Gopls contains a driver for pluggable, modular static analyzers, such as those used by go vet.
Most analyzers report mistakes in your code; some suggest "quick fixes" that can be directly applied in your editor. Every time you edit your code, gopls re-runs its analyzers. Analyzer diagnostics help you detect bugs sooner, before you run your tests, or even before you save your files.
This document describes the suite of analyzers available in gopls, which aggregates analyzers from a variety of sources:
- all the usual bug-finding analyzers from the
go vetsuite (e.g.printf; seego tool vet helpfor the complete list); - a number of analyzers with more substantial dependencies that prevent them from being used in
go vet(e.g.nilness); - analyzers that augment compilation errors by suggesting quick fixes to common mistakes (e.g.
fillreturns); and - a handful of analyzers that suggest possible style improvements (e.g.
simplifyrange).
To enable or disable analyzers, use the analyses setting.
In addition, gopls includes the staticcheck suite.
When the staticcheck boolean option is
unset, slightly more than half of these analyzers are enabled by
default; this subset has been chosen for precision and efficiency. Set
staticcheck to true to enable the complete set, or to false to
disable the complete set.
Staticcheck analyzers, like all other analyzers, can be explicitly
enabled or disabled using the analyzers configuration setting; this
setting takes precedence over the staticcheck setting, so,
regardless of what value of staticcheck you use (true/false/unset),
you can make adjustments to your preferred set of analyzers.
Available since 2021.1
Default: off. Enable by setting "analyses": {"QF1001": true}.
Package documentation: QF1001
An untagged switch that compares a single variable against a series of values can be replaced with a tagged switch.
Before:
switch {
case x == 1 || x == 2, x == 3:
...
case x == 4:
...
default:
...
}
After:
switch x {
case 1, 2, 3:
...
case 4:
...
default:
...
}
Available since 2021.1
Default: on.
Package documentation: QF1002
A series of if/else-if checks comparing the same variable against values can be replaced with a tagged switch.
Before:
if x == 1 || x == 2 {
...
} else if x == 3 {
...
} else {
...
}
After:
switch x {
case 1, 2:
...
case 3:
...
default:
...
}
Available since 2021.1
Default: on.
Package documentation: QF1003
Available since 2021.1
Default: on.
Package documentation: QF1004
Some uses of math.Pow can be simplified to basic multiplication.
Before:
math.Pow(x, 2)
After:
x * x
Available since 2021.1
Default: off. Enable by setting "analyses": {"QF1005": true}.
Package documentation: QF1005
Before:
for {
if done {
break
}
...
}
After:
for !done {
...
}
Available since 2021.1
Default: off. Enable by setting "analyses": {"QF1006": true}.
Package documentation: QF1006
Before:
x := false
if someCondition {
x = true
}
After:
x := someCondition
Available since 2021.1
Default: off. Enable by setting "analyses": {"QF1007": true}.
Package documentation: QF1007
Available since 2021.1
Default: off. Enable by setting "analyses": {"QF1008": true}.
Package documentation: QF1008
Available since 2021.1
Default: on.
Package documentation: QF1009
Available since 2021.1
Default: on.
Package documentation: QF1010
Available since 2021.1
Default: off. Enable by setting "analyses": {"QF1011": true}.
Package documentation: QF1011
Available since 2022.1
Default: on.
Package documentation: QF1012
Select statements with a single case can be replaced with a simple send or receive.
Before:
select {
case x := <-ch:
fmt.Println(x)
}
After:
x := <-ch
fmt.Println(x)
Available since 2017.1
Default: on.
Package documentation: S1000
Use copy() for copying elements from one slice to another. For arrays of identical size, you can use simple assignment.
Before:
for i, x := range src {
dst[i] = x
}
After:
copy(dst, src)
Available since 2017.1
Default: on.
Package documentation: S1001
Before:
if x == true {}
After:
if x {}
Available since 2017.1
Default: off. Enable by setting "analyses": {"S1002": true}.
Package documentation: S1002
Before:
if strings.Index(x, y) != -1 {}
After:
if strings.Contains(x, y) {}
Available since 2017.1
Default: on.
Package documentation: S1003
Before:
if bytes.Compare(x, y) == 0 {}
After:
if bytes.Equal(x, y) {}
Available since 2017.1
Default: on.
Package documentation: S1004
In many cases, assigning to the blank identifier is unnecessary.
Before:
for _ = range s {}
x, _ = someMap[key]
_ = <-ch
After:
for range s{}
x = someMap[key]
<-ch
Available since 2017.1
Default: off. Enable by setting "analyses": {"S1005": true}.
Package documentation: S1005
For infinite loops, using for { ... } is the most idiomatic choice.
Available since 2017.1
Default: off. Enable by setting "analyses": {"S1006": true}.
Package documentation: S1006
Raw string literals use backticks instead of quotation marks and do not support any escape sequences. This means that the backslash can be used freely, without the need of escaping.
Since regular expressions have their own escape sequences, raw strings can improve their readability.
Before:
regexp.Compile("\\A(\\w+) profile: total \\d+\\n\\z")
After:
regexp.Compile(`\A(\w+) profile: total \d+\n\z`)
Available since 2017.1
Default: on.
Package documentation: S1007
Before:
if <expr> {
return true
}
return false
After:
return <expr>
Available since 2017.1
Default: off. Enable by setting "analyses": {"S1008": true}.
Package documentation: S1008
The len function is defined for all slices, maps, and channels, even nil ones, which have a length of zero. It is not necessary to check for nil before checking that their length is not zero.
Before:
if x != nil && len(x) != 0 {}
After:
if len(x) != 0 {}
Available since 2017.1
Default: on.
Package documentation: S1009
When slicing, the second index defaults to the length of the value, making s[n:len(s)] and s[n:] equivalent.
Available since 2017.1
Default: on.
Package documentation: S1010
Before:
for _, e := range y {
x = append(x, e)
}
for i := range y {
x = append(x, y[i])
}
for i := range y {
v := y[i]
x = append(x, v)
}
After:
x = append(x, y...)
x = append(x, y...)
x = append(x, y...)
Available since 2017.1
Default: off. Enable by setting "analyses": {"S1011": true}.
Package documentation: S1011
The time.Since helper has the same effect as using time.Now().Sub(x) but is easier to read.
Before:
time.Now().Sub(x)
After:
time.Since(x)
Available since 2017.1
Default: on.
Package documentation: S1012
Two struct types with identical fields can be converted between each other. In older versions of Go, the fields had to have identical struct tags. Since Go 1.8, however, struct tags are ignored during conversions. It is thus not necessary to manually copy every field individually.
Before:
var x T1
y := T2{
Field1: x.Field1,
Field2: x.Field2,
}
After:
var x T1
y := T2(x)
Available since 2017.1
Default: off. Enable by setting "analyses": {"S1016": true}.
Package documentation: S1016
Instead of using strings.HasPrefix and manual slicing, use the strings.TrimPrefix function. If the string doesn't start with the prefix, the original string will be returned. Using strings.TrimPrefix reduces complexity, and avoids common bugs, such as off-by-one mistakes.
Before:
if strings.HasPrefix(str, prefix) {
str = str[len(prefix):]
}
After:
str = strings.TrimPrefix(str, prefix)
Available since 2017.1
Default: on.
Package documentation: S1017
copy() permits using the same source and destination slice, even with overlapping ranges. This makes it ideal for sliding elements in a slice.
Before:
for i := 0; i < n; i++ {
bs[i] = bs[offset+i]
}
After:
copy(bs[:n], bs[offset:])
Available since 2017.1
Default: on.
Package documentation: S1018
The 'make' function has default values for the length and capacity arguments. For channels, the length defaults to zero, and for slices, the capacity defaults to the length.
Available since 2017.1
Default: on.
Package documentation: S1019
Before:
if _, ok := i.(T); ok && i != nil {}
After:
if _, ok := i.(T); ok {}
Available since 2017.1
Default: on.
Package documentation: S1020
Before:
var x uint
x = 1
After:
var x uint = 1
Available since 2017.1
Default: off. Enable by setting "analyses": {"S1021": true}.
Package documentation: S1021
Functions that have no return value do not need a return statement as the final statement of the function.
Switches in Go do not have automatic fallthrough, unlike languages like C. It is not necessary to have a break statement as the final statement in a case block.
Available since 2017.1
Default: on.
Package documentation: S1023
The time.Until helper has the same effect as using x.Sub(time.Now()) but is easier to read.
Before:
x.Sub(time.Now())
After:
time.Until(x)
Available since 2017.1
Default: on.
Package documentation: S1024
In many instances, there are easier and more efficient ways of getting a value's string representation. Whenever a value's underlying type is a string already, or the type has a String method, they should be used directly.
Given the following shared definitions
type T1 string
type T2 int
func (T2) String() string { return "Hello, world" }
var x string
var y T1
var z T2
we can simplify
fmt.Sprintf("%s", x)
fmt.Sprintf("%s", y)
fmt.Sprintf("%s", z)
to
x
string(y)
z.String()
Available since 2017.1
Default: off. Enable by setting "analyses": {"S1025": true}.
Package documentation: S1025
Before:
errors.New(fmt.Sprintf(...))
After:
fmt.Errorf(...)
Available since 2017.1
Default: on.
Package documentation: S1028
Ranging over a string will yield byte offsets and runes. If the offset isn't used, this is functionally equivalent to converting the string to a slice of runes and ranging over that. Ranging directly over the string will be more performant, however, as it avoids allocating a new slice, the size of which depends on the length of the string.
Before:
for _, r := range []rune(s) {}
After:
for _, r := range s {}
Available since 2017.1
Default: off. Enable by setting "analyses": {"S1029": true}.
Package documentation: S1029
bytes.Buffer has both a String and a Bytes method. It is almost never necessary to use string(buf.Bytes()) or []byte(buf.String()) – simply use the other method.
The only exception to this are map lookups. Due to a compiler optimization, m[string(buf.Bytes())] is more efficient than m[buf.String()].
Available since 2017.1
Default: on.
Package documentation: S1030
You can use range on nil slices and maps, the loop will simply never execute. This makes an additional nil check around the loop unnecessary.
Before:
if s != nil {
for _, x := range s {
...
}
}
After:
for _, x := range s {
...
}
Available since 2017.1
Default: on.
Package documentation: S1031
The sort.Ints, sort.Float64s and sort.Strings functions are easier to read than sort.Sort(sort.IntSlice(x)), sort.Sort(sort.Float64Slice(x)) and sort.Sort(sort.StringSlice(x)).
Before:
sort.Sort(sort.StringSlice(x))
After:
sort.Strings(x)
Available since 2019.1
Default: on.
Package documentation: S1032
Calling delete on a nil map is a no-op.
Available since 2019.2
Default: on.
Package documentation: S1033
Available since 2019.2
Default: on.
Package documentation: S1034
The methods on net/http.Header, namely Add, Del, Get and Set, already canonicalize the given header name.
Available since 2020.1
Default: on.
Package documentation: S1035
When accessing a map key that doesn't exist yet, one receives a zero value. Often, the zero value is a suitable value, for example when using append or doing integer math.
The following
if _, ok := m["foo"]; ok {
m["foo"] = append(m["foo"], "bar")
} else {
m["foo"] = []string{"bar"}
}
can be simplified to
m["foo"] = append(m["foo"], "bar")
and
if _, ok := m2["k"]; ok {
m2["k"] += 4
} else {
m2["k"] = 4
}
can be simplified to
m["k"] += 4
Available since 2020.1
Default: on.
Package documentation: S1036
Using a select statement with a single case receiving from the result of time.After is a very elaborate way of sleeping that can much simpler be expressed with a simple call to time.Sleep.
Available since 2020.1
Default: on.
Package documentation: S1037
Instead of using fmt.Print(fmt.Sprintf(...)), one can use fmt.Printf(...).
Available since 2020.1
Default: on.
Package documentation: S1038
Calling fmt.Sprint with a single string argument is unnecessary and identical to using the string directly.
Available since 2020.1
Default: on.
Package documentation: S1039
The type assertion x.(SomeInterface), when x already has type SomeInterface, can only fail if x is nil. Usually, this is left-over code from when x had a different type and you can safely delete the type assertion. If you want to check that x is not nil, consider being explicit and using an actual if x == nil comparison instead of relying on the type assertion panicking.
Available since 2021.1
Default: on.
Package documentation: S1040
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA1000": true}.
Package documentation: SA1000
Available since 2017.1
Default: on.
Package documentation: SA1001
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA1002": true}.
Package documentation: SA1002
The encoding/binary package can only serialize types with known sizes. This precludes the use of the int and uint types, as their sizes differ on different architectures. Furthermore, it doesn't support serializing maps, channels, strings, or functions.
Before Go 1.8, bool wasn't supported, either.
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA1003": true}.
Package documentation: SA1003
The time.Sleep function takes a time.Duration as its only argument. Durations are expressed in nanoseconds. Thus, calling time.Sleep(1) will sleep for 1 nanosecond. This is a common source of bugs, as sleep functions in other languages often accept seconds or milliseconds.
The time package provides constants such as time.Second to express large durations. These can be combined with arithmetic to express arbitrary durations, for example 5 * time.Second for 5 seconds.
If you truly meant to sleep for a tiny amount of time, use n * time.Nanosecond to signal to Staticcheck that you did mean to sleep for some amount of nanoseconds.
Available since 2017.1
Default: on.
Package documentation: SA1004
os/exec runs programs directly (using variants of the fork and exec system calls on Unix systems). This shouldn't be confused with running a command in a shell. The shell will allow for features such as input redirection, pipes, and general scripting. The shell is also responsible for splitting the user's input into a program name and its arguments. For example, the equivalent to
ls / /tmp
would be
exec.Command("ls", "/", "/tmp")
If you want to run a command in a shell, consider using something like the following – but be aware that not all systems, particularly Windows, will have a /bin/sh program:
exec.Command("/bin/sh", "-c", "ls | grep Awesome")
Available since 2017.1
Default: on.
Package documentation: SA1005
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA1007": true}.
Package documentation: SA1007
Keys in http.Header maps are canonical, meaning they follow a specific combination of uppercase and lowercase letters. Methods such as http.Header.Add and http.Header.Del convert inputs into this canonical form before manipulating the map.
When manipulating http.Header maps directly, as opposed to using the provided methods, care should be taken to stick to canonical form in order to avoid inconsistencies. The following piece of code demonstrates one such inconsistency:
h := http.Header{}
h["etag"] = []string{"1234"}
h.Add("etag", "5678")
fmt.Println(h)
// Output:
// map[Etag:[5678] etag:[1234]]
The easiest way of obtaining the canonical form of a key is to use http.CanonicalHeaderKey.
Available since 2017.1
Default: on.
Package documentation: SA1008
If n >= 0, the function returns at most n matches/submatches. To return all results, specify a negative number.
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA1010": true}.
Package documentation: SA1010
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA1011": true}.
Package documentation: SA1011
Available since 2017.1
Default: on.
Package documentation: SA1012
SA1013: io.Seeker.Seek is being called with the whence constant as the first argument, but it should be the second
Available since 2017.1
Default: on.
Package documentation: SA1013
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA1014": true}.
Package documentation: SA1014
SA1015: Using time.Tick in a way that will leak. Consider using time.NewTicker, and only use time.Tick in tests, commands and endless functions
Before Go 1.23, time.Tickers had to be closed to be able to be garbage collected. Since time.Tick doesn't make it possible to close the underlying ticker, using it repeatedly would leak memory.
Go 1.23 fixes this by allowing tickers to be collected even if they weren't closed.
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA1015": true}.
Package documentation: SA1015
Not all signals can be intercepted by a process. Specifically, on UNIX-like systems, the syscall.SIGKILL and syscall.SIGSTOP signals are never passed to the process, but instead handled directly by the kernel. It is therefore pointless to try and handle these signals.
Available since 2017.1
Default: on.
Package documentation: SA1016
The os/signal package uses non-blocking channel sends when delivering signals. If the receiving end of the channel isn't ready and the channel is either unbuffered or full, the signal will be dropped. To avoid missing signals, the channel should be buffered and of the appropriate size. For a channel used for notification of just one signal value, a buffer of size 1 is sufficient.
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA1017": true}.
Package documentation: SA1017
With n == 0, zero instances will be replaced. To replace all instances, use a negative number, or use strings.ReplaceAll.
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA1018": true}.
Package documentation: SA1018
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA1020": true}.
Package documentation: SA1020
A net.IP stores an IPv4 or IPv6 address as a slice of bytes. The length of the slice for an IPv4 address, however, can be either 4 or 16 bytes long, using different ways of representing IPv4 addresses. In order to correctly compare two net.IPs, the net.IP.Equal method should be used, as it takes both representations into account.
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA1021": true}.
Package documentation: SA1021
Write must not modify the slice data, even temporarily.
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA1023": true}.
Package documentation: SA1023
The strings.TrimLeft and strings.TrimRight functions take cutsets, not prefixes. A cutset is treated as a set of characters to remove from a string. For example,
strings.TrimLeft("42133word", "1234")
will result in the string "word" – any characters that are 1, 2, 3 or 4 are cut from the left of the string.
In order to remove one string from another, use strings.TrimPrefix instead.
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA1024": true}.
Package documentation: SA1024
Available since 2019.1
Default: off. Enable by setting "analyses": {"SA1025": true}.
Package documentation: SA1025
Available since 2019.2
Default: off. Enable by setting "analyses": {"SA1026": true}.
Package documentation: SA1026
On ARM, x86-32, and 32-bit MIPS, it is the caller's responsibility to arrange for 64-bit alignment of 64-bit words accessed atomically. The first word in a variable or in an allocated struct, array, or slice can be relied upon to be 64-bit aligned.
You can use the structlayout tool to inspect the alignment of fields in a struct.
Available since 2019.2
Default: off. Enable by setting "analyses": {"SA1027": true}.
Package documentation: SA1027
The first argument of sort.Slice must be a slice.
Available since 2020.1
Default: off. Enable by setting "analyses": {"SA1028": true}.
Package documentation: SA1028
The provided key must be comparable and should not be of type string or any other built-in type to avoid collisions between packages using context. Users of WithValue should define their own types for keys.
To avoid allocating when assigning to an interface{}, context keys often have concrete type struct{}. Alternatively, exported context key variables' static type should be a pointer or interface.
Available since 2020.1
Default: off. Enable by setting "analyses": {"SA1029": true}.
Package documentation: SA1029
This check validates the format, number base and bit size arguments of the various parsing and formatting functions in strconv.
Available since 2021.1
Default: off. Enable by setting "analyses": {"SA1030": true}.
Package documentation: SA1030
In an encoding function of the form Encode(dst, src), dst and src were found to reference the same memory. This can result in src bytes being overwritten before they are read, when the encoder writes more than one byte per src byte.
Available since 2024.1
Default: off. Enable by setting "analyses": {"SA1031": true}.
Package documentation: SA1031
The first argument of the function errors.Is is the error that we have and the second argument is the error we're trying to match against. For example:
if errors.Is(err, io.EOF) { ... }
This check detects some cases where the two arguments have been swapped. It flags any calls where the first argument is referring to a package-level error variable, such as
if errors.Is(io.EOF, err) { /* this is wrong */ }
Available since 2024.1
Default: off. Enable by setting "analyses": {"SA1032": true}.
Package documentation: SA1032
Empty critical sections of the kind
mu.Lock()
mu.Unlock()
are very often a typo, and the following was intended instead:
mu.Lock()
defer mu.Unlock()
Do note that sometimes empty critical sections can be useful, as a form of signaling to wait on another goroutine. Many times, there are simpler ways of achieving the same effect. When that isn't the case, the code should be amply commented to avoid confusion. Combining such comments with a //lint:ignore directive can be used to suppress this rare false positive.
Available since 2017.1
Default: on.
Package documentation: SA2001
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA2002": true}.
Package documentation: SA2002
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA2003": true}.
Package documentation: SA2003
Test executables (and in turn 'go test') exit with a non-zero status code if any tests failed. When specifying your own TestMain function, it is your responsibility to arrange for this, by calling os.Exit with the correct code. The correct code is returned by (*testing.M).Run, so the usual way of implementing TestMain is to end it with os.Exit(m.Run()).
Available since 2017.1
Default: on.
Package documentation: SA3000
The testing package dynamically sets b.N to improve the reliability of benchmarks and uses it in computations to determine the duration of a single operation. Benchmark code must not alter b.N as this would falsify results.
Available since 2017.1
Default: on.
Package documentation: SA3001
Available since 2017.1
Default: on.
Package documentation: SA4000
Available since 2017.1
Default: on.
Package documentation: SA4001
Available since 2017.1
Default: on.
Package documentation: SA4003
Available since 2017.1
Default: on.
Package documentation: SA4004
Available since 2021.1
Default: off. Enable by setting "analyses": {"SA4005": true}.
Package documentation: SA4005
SA4006: A value assigned to a variable is never read before being overwritten. Forgotten error check or dead code?
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA4006": true}.
Package documentation: SA4006
For example:
for i := 0; i < 10; j++ { ... }
This may also occur when a loop can only execute once because of unconditional control flow that terminates the loop. For example, when a loop body contains an unconditional break, return, or panic:
func f() {
panic("oops")
}
func g() {
for i := 0; i < 10; i++ {
// f unconditionally calls panic, which means "i" is
// never incremented.
f()
}
}
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA4008": true}.
Package documentation: SA4008
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA4009": true}.
Package documentation: SA4009
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA4010": true}.
Package documentation: SA4010
Available since 2017.1
Default: on.
Package documentation: SA4011
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA4012": true}.
Package documentation: SA4012
SA4013: Negating a boolean twice (!!b) is the same as writing b. This is either redundant, or a typo.
Available since 2017.1
Default: on.
Package documentation: SA4013
SA4014: An if/else if chain has repeated conditions and no side-effects; if the condition didn't match the first time, it won't match the second time, either
Available since 2017.1
Default: on.
Package documentation: SA4014
SA4015: Calling functions like math.Ceil on floats converted from integers doesn't do anything useful
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA4015": true}.
Package documentation: SA4015
Available since 2017.1
Default: on.
Package documentation: SA4016
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA4017": true}.
Package documentation: SA4017
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA4018": true}.
Package documentation: SA4018
Available since 2017.1
Default: on.
Package documentation: SA4019
In a type switch like the following
type T struct{}
func (T) Read(b []byte) (int, error) { return 0, nil }
var v interface{} = T{}
switch v.(type) {
case io.Reader:
// ...
case T:
// unreachable
}
the second case clause can never be reached because T implements io.Reader and case clauses are evaluated in source order.
Another example:
type T struct{}
func (T) Read(b []byte) (int, error) { return 0, nil }
func (T) Close() error { return nil }
var v interface{} = T{}
switch v.(type) {
case io.Reader:
// ...
case io.ReadCloser:
// unreachable
}
Even though T has a Close method and thus implements io.ReadCloser, io.Reader will always match first. The method set of io.Reader is a subset of io.ReadCloser. Thus it is impossible to match the second case without matching the first case.
Structurally equivalent interfaces
A special case of the previous example are structurally identical interfaces. Given these declarations
type T error
type V error
func doSomething() error {
err, ok := doAnotherThing()
if ok {
return T(err)
}
return U(err)
}
the following type switch will have an unreachable case clause:
switch doSomething().(type) {
case T:
// ...
case V:
// unreachable
}
T will always match before V because they are structurally equivalent and therefore doSomething()'s return value implements both.
Available since 2019.2
Default: on.
Package documentation: SA4020
Code such as 'if &x == nil' is meaningless, because taking the address of a variable always yields a non-nil pointer.
Available since 2020.1
Default: on.
Package documentation: SA4022
Under the covers, interfaces are implemented as two elements, a type T and a value V. V is a concrete value such as an int, struct or pointer, never an interface itself, and has type T. For instance, if we store the int value 3 in an interface, the resulting interface value has, schematically, (T=int, V=3). The value V is also known as the interface's dynamic value, since a given interface variable might hold different values V (and corresponding types T) during the execution of the program.
An interface value is nil only if the V and T are both unset, (T=nil, V is not set), In particular, a nil interface will always hold a nil type. If we store a nil pointer of type *int inside an interface value, the inner type will be *int regardless of the value of the pointer: (T=*int, V=nil). Such an interface value will therefore be non-nil even when the pointer value V inside is nil.
This situation can be confusing, and arises when a nil value is stored inside an interface value such as an error return:
func returnsError() error {
var p *MyError = nil
if bad() {
p = ErrBad
}
return p // Will always return a non-nil error.
}
If all goes well, the function returns a nil p, so the return value is an error interface value holding (T=*MyError, V=nil). This means that if the caller compares the returned error to nil, it will always look as if there was an error even if nothing bad happened. To return a proper nil error to the caller, the function must return an explicit nil:
func returnsError() error {
if bad() {
return ErrBad
}
return nil
}
It's a good idea for functions that return errors always to use the error type in their signature (as we did above) rather than a concrete type such as *MyError, to help guarantee the error is created correctly. As an example, os.Open returns an error even though, if not nil, it's always of concrete type *os.PathError.
Similar situations to those described here can arise whenever interfaces are used. Just keep in mind that if any concrete value has been stored in the interface, the interface will not be nil. For more information, see The Laws of Reflection at https://golang.org/doc/articles/laws_of_reflection.html.
This text has been copied from https://golang.org/doc/faq#nil_error, licensed under the Creative Commons Attribution 3.0 License.
Available since 2020.2
Default: off. Enable by setting "analyses": {"SA4023": true}.
Package documentation: SA4023
Return values of the len and cap builtins cannot be negative.
See https://golang.org/pkg/builtin/#len and https://golang.org/pkg/builtin/#cap.
Example:
if len(slice) < 0 {
fmt.Println("unreachable code")
}
Available since 2021.1
Default: on.
Package documentation: SA4024
When dividing two integer constants, the result will also be an integer. Thus, a division such as 2 / 3 results in 0. This is true for all of the following examples:
_ = 2 / 3
const _ = 2 / 3
const _ float64 = 2 / 3
_ = float64(2 / 3)
Staticcheck will flag such divisions if both sides of the division are integer literals, as it is highly unlikely that the division was intended to truncate to zero. Staticcheck will not flag integer division involving named constants, to avoid noisy positives.
Available since 2021.1
Default: on.
Package documentation: SA4025
In IEEE 754 floating point math, zero has a sign and can be positive or negative. This can be useful in certain numerical code.
Go constants, however, cannot express negative zero. This means that the literals -0.0 and 0.0 have the same ideal value (zero) and will both represent positive zero at runtime.
To explicitly and reliably create a negative zero, you can use the math.Copysign function: math.Copysign(0, -1).
Available since 2021.1
Default: on.
Package documentation: SA4026
(*net/url.URL).Query parses the current value of net/url.URL.RawQuery and returns it as a map of type net/url.Values. Subsequent changes to this map will not affect the URL unless the map gets encoded and assigned to the URL's RawQuery.
As a consequence, the following code pattern is an expensive no-op: u.Query().Add(key, value).
Available since 2021.1
Default: on.
Package documentation: SA4027
Available since 2022.1
Default: on.
Package documentation: SA4028
sort.Float64Slice, sort.IntSlice, and sort.StringSlice are types, not functions. Doing x = sort.StringSlice(x) does nothing, especially not sort any values. The correct usage is sort.Sort(sort.StringSlice(x)) or sort.StringSlice(x).Sort(), but there are more convenient helpers, namely sort.Float64s, sort.Ints, and sort.Strings.
Available since 2022.1
Default: on.
Package documentation: SA4029
Functions in the math/rand package that accept upper limits, such as Intn, generate random numbers in the half-open interval [0,n). In other words, the generated numbers will be >= 0 and < n – they don't include n. rand.Intn(1) therefore doesn't generate 0 or 1, it always generates 0.
Available since 2022.1
Default: on.
Package documentation: SA4030
Available since 2022.1
Default: off. Enable by setting "analyses": {"SA4031": true}.
Package documentation: SA4031
Available since 2024.1
Default: on.
Package documentation: SA4032
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA5000": true}.
Package documentation: SA5000
Available since 2017.1
Default: on.
Package documentation: SA5001
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA5002": true}.
Package documentation: SA5002
Defers are scoped to the surrounding function, not the surrounding block. In a function that never returns, i.e. one containing an infinite loop, defers will never execute.
Available since 2017.1
Default: on.
Package documentation: SA5003
Available since 2017.1
Default: on.
Package documentation: SA5004
A finalizer is a function associated with an object that runs when the garbage collector is ready to collect said object, that is when the object is no longer referenced by anything.
If the finalizer references the object, however, it will always remain as the final reference to that object, preventing the garbage collector from collecting the object. The finalizer will never run, and the object will never be collected, leading to a memory leak. That is why the finalizer should instead use its first argument to operate on the object. That way, the number of references can temporarily go to zero before the object is being passed to the finalizer.
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA5005": true}.
Package documentation: SA5005
A function that calls itself recursively needs to have an exit condition. Otherwise it will recurse forever, until the system runs out of memory.
This issue can be caused by simple bugs such as forgetting to add an exit condition. It can also happen "on purpose". Some languages have tail call optimization which makes certain infinite recursive calls safe to use. Go, however, does not implement TCO, and as such a loop should be used instead.
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA5007": true}.
Package documentation: SA5007
Available since 2019.2
Default: on.
Package documentation: SA5008
Some type assertions can be statically proven to be impossible. This is the case when the method sets of both arguments of the type assertion conflict with each other, for example by containing the same method with different signatures.
The Go compiler already applies this check when asserting from an interface value to a concrete type. If the concrete type misses methods from the interface, or if function signatures don't match, then the type assertion can never succeed.
This check applies the same logic when asserting from one interface to another. If both interface types contain the same method but with different signatures, then the type assertion can never succeed, either.
Available since 2020.1
Default: off. Enable by setting "analyses": {"SA5010": true}.
Package documentation: SA5010
A pointer is being dereferenced unconditionally, while also being checked against nil in another place. This suggests that the pointer may be nil and dereferencing it may panic. This is commonly a result of improperly ordered code or missing return statements. Consider the following examples:
func fn(x *int) {
fmt.Println(*x)
// This nil check is equally important for the previous dereference
if x != nil {
foo(*x)
}
}
func TestFoo(t *testing.T) {
x := compute()
if x == nil {
t.Errorf("nil pointer received")
}
// t.Errorf does not abort the test, so if x is nil, the next line will panic.
foo(*x)
}
Staticcheck tries to deduce which functions abort control flow. For example, it is aware that a function will not continue execution after a call to panic or log.Fatal. However, sometimes this detection fails, in particular in the presence of conditionals. Consider the following example:
func Log(msg string, level int) {
fmt.Println(msg)
if level == levelFatal {
os.Exit(1)
}
}
func Fatal(msg string) {
Log(msg, levelFatal)
}
func fn(x *int) {
if x == nil {
Fatal("unexpected nil pointer")
}
fmt.Println(*x)
}
Staticcheck will flag the dereference of x, even though it is perfectly safe. Staticcheck is not able to deduce that a call to Fatal will exit the program. For the time being, the easiest workaround is to modify the definition of Fatal like so:
func Fatal(msg string) {
Log(msg, levelFatal)
panic("unreachable")
}
We also hard-code functions from common logging packages such as logrus. Please file an issue if we're missing support for a popular package.
Available since 2020.1
Default: off. Enable by setting "analyses": {"SA5011": true}.
Package documentation: SA5011
Some functions that take slices as parameters expect the slices to have an even number of elements. Often, these functions treat elements in a slice as pairs. For example, strings.NewReplacer takes pairs of old and new strings, and calling it with an odd number of elements would be an error.
Available since 2020.2
Default: off. Enable by setting "analyses": {"SA5012": true}.
Package documentation: SA5012
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA6000": true}.
Package documentation: SA6000
Map keys must be comparable, which precludes the use of byte slices. This usually leads to using string keys and converting byte slices to strings.
Normally, a conversion of a byte slice to a string needs to copy the data and causes allocations. The compiler, however, recognizes m[string(b)] and uses the data of b directly, without copying it, because it knows that the data can't change during the map lookup. This leads to the counter-intuitive situation that
k := string(b)
println(m[k])
println(m[k])
will be less efficient than
println(m[string(b)])
println(m[string(b)])
because the first version needs to copy and allocate, while the second one does not.
For some history on this optimization, check out commit f5f5a8b6209f84961687d993b93ea0d397f5d5bf in the Go repository.
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA6001": true}.
Package documentation: SA6001
A sync.Pool is used to avoid unnecessary allocations and reduce the amount of work the garbage collector has to do.
When passing a value that is not a pointer to a function that accepts an interface, the value needs to be placed on the heap, which means an additional allocation. Slices are a common thing to put in sync.Pools, and they're structs with 3 fields (length, capacity, and a pointer to an array). In order to avoid the extra allocation, one should store a pointer to the slice instead.
See the comments on https://go-review.googlesource.com/c/go/+/24371 that discuss this problem.
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA6002": true}.
Package documentation: SA6002
You may want to loop over the runes in a string. Instead of converting the string to a slice of runes and looping over that, you can loop over the string itself. That is,
for _, r := range s {}
and
for _, r := range []rune(s) {}
will yield the same values. The first version, however, will be faster and avoid unnecessary memory allocations.
Do note that if you are interested in the indices, ranging over a string and over a slice of runes will yield different indices. The first one yields byte offsets, while the second one yields indices in the slice of runes.
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA6003": true}.
Package documentation: SA6003
Converting two strings to the same case and comparing them like so
if strings.ToLower(s1) == strings.ToLower(s2) {
...
}
is significantly more expensive than comparing them with strings.EqualFold(s1, s2). This is due to memory usage as well as computational complexity.
strings.ToLower will have to allocate memory for the new strings, as well as convert both strings fully, even if they differ on the very first byte. strings.EqualFold, on the other hand, compares the strings one character at a time. It doesn't need to create two intermediate strings and can return as soon as the first non-matching character has been found.
For a more in-depth explanation of this issue, see https://blog.digitalocean.com/how-to-efficiently-compare-strings-in-go/
Available since 2019.2
Default: on.
Package documentation: SA6005
Using io.WriteString to write a slice of bytes, as in
io.WriteString(w, string(b))
is both unnecessary and inefficient. Converting from []byte to string has to allocate and copy the data, and we could simply use w.Write(b) instead.
Available since 2024.1
Default: on.
Package documentation: SA6006
Available since 2017.1
Default: off. Enable by setting "analyses": {"SA9001": true}.
Package documentation: SA9001
Available since 2017.1
Default: on.
Package documentation: SA9002
Available since 2017.1, non-default
Default: off. Enable by setting "analyses": {"SA9003": true}.
Package documentation: SA9003
In a constant declaration such as the following:
const (
First byte = 1
Second = 2
)
the constant Second does not have the same type as the constant First. This construct shouldn't be confused with
const (
First byte = iota
Second
)
where First and Second do indeed have the same type. The type is only passed on when no explicit value is assigned to the constant.
When declaring enumerations with explicit values it is therefore important not to write
const (
EnumFirst EnumType = 1
EnumSecond = 2
EnumThird = 3
)
This discrepancy in types can cause various confusing behaviors and bugs.
Wrong type in variable declarations
The most obvious issue with such incorrect enumerations expresses itself as a compile error:
package pkg
const (
EnumFirst uint8 = 1
EnumSecond = 2
)
func fn(useFirst bool) {
x := EnumSecond
if useFirst {
x = EnumFirst
}
}
fails to compile with
./const.go:11:5: cannot use EnumFirst (type uint8) as type int in assignment
Losing method sets
A more subtle issue occurs with types that have methods and optional interfaces. Consider the following:
package main
import "fmt"
type Enum int
func (e Enum) String() string {
return "an enum"
}
const (
EnumFirst Enum = 1
EnumSecond = 2
)
func main() {
fmt.Println(EnumFirst)
fmt.Println(EnumSecond)
}
This code will output
an enum
2
as EnumSecond has no explicit type, and thus defaults to int.
Available since 2019.1
Default: on.
Package documentation: SA9004
The encoding/json and encoding/xml packages only operate on exported fields in structs, not unexported ones. It is usually an error to try to (un)marshal structs that only consist of unexported fields.
This check will not flag calls involving types that define custom marshaling behavior, e.g. via MarshalJSON methods. It will also not flag empty structs.
Available since 2019.2
Default: off. Enable by setting "analyses": {"SA9005": true}.
Package documentation: SA9005
Bit shifting a value past its size will always clear the value.
For instance:
v := int8(42)
v >>= 8
will always result in 0.
This check flags bit shifting operations on fixed size integer values only. That is, int, uint and uintptr are never flagged to avoid potential false positives in somewhat exotic but valid bit twiddling tricks:
// Clear any value above 32 bits if integers are more than 32 bits.
func f(i int) int {
v := i >> 32
v = v << 32
return i-v
}
Available since 2020.2
Default: on.
Package documentation: SA9006
It is virtually never correct to delete system directories such as /tmp or the user's home directory. However, it can be fairly easy to do by mistake, for example by mistakenly using os.TempDir instead of ioutil.TempDir, or by forgetting to add a suffix to the result of os.UserHomeDir.
Writing
d := os.TempDir()
defer os.RemoveAll(d)
in your unit tests will have a devastating effect on the stability of your system.
This check flags attempts at deleting the following directories:
- os.TempDir
- os.UserCacheDir
- os.UserConfigDir
- os.UserHomeDir
Available since 2022.1
Default: off. Enable by setting "analyses": {"SA9007": true}.
Package documentation: SA9007
When declaring variables as part of an if statement (like in 'if foo := ...; foo {'), the same variables will also be in the scope of the else branch. This means that in the following example
if x, ok := x.(int); ok {
// ...
} else {
fmt.Printf("unexpected type %T", x)
}
x in the else branch will refer to the x from x, ok :=; it will not refer to the x that is being type-asserted. The result of a failed type assertion is the zero value of the type that is being asserted to, so x in the else branch will always have the value 0 and the type int.
Available since 2022.1
Default: off. Enable by setting "analyses": {"SA9008": true}.
Package documentation: SA9008
A potential Go compiler directive was found, but is ineffectual as it begins with whitespace.
Available since 2024.1
Default: on.
Package documentation: SA9009
Packages must have a package comment that is formatted according to the guidelines laid out in https://go.dev/wiki/CodeReviewComments#package-comments.
Available since 2019.1, non-default
Default: off. Enable by setting "analyses": {"ST1000": true}.
Package documentation: ST1000
Dot imports that aren't in external test packages are discouraged.
The dot_import_whitelist option can be used to whitelist certain imports.
Quoting Go Code Review Comments:
The import . form can be useful in tests that, due to circular dependencies, cannot be made part of the package being tested:
package foo_test import ( "bar/testutil" // also imports "foo" . "foo" )In this case, the test file cannot be in package foo because it uses bar/testutil, which imports foo. So we use the import . form to let the file pretend to be part of package foo even though it is not. Except for this one case, do not use import . in your programs. It makes the programs much harder to read because it is unclear whether a name like Quux is a top-level identifier in the current package or in an imported package.
Available since 2019.1
Options dot_import_whitelist
Default: off. Enable by setting "analyses": {"ST1001": true}.
Package documentation: ST1001
Identifiers, such as variable and package names, follow certain rules.
See the following links for details:
- https://go.dev/doc/effective_go#package-names
- https://go.dev/doc/effective_go#mixed-caps
- https://go.dev/wiki/CodeReviewComments#initialisms
- https://go.dev/wiki/CodeReviewComments#variable-names
Available since 2019.1, non-default
Options initialisms
Default: off. Enable by setting "analyses": {"ST1003": true}.
Package documentation: ST1003
Error strings follow a set of guidelines to ensure uniformity and good composability.
Quoting Go Code Review Comments:
Error strings should not be capitalized (unless beginning with proper nouns or acronyms) or end with punctuation, since they are usually printed following other context. That is, use fmt.Errorf("something bad") not fmt.Errorf("Something bad"), so that log.Printf("Reading %s: %v", filename, err) formats without a spurious capital letter mid-message.
Available since 2019.1
Default: off. Enable by setting "analyses": {"ST1005": true}.
Package documentation: ST1005
Quoting Go Code Review Comments:
The name of a method's receiver should be a reflection of its identity; often a one or two letter abbreviation of its type suffices (such as "c" or "cl" for "Client"). Don't use generic names such as "me", "this" or "self", identifiers typical of object-oriented languages that place more emphasis on methods as opposed to functions. The name need not be as descriptive as that of a method argument, as its role is obvious and serves no documentary purpose. It can be very short as it will appear on almost every line of every method of the type; familiarity admits brevity. Be consistent, too: if you call the receiver "c" in one method, don't call it "cl" in another.
Available since 2019.1
Default: off. Enable by setting "analyses": {"ST1006": true}.
Package documentation: ST1006
A function's error value should be its last return value.
Available since 2019.1
Default: off. Enable by setting "analyses": {"ST1008": true}.
Package documentation: ST1008
time.Duration values represent an amount of time, which is represented as a count of nanoseconds. An expression like 5 * time.Microsecond yields the value 5000. It is therefore not appropriate to suffix a variable of type time.Duration with any time unit, such as Msec or Milli.
Available since 2019.1
Default: off. Enable by setting "analyses": {"ST1011": true}.
Package documentation: ST1011
Error variables that are part of an API should be called errFoo or ErrFoo.
Available since 2019.1
Default: off. Enable by setting "analyses": {"ST1012": true}.
Package documentation: ST1012
HTTP has a tremendous number of status codes. While some of those are well known (200, 400, 404, 500), most of them are not. The net/http package provides constants for all status codes that are part of the various specifications. It is recommended to use these constants instead of hard-coding magic numbers, to vastly improve the readability of your code.
Available since 2019.1
Options http_status_code_whitelist
Default: off. Enable by setting "analyses": {"ST1013": true}.
Package documentation: ST1013
Available since 2019.1
Default: off. Enable by setting "analyses": {"ST1015": true}.
Package documentation: ST1015
Available since 2019.1, non-default
Default: off. Enable by setting "analyses": {"ST1016": true}.
Package documentation: ST1016
Yoda conditions are conditions of the kind 'if 42 == x', where the literal is on the left side of the comparison. These are a common idiom in languages in which assignment is an expression, to avoid bugs of the kind 'if (x = 42)'. In Go, which doesn't allow for this kind of bug, we prefer the more idiomatic 'if x == 42'.
Available since 2019.2
Default: off. Enable by setting "analyses": {"ST1017": true}.
Package documentation: ST1017
Available since 2019.2
Default: off. Enable by setting "analyses": {"ST1018": true}.
Package documentation: ST1018
Go allows importing the same package multiple times, as long as different import aliases are being used. That is, the following bit of code is valid:
import (
"fmt"
fumpt "fmt"
format "fmt"
_ "fmt"
)
However, this is very rarely done on purpose. Usually, it is a sign of code that got refactored, accidentally adding duplicate import statements. It is also a rarely known feature, which may contribute to confusion.
Do note that sometimes, this feature may be used intentionally (see for example https://github.com/golang/go/commit/3409ce39bfd7584523b7a8c150a310cea92d879d) – if you want to allow this pattern in your code base, you're advised to disable this check.
Available since 2020.1
Default: off. Enable by setting "analyses": {"ST1019": true}.
Package documentation: ST1019
Doc comments work best as complete sentences, which allow a wide variety of automated presentations. The first sentence should be a one-sentence summary that starts with the name being declared.
If every doc comment begins with the name of the item it describes, you can use the doc subcommand of the go tool and run the output through grep.
See https://go.dev/doc/effective_go#commentary for more information on how to write good documentation.
Available since 2020.1, non-default
Default: off. Enable by setting "analyses": {"ST1020": true}.
Package documentation: ST1020
Doc comments work best as complete sentences, which allow a wide variety of automated presentations. The first sentence should be a one-sentence summary that starts with the name being declared.
If every doc comment begins with the name of the item it describes, you can use the doc subcommand of the go tool and run the output through grep.
See https://go.dev/doc/effective_go#commentary for more information on how to write good documentation.
Available since 2020.1, non-default
Default: off. Enable by setting "analyses": {"ST1021": true}.
Package documentation: ST1021
Doc comments work best as complete sentences, which allow a wide variety of automated presentations. The first sentence should be a one-sentence summary that starts with the name being declared.
If every doc comment begins with the name of the item it describes, you can use the doc subcommand of the go tool and run the output through grep.
See https://go.dev/doc/effective_go#commentary for more information on how to write good documentation.
Available since 2020.1, non-default
Default: off. Enable by setting "analyses": {"ST1022": true}.
Package documentation: ST1022
Available since 2021.1, non-default
Default: off. Enable by setting "analyses": {"ST1023": true}.
Package documentation: ST1023
This checker reports calls to append that pass no values to be appended to the slice.
s := []string{"a", "b", "c"}
_ = append(s)
Such calls are always no-ops and often indicate an underlying mistake.
Default: on.
Package documentation: appends
Default: on.
Package documentation: asmdecl
This checker reports assignments of the form x = x or a[i] = a[i]. These are almost always useless, and even when they aren't they are usually a mistake.
Default: on.
Package documentation: assign
The atomic checker looks for assignment statements of the form:
x = atomic.AddUint64(&x, 1)
which are not atomic.
Default: on.
Package documentation: atomic
Default: on.
Package documentation: atomicalign
Default: on.
Package documentation: bools
Default: on.
Package documentation: buildtag
Check for invalid cgo pointer passing. This looks for code that uses cgo to call C code passing values whose types are almost always invalid according to the cgo pointer sharing rules. Specifically, it warns about attempts to pass a Go chan, map, func, or slice to C, either directly, or via a pointer, array, or struct.
Default: on.
Package documentation: cgocall
This analyzer reports a diagnostic for composite literals of struct types imported from another package that do not use the field-keyed syntax. Such literals are fragile because the addition of a new field (even if unexported) to the struct will cause compilation to fail.
As an example,
err = &net.DNSConfigError{err}
should be replaced by:
err = &net.DNSConfigError{Err: err}
Default: on.
Package documentation: composites
Inadvertently copying a value containing a lock, such as sync.Mutex or sync.WaitGroup, may cause both copies to malfunction. Generally such values should be referred to through a pointer.
Default: on.
Package documentation: copylocks
The deepequalerrors checker looks for calls of the form:
reflect.DeepEqual(err1, err2)
where err1 and err2 are errors. Using reflect.DeepEqual to compare errors is discouraged.
Default: on.
Package documentation: deepequalerrors
The defers analyzer reports a diagnostic when a defer statement would result in a non-deferred call to time.Since, as experience has shown that this is nearly always a mistake.
For example:
start := time.Now()
...
defer recordLatency(time.Since(start)) // error: call to time.Since is not deferred
The correct code is:
defer func() { recordLatency(time.Since(start)) }()
Default: on.
Package documentation: defers
The deprecated analyzer looks for deprecated symbols and package imports.
See https://go.dev/wiki/Deprecated to learn about Go's convention for documenting and signaling deprecated identifiers.
Default: on.
Package documentation: deprecated
This analyzer checks for problems with known Go toolchain directives in all Go source files in a package directory, even those excluded by //go:build constraints, and all non-Go source files too.
For //go:debug (see https://go.dev/doc/godebug), the analyzer checks that the directives are placed only in Go source files, only above the package comment, and only in package main or *_test.go files.
Support for other known directives may be added in the future.
This analyzer does not check //go:build, which is handled by the buildtag analyzer.
Default: on.
Package documentation: directive
This analyzer checks that the embed package is imported if //go:embed directives are present, providing a suggested fix to add the import if it is missing.
This analyzer also checks that //go:embed directives precede the declaration of a single variable.
Default: on.
Package documentation: embed
The errorsas analysis reports calls to errors.As where the type of the second argument is not a pointer to a type implementing error.
Default: on.
Package documentation: errorsas
This checker provides suggested fixes for type errors of the type "wrong number of return values (want %d, got %d)". For example:
func m() (int, string, *bool, error) {
return
}
will turn into
func m() (int, string, *bool, error) {
return 0, "", nil, nil
}
This functionality is similar to https://github.com/sqs/goreturns.
Default: on.
Package documentation: fillreturns
Default: on.
Package documentation: framepointer
The gofix analyzer inlines functions and constants that are marked for inlining.
Default: on.
Package documentation: gofix
This analyzer flags code that produce network address strings using fmt.Sprintf, as in this example:
addr := fmt.Sprintf("%s:%d", host, 12345) // "will not work with IPv6"
...
conn, err := net.Dial("tcp", addr) // "when passed to dial here"
The analyzer suggests a fix to use the correct approach, a call to net.JoinHostPort:
addr := net.JoinHostPort(host, "12345")
...
conn, err := net.Dial("tcp", addr)
A similar diagnostic and fix are produced for a format string of "%s:%s".
Default: on.
Package documentation: hostport
A common mistake when using the net/http package is to defer a function call to close the http.Response Body before checking the error that determines whether the response is valid:
resp, err := http.Head(url)
defer resp.Body.Close()
if err != nil {
log.Fatal(err)
}
// (defer statement belongs here)
This checker helps uncover latent nil dereference bugs by reporting a diagnostic for such mistakes.
Default: on.
Package documentation: httpresponse
This checker flags type assertions v.(T) and corresponding type-switch cases in which the static type V of v is an interface that cannot possibly implement the target interface T. This occurs when V and T contain methods with the same name but different signatures. Example:
var v interface {
Read()
}
_ = v.(io.Reader)
The Read method in v has a different signature than the Read method in io.Reader, so this assertion cannot succeed.
Default: on.
Package documentation: ifaceassert
Explicit type arguments may be omitted from call expressions if they can be inferred from function arguments, or from other type arguments:
func f[T any](T) {}
func _() {
f[string]("foo") // string could be inferred
}
Default: on.
Package documentation: infertypeargs
This analyzer reports places where a function literal references the iteration variable of an enclosing loop, and the loop calls the function in such a way (e.g. with go or defer) that it may outlive the loop iteration and possibly observe the wrong value of the variable.
Note: An iteration variable can only outlive a loop iteration in Go versions <=1.21. In Go 1.22 and later, the loop variable lifetimes changed to create a new iteration variable per loop iteration. (See go.dev/issue/60078.)
In this example, all the deferred functions run after the loop has completed, so all observe the final value of v [<go1.22].
for _, v := range list {
defer func() {
use(v) // incorrect
}()
}
One fix is to create a new variable for each iteration of the loop:
for _, v := range list {
v := v // new var per iteration
defer func() {
use(v) // ok
}()
}
After Go version 1.22, the previous two for loops are equivalent and both are correct.
The next example uses a go statement and has a similar problem [<go1.22]. In addition, it has a data race because the loop updates v concurrent with the goroutines accessing it.
for _, v := range elem {
go func() {
use(v) // incorrect, and a data race
}()
}
A fix is the same as before. The checker also reports problems in goroutines started by golang.org/x/sync/errgroup.Group. A hard-to-spot variant of this form is common in parallel tests:
func Test(t *testing.T) {
for _, test := range tests {
t.Run(test.name, func(t *testing.T) {
t.Parallel()
use(test) // incorrect, and a data race
})
}
}
The t.Parallel() call causes the rest of the function to execute concurrent with the loop [<go1.22].
The analyzer reports references only in the last statement, as it is not deep enough to understand the effects of subsequent statements that might render the reference benign. ("Last statement" is defined recursively in compound statements such as if, switch, and select.)
See: https://golang.org/doc/go_faq.html#closures_and_goroutines
Default: on.
Package documentation: loopclosure
The cancellation function returned by context.WithCancel, WithTimeout, WithDeadline and variants such as WithCancelCause must be called, or the new context will remain live until its parent context is cancelled. (The background context is never cancelled.)
Default: on.
Package documentation: lostcancel
This analyzer reports opportunities for simplifying and clarifying existing code by using more modern features of Go and its standard library.
Each diagnostic provides a fix. Our intent is that these fixes may be safely applied en masse without changing the behavior of your program. In some cases the suggested fixes are imperfect and may lead to (for example) unused imports or unused local variables, causing build breakage. However, these problems are generally trivial to fix. We regard any modernizer whose fix changes program behavior to have a serious bug and will endeavor to fix it.
To apply all modernization fixes en masse, you can use the following command:
$ go run golang.org/x/tools/gopls/internal/analysis/modernize/cmd/modernize@latest -fix -test ./...
(Do not use "go get -tool" to add gopls as a dependency of your module; gopls commands must be built from their release branch.)
If the tool warns of conflicting fixes, you may need to run it more than once until it has applied all fixes cleanly. This command is not an officially supported interface and may change in the future.
Changes produced by this tool should be reviewed as usual before being merged. In some cases, a loop may be replaced by a simple function call, causing comments within the loop to be discarded. Human judgment may be required to avoid losing comments of value.
Each diagnostic reported by modernize has a specific category. (The categories are listed below.) Diagnostics in some categories, such as "efaceany" (which replaces "interface{}" with "any" where it is safe to do so) are particularly numerous. It may ease the burden of code review to apply fixes in two passes, the first change consisting only of fixes of category "efaceany", the second consisting of all others. This can be achieved using the -category flag:
$ modernize -category=efaceany -fix -test ./...
$ modernize -category=-efaceany -fix -test ./...
Categories of modernize diagnostic:
-
forvar: remove x := x variable declarations made unnecessary by the new semantics of loops in go1.22.
-
slicescontains: replace 'for i, elem := range s { if elem == needle { ...; break }' by a call to slices.Contains, added in go1.21.
-
minmax: replace an if/else conditional assignment by a call to the built-in min or max functions added in go1.21.
-
sortslice: replace sort.Slice(x, func(i, j int) bool) { return s[i] < s[j] } by a call to slices.Sort(s), added in go1.21.
-
efaceany: replace interface{} by the 'any' type added in go1.18.
-
slicesclone: replace append([]T(nil), s...) by slices.Clone(s) or slices.Concat(s), added in go1.21.
-
mapsloop: replace a loop around an m[k]=v map update by a call to one of the Collect, Copy, Clone, or Insert functions from the maps package, added in go1.21.
-
fmtappendf: replace []byte(fmt.Sprintf...) by fmt.Appendf(nil, ...), added in go1.19.
-
testingcontext: replace uses of context.WithCancel in tests with t.Context, added in go1.24.
-
omitzero: replace omitempty by omitzero on structs, added in go1.24.
-
bloop: replace "for i := range b.N" or "for range b.N" in a benchmark with "for b.Loop()", and remove any preceding calls to b.StopTimer, b.StartTimer, and b.ResetTimer.
-
slicesdelete: replace append(s[:i], s[i+1]...) by slices.Delete(s, i, i+1), added in go1.21.
-
rangeint: replace a 3-clause "for i := 0; i < n; i++" loop by "for i := range n", added in go1.22.
-
stringsseq: replace Split in "for range strings.Split(...)" by go1.24's more efficient SplitSeq, or Fields with FieldSeq.
-
stringscutprefix: replace some uses of HasPrefix followed by TrimPrefix with CutPrefix, added to the strings package in go1.20.
-
waitgroup: replace old complex usages of sync.WaitGroup by less complex WaitGroup.Go method in go1.25.
Default: on.
Package documentation: modernize
A useless comparison is one like f == nil as opposed to f() == nil.
Default: on.
Package documentation: nilfunc
The nilness checker inspects the control-flow graph of each function in a package and reports nil pointer dereferences, degenerate nil pointers, and panics with nil values. A degenerate comparison is of the form x==nil or x!=nil where x is statically known to be nil or non-nil. These are often a mistake, especially in control flow related to errors. Panics with nil values are checked because they are not detectable by
if r := recover(); r != nil {
This check reports conditions such as:
if f == nil { // impossible condition (f is a function)
}
and:
p := &v
...
if p != nil { // tautological condition
}
and:
if p == nil {
print(*p) // nil dereference
}
and:
if p == nil {
panic(p)
}
Sometimes the control flow may be quite complex, making bugs hard to spot. In the example below, the err.Error expression is guaranteed to panic because, after the first return, err must be nil. The intervening loop is just a distraction.
...
err := g.Wait()
if err != nil {
return err
}
partialSuccess := false
for _, err := range errs {
if err == nil {
partialSuccess = true
break
}
}
if partialSuccess {
reportStatus(StatusMessage{
Code: code.ERROR,
Detail: err.Error(), // "nil dereference in dynamic method call"
})
return nil
}
...
Default: on.
Package documentation: nilness
This checker provides suggested fixes for type errors of the type "no new vars on left side of :=". For example:
z := 1
z := 2
will turn into
z := 1
z = 2
Default: on.
Package documentation: nonewvars
This checker provides suggested fixes for type errors of the type "no result values expected" or "too many return values". For example:
func z() { return nil }
will turn into
func z() { return }
Default: on.
Package documentation: noresultvalues
The check applies to calls of the formatting functions such as [fmt.Printf] and [fmt.Sprintf], as well as any detected wrappers of those functions such as [log.Printf]. It reports a variety of mistakes such as syntax errors in the format string and mismatches (of number and type) between the verbs and their arguments.
See the documentation of the fmt package for the complete set of format operators and their operand types.
Default: on.
Package documentation: printf
This analyzer check for shadowed variables. A shadowed variable is a variable declared in an inner scope with the same name and type as a variable in an outer scope, and where the outer variable is mentioned after the inner one is declared.
(This definition can be refined; the module generates too many false positives and is not yet enabled by default.)
For example:
func BadRead(f *os.File, buf []byte) error {
var err error
for {
n, err := f.Read(buf) // shadows the function variable 'err'
if err != nil {
break // causes return of wrong value
}
foo(buf)
}
return err
}
Default: off. Enable by setting "analyses": {"shadow": true}.
Package documentation: shadow
Default: on.
Package documentation: shift
This checker reports call expression of the form
signal.Notify(c <-chan os.Signal, sig ...os.Signal),
where c is an unbuffered channel, which can be at risk of missing the signal.
Default: on.
Package documentation: sigchanyzer
An array, slice, or map composite literal of the form:
[]T{T{}, T{}}
will be simplified to:
[]T{{}, {}}
This is one of the simplifications that "gofmt -s" applies.
This analyzer ignores generated code.
Default: on.
Package documentation: simplifycompositelit
A range of the form:
for x, _ = range v {...}
will be simplified to:
for x = range v {...}
A range of the form:
for _ = range v {...}
will be simplified to:
for range v {...}
This is one of the simplifications that "gofmt -s" applies.
This analyzer ignores generated code.
Default: on.
Package documentation: simplifyrange
A slice expression of the form:
s[a:len(s)]
will be simplified to:
s[a:]
This is one of the simplifications that "gofmt -s" applies.
This analyzer ignores generated code.
Default: on.
Package documentation: simplifyslice
The slog checker looks for calls to functions from the log/slog package that take alternating key-value pairs. It reports calls where an argument in a key position is neither a string nor a slog.Attr, and where a final key is missing its value. For example,it would report
slog.Warn("message", 11, "k") // slog.Warn arg "11" should be a string or a slog.Attr
and
slog.Info("message", "k1", v1, "k2") // call to slog.Info missing a final value
Default: on.
Package documentation: slog
sort.Slice requires an argument of a slice type. Check that the interface{} value passed to sort.Slice is actually a slice.
Default: on.
Package documentation: sortslice
Sometimes a type may be intended to satisfy an interface but may fail to do so because of a mistake in its method signature. For example, the result of this WriteTo method should be (int64, error), not error, to satisfy io.WriterTo:
type myWriterTo struct{...}
func (myWriterTo) WriteTo(w io.Writer) error { ... }
This check ensures that each method whose name matches one of several well-known interface methods from the standard library has the correct signature for that interface.
Checked method names include:
Format GobEncode GobDecode MarshalJSON MarshalXML
Peek ReadByte ReadFrom ReadRune Scan Seek
UnmarshalJSON UnreadByte UnreadRune WriteByte
WriteTo
Default: on.
Package documentation: stdmethods
The stdversion analyzer reports references to symbols in the standard library that were introduced by a Go release higher than the one in force in the referring file. (Recall that the file's Go version is defined by the 'go' directive its module's go.mod file, or by a "//go:build go1.X" build tag at the top of the file.)
The analyzer does not report a diagnostic for a reference to a "too new" field or method of a type that is itself "too new", as this may have false positives, for example if fields or methods are accessed through a type alias that is guarded by a Go version constraint.
Default: on.
Package documentation: stdversion
This checker flags conversions of the form string(x) where x is an integer (but not byte or rune) type. Such conversions are discouraged because they return the UTF-8 representation of the Unicode code point x, and not a decimal string representation of x as one might expect. Furthermore, if x denotes an invalid code point, the conversion cannot be statically rejected.
For conversions that intend on using the code point, consider replacing them with string(rune(x)). Otherwise, strconv.Itoa and its equivalents return the string representation of the value in the desired base.
Default: on.
Package documentation: stringintconv
Also report certain struct tags (json, xml) used with unexported fields.
Default: on.
Package documentation: structtag
Functions that abruptly terminate a test, such as the Fatal, Fatalf, FailNow, and Skip{,f,Now} methods of *testing.T, must be called from the test goroutine itself. This checker detects calls to these functions that occur within a goroutine started by the test. For example:
func TestFoo(t *testing.T) {
go func() {
t.Fatal("oops") // error: (*T).Fatal called from non-test goroutine
}()
}
Default: on.
Package documentation: testinggoroutine
The tests checker walks Test, Benchmark, Fuzzing and Example functions checking malformed names, wrong signatures and examples documenting non-existent identifiers.
Please see the documentation for package testing in golang.org/pkg/testing for the conventions that are enforced for Tests, Benchmarks, and Examples.
Default: on.
Package documentation: tests
The timeformat checker looks for time formats with the 2006-02-01 (yyyy-dd-mm) format. Internationally, "yyyy-dd-mm" does not occur in common calendar date standards, and so it is more likely that 2006-01-02 (yyyy-mm-dd) was intended.
Default: on.
Package documentation: timeformat
The unmarshal analysis reports calls to functions such as json.Unmarshal in which the argument type is not a pointer or an interface.
Default: on.
Package documentation: unmarshal
The unreachable analyzer finds statements that execution can never reach because they are preceded by a return statement, a call to panic, an infinite loop, or similar constructs.
Default: on.
Package documentation: unreachable
The unsafeptr analyzer reports likely incorrect uses of unsafe.Pointer to convert integers to pointers. A conversion from uintptr to unsafe.Pointer is invalid if it implies that there is a uintptr-typed word in memory that holds a pointer value, because that word will be invisible to stack copying and to the garbage collector.
Default: on.
Package documentation: unsafeptr
The unusedfunc analyzer reports functions and methods that are never referenced outside of their own declaration.
A function is considered unused if it is unexported and not referenced (except within its own declaration).
A method is considered unused if it is unexported, not referenced (except within its own declaration), and its name does not match that of any method of an interface type declared within the same package.
The tool may report false positives in some situations, for example:
-
For a declaration of an unexported function that is referenced from another package using the go:linkname mechanism, if the declaration's doc comment does not also have a go:linkname comment.
(Such code is in any case strongly discouraged: linkname annotations, if they must be used at all, should be used on both the declaration and the alias.)
-
For compiler intrinsics in the "runtime" package that, though never referenced, are known to the compiler and are called indirectly by compiled object code.
-
For functions called only from assembly.
-
For functions called only from files whose build tags are not selected in the current build configuration.
See golang/go#71686 for discussion of these limitations.
The unusedfunc algorithm is not as precise as the golang.org/x/tools/cmd/deadcode tool, but it has the advantage that it runs within the modular analysis framework, enabling near real-time feedback within gopls.
Default: on.
Package documentation: unusedfunc
The unusedparams analyzer checks functions to see if there are any parameters that are not being used.
To ensure soundness, it ignores:
- "address-taken" functions, that is, functions that are used as a value rather than being called directly; their signatures may be required to conform to a func type.
- exported functions or methods, since they may be address-taken in another package.
- unexported methods whose name matches an interface method declared in the same package, since the method's signature may be required to conform to the interface type.
- functions with empty bodies, or containing just a call to panic.
- parameters that are unnamed, or named "_", the blank identifier.
The analyzer suggests a fix of replacing the parameter name by "_", but in such cases a deeper fix can be obtained by invoking the "Refactor: remove unused parameter" code action, which will eliminate the parameter entirely, along with all corresponding arguments at call sites, while taking care to preserve any side effects in the argument expressions; see https://github.com/golang/tools/releases/tag/gopls%2Fv0.14.
This analyzer ignores generated code.
Default: on.
Package documentation: unusedparams
Some functions like fmt.Errorf return a result and have no side effects, so it is always a mistake to discard the result. Other functions may return an error that must not be ignored, or a cleanup operation that must be called. This analyzer reports calls to functions like these when the result of the call is ignored.
The set of functions may be controlled using flags.
Default: on.
Package documentation: unusedresult
Default: on.
Package documentation: unusedvariable
The analyzer reports instances of writes to struct fields and arrays that are never read. Specifically, when a struct object or an array is copied, its elements are copied implicitly by the compiler, and any element write to this copy does nothing with the original object.
For example:
type T struct { x int }
func f(input []T) {
for i, v := range input { // v is a copy
v.x = i // unused write to field x
}
}
Another example is about non-pointer receiver:
type T struct { x int }
func (t T) f() { // t is a copy
t.x = i // unused write to field x
}
Default: on.
Package documentation: unusedwrite
This analyzer detects mistaken calls to the (*sync.WaitGroup).Add method from inside a new goroutine, causing Add to race with Wait:
// WRONG
var wg sync.WaitGroup
go func() {
wg.Add(1) // "WaitGroup.Add called from inside new goroutine"
defer wg.Done()
...
}()
wg.Wait() // (may return prematurely before new goroutine starts)
The correct code calls Add before starting the goroutine:
// RIGHT
var wg sync.WaitGroup
wg.Add(1)
go func() {
defer wg.Done()
...
}()
wg.Wait()
Default: on.
Package documentation: waitgroup
After a yield function returns false, the caller should not call the yield function again; generally the iterator should return promptly.
This example fails to check the result of the call to yield, causing this analyzer to report a diagnostic:
yield(1) // yield may be called again (on L2) after returning false
yield(2)
The corrected code is either this:
if yield(1) { yield(2) }
or simply:
_ = yield(1) && yield(2)
It is not always a mistake to ignore the result of yield. For example, this is a valid single-element iterator:
yield(1) // ok to ignore result
return
It is only a mistake when the yield call that returned false may be followed by another call.
Default: on.
Package documentation: yield