluated. The address of such an expression therefore points to a fresh instance each time. Use such expressions to avoid the repetition of filling out a data structure.
length := Point{x, y}.Abs();
// Prepare RPCMessage to send to server
rpc := &RPCMessage {
Version: 1,
Header: &RPCHeader {
Id: nextId(),
Signature: sign(body),
Method: method,
},
Body: body,
};
Array, slice, and map literals behave similarly, although it is unusual to need the address of a slice or map.
header, body, checksum := buf[0:20], buf[20:n-4], buf[n-4:n];
When an if statement doesn't flow into the next statement—that is,
the body ends in break, continue,
goto, or return—omit the else.
f, err := os.Open(name, os.O_RDONLY, 0);
if err != nil {
return err;
}
codeUsing(f);
Go's switch is more general than C's.
When an if-else if-else chain has three or more bodies,
or an if condition has a long list of alternatives,
it will be clearer if rewritten as a switch.
func unhex(c byte) byte {
switch {
case '0' <= c && c <= '9':
return c - '0'
case 'a' <= c && c <= 'f':
return c - 'a' + 10
case 'A' <= c && c <= 'F':
return c - 'A' + 10
}
return 0
}
go/src/pkg/http/url.go:
func shouldEscape(c byte) bool {
switch c {
case ' ', '?', '&', '=', '#', '+', '%':
return true
}
return false
}
go/src/pkg/bytes/bytes.go:
// Compare returns an integer comparing the two byte arrays
// lexicographically.
// The result will be 0 if a==b, -1 if a < b, and +1 if a > b
func Compare(a, b []byte) int {
for i := 0; i < len(a) && i < len(b); i++ {
switch {
case a[i] > b[i]:
return 1
case a[i] < b[i]:
return -1
}
}
switch {
case len(a) < len(b):
return -1
case len(a) > len(b):
return 1
}
return 0
}
Functions are great for factoring out common code, but if a function is only called once, ask whether it is necessary, especially if it is just a short wrapper around another function. This style is rampant in C++ code: wrappers call wrappers that call wrappers that call wrappers. This style hinders people trying to understand the program, not to mention computers trying to execute it.
If a function must return multiple values, it can do so directly. There is no need to pass a pointer to a return value.
os.Error, not bool
Especially in libraries, functions tend to have multiple error modes.
Instead of returning a boolean to signal success,
return an os.Error that describes the failure.
Even if there is only one failure mode now,
there may be more later.
Error cases tend to be simpler than non-error cases, and it helps readability when the non-error flow of control is always down the page. Also, error cases tend to end in jumps, so that there is no need for an explicit else.
if len(name) == 0 {
return os.EINVAL;
}
if IsDir(name) {
return os.EISDIR;
}
f, err := os.Open(name, os.O_RDONLY, 0);
if err != nil {
return err;
}
codeUsing(f);
os.Error should
describe the error and provide context.
For example, os.Open returns an os.PathError:
/src/pkg/os/file.go:
// PathError records an error and the operation and
// file path that caused it.
type PathError struct {
Op string;
Path string;
Error Error;
}
func (e *PathError) String() string {
return e.Op + " " + e.Path + ": " + e.Error.String();
}
PathError's String formats
the error nicely, including the operation and file name
tha failed; just printing the error generates a
message, such as
open /etc/passwx: no such file or directory
that is useful even if printed far from the call that triggered it.
Callers that care about the precise error details can
use a type switch or a type guard to look for specific
errors and extract details. For PathErrors
this might include examining the internal Error
to see if it is os.EPERM or os.ENOENT,
for instance.
NewTypeName for constructors
The constructor for the type pkg.MyType should
be named pkg.NewMyType and should return *pkg.MyType.
The implementation of NewTypeName often uses the
struct allocation idiom.
func NewFile(fd int, name string) *File {
if file < 0 {
return nil
}
return &File{fd, name, nil, 0}
}
Packages that export only a single type sometimes
shorten NewTypeName to New;
the vector constructor is
vector.New, not vector.NewVector.
A type that is intended to be allocated
as part of a larger struct may have an Init method
that must be called explicitly.
Conventionally, the Init method returns
the object being initialized, to make the constructor trivial:
func New(len int) *Vector {
return new(Vector).Init(len)
}
In Go, newly allocated memory and newly declared variables are zeroed. If a type is intended to be allocated without using a constructor (for example, as part of a larger struct or declared as a local variable), define the meaning of the zero value and arrange for that meaning to be useful.
For example, sync.Mutex does not
have an explicit constructor or Init method.
Instead, the zero value for a sync.Mutex
is defined to be an unlocked mutex.
If a type exists only to implement an interface and has no exported methods beyond that interface, there is no need to publish the type itself. Instead, write a constructor that returns an interface value.
For example, both crc32.NewIEEE() and adler32.New()
return type hash.Hash32.
Substituting the CRC-32 algorithm for Adler-32 in a Go program
requires only changing the constructor call:
the rest of the code is unaffected by the change of algorithm.
tables
XXX struct tags for marshalling. template eventually datafmt
Do not communicate by sharing memory; instead, share memory by communicating.
XXX, more here.
Tests should not stop early just because one case has misbehaved.
If at all possible, let tests continue, in order to characterize the
problem in more detail.
For example, it is more useful for a test to report that isPrime
gives the wrong answer for 2, 3, 5, and 7 (or for 2, 4, 8, and 16) than to report
that isPrime gives the wrong answer for 2 and therefore
no more tests were run.
If a test fails, print a concise message explaining the context, what happened, and what was expected. Many testing environments encourage causing the program to crash, but stack traces and core dumps have low signal to noise ratios and require reconstructing the situation from scratch. The programmer who triggers the test failure may be someone editing the code months later or even someone editing a different package on which the code depends. Time invested writing a good error message now pays off when the test breaks later.
Many tests reduce to running the same code multiple times,
with different input and expected output.
Instead of using cut and paste to write this code,
create a table of test cases and write a single test that
iterates over the table.
Once the table is written, you might find that it
serves well as input to multiple tests. For example,
a single table of encoded/decoded pairs can be
used by both TestEncoder and TestDecoder.
This data-driven style dominates in the Go package tests.
The reflect.DeepEqual function tests
whether two complex data structures have equal values.
If a function returns a complex data structure,
reflect.DeepEqual combined with table-driven testing
makes it easy to check that the return value is
exactly as expected.
Programmers often want their style to be distinctive, writing loops backwards or using custom spacing and naming conventions. Such idiosyncracies come at a price, however: by making the code look different, they make it harder to understand. Consistency trumps personal expression in programming.
If a program does the same thing twice, it should do it the same way both times. Conversely, if two different sections of a program look different, the reader will expect them to do different things.
Consider for loops.
Traditionally, a loop over n
elements begins:
for i := 0; i < n; i++ {
Much of the time, the loop could run in the opposite order and still be correct:
for i := n-1; i >= 0; i-- {
The convention is to count up unless to do so would be incorrect. A loop that counts down implicitly says “something special is happening here.” A reader who finds a program in which some loops count up and the rest count down will spend time trying to understand why.
Loop direction is just one
programming decision that must be made
consistently; others include
formatting, naming variables and methods,
whether a type
has a constructor, what tests look like, and so on.
Why is this variable called n here and cnt there?
Why is the Log constructor CreateLog when
the List constructor is NewList?
Why is this data structure initialized using
a structure literal when that one
is initialized using individual assignments?
These questions distract from the important one:
what does the code do?
Moreover, internal consistency is important not only within a single file,
but also within the the surrounding source files.
When editing code, read the surrounding context
and try to mimic it as much as possible, even if it
disagrees with the rules here.
It should not be possible to tell which lines
you wrote or edited based on style alone.
Consistency about little things
lets readers concentrate on big ones.