open-nomad/vendor/honnef.co/go/tools/lint/lint.go

540 lines
12 KiB
Go
Raw Normal View History

// Package lint provides the foundation for tools like staticcheck
package lint // import "honnef.co/go/tools/lint"
import (
"bytes"
"encoding/gob"
"fmt"
"go/scanner"
"go/token"
"go/types"
"path/filepath"
"sort"
"strings"
"sync"
"sync/atomic"
"unicode"
"golang.org/x/tools/go/analysis"
"golang.org/x/tools/go/packages"
"honnef.co/go/tools/config"
"honnef.co/go/tools/internal/cache"
)
type Documentation struct {
Title string
Text string
Since string
NonDefault bool
Options []string
}
func (doc *Documentation) String() string {
b := &strings.Builder{}
fmt.Fprintf(b, "%s\n\n", doc.Title)
if doc.Text != "" {
fmt.Fprintf(b, "%s\n\n", doc.Text)
}
fmt.Fprint(b, "Available since\n ")
if doc.Since == "" {
fmt.Fprint(b, "unreleased")
} else {
fmt.Fprintf(b, "%s", doc.Since)
}
if doc.NonDefault {
fmt.Fprint(b, ", non-default")
}
fmt.Fprint(b, "\n")
if len(doc.Options) > 0 {
fmt.Fprintf(b, "\nOptions\n")
for _, opt := range doc.Options {
fmt.Fprintf(b, " %s", opt)
}
fmt.Fprint(b, "\n")
}
return b.String()
}
type Ignore interface {
Match(p Problem) bool
}
type LineIgnore struct {
File string
Line int
Checks []string
Matched bool
Pos token.Position
}
func (li *LineIgnore) Match(p Problem) bool {
pos := p.Pos
if pos.Filename != li.File || pos.Line != li.Line {
return false
}
for _, c := range li.Checks {
if m, _ := filepath.Match(c, p.Check); m {
li.Matched = true
return true
}
}
return false
}
func (li *LineIgnore) String() string {
matched := "not matched"
if li.Matched {
matched = "matched"
}
return fmt.Sprintf("%s:%d %s (%s)", li.File, li.Line, strings.Join(li.Checks, ", "), matched)
}
type FileIgnore struct {
File string
Checks []string
}
func (fi *FileIgnore) Match(p Problem) bool {
if p.Pos.Filename != fi.File {
return false
}
for _, c := range fi.Checks {
if m, _ := filepath.Match(c, p.Check); m {
return true
}
}
return false
}
type Severity uint8
const (
Error Severity = iota
Warning
Ignored
)
// Problem represents a problem in some source code.
type Problem struct {
Pos token.Position
End token.Position
Message string
Check string
Severity Severity
Related []Related
}
type Related struct {
Pos token.Position
End token.Position
Message string
}
func (p Problem) Equal(o Problem) bool {
return p.Pos == o.Pos &&
p.End == o.End &&
p.Message == o.Message &&
p.Check == o.Check &&
p.Severity == o.Severity
}
func (p *Problem) String() string {
return fmt.Sprintf("%s (%s)", p.Message, p.Check)
}
// A Linter lints Go source code.
type Linter struct {
Checkers []*analysis.Analyzer
CumulativeCheckers []CumulativeChecker
GoVersion int
Config config.Config
Stats Stats
RepeatAnalyzers uint
}
type CumulativeChecker interface {
Analyzer() *analysis.Analyzer
Result() []types.Object
ProblemObject(*token.FileSet, types.Object) Problem
}
func (l *Linter) Lint(cfg *packages.Config, patterns []string) ([]Problem, error) {
var allAnalyzers []*analysis.Analyzer
allAnalyzers = append(allAnalyzers, l.Checkers...)
for _, cum := range l.CumulativeCheckers {
allAnalyzers = append(allAnalyzers, cum.Analyzer())
}
// The -checks command line flag overrules all configuration
// files, which means that for `-checks="foo"`, no check other
// than foo can ever be reported to the user. Make use of this
// fact to cull the list of analyses we need to run.
// replace "inherit" with "all", as we don't want to base the
// list of all checks on the default configuration, which
// disables certain checks.
checks := make([]string, len(l.Config.Checks))
copy(checks, l.Config.Checks)
for i, c := range checks {
if c == "inherit" {
checks[i] = "all"
}
}
allowed := FilterChecks(allAnalyzers, checks)
var allowedAnalyzers []*analysis.Analyzer
for _, c := range l.Checkers {
if allowed[c.Name] {
allowedAnalyzers = append(allowedAnalyzers, c)
}
}
hasCumulative := false
for _, cum := range l.CumulativeCheckers {
a := cum.Analyzer()
if allowed[a.Name] {
hasCumulative = true
allowedAnalyzers = append(allowedAnalyzers, a)
}
}
r, err := NewRunner(&l.Stats)
if err != nil {
return nil, err
}
r.goVersion = l.GoVersion
r.repeatAnalyzers = l.RepeatAnalyzers
pkgs, err := r.Run(cfg, patterns, allowedAnalyzers, hasCumulative)
if err != nil {
return nil, err
}
tpkgToPkg := map[*types.Package]*Package{}
for _, pkg := range pkgs {
tpkgToPkg[pkg.Types] = pkg
for _, e := range pkg.errs {
switch e := e.(type) {
case types.Error:
p := Problem{
Pos: e.Fset.PositionFor(e.Pos, false),
Message: e.Msg,
Severity: Error,
Check: "compile",
}
pkg.problems = append(pkg.problems, p)
case packages.Error:
msg := e.Msg
if len(msg) != 0 && msg[0] == '\n' {
// TODO(dh): See https://github.com/golang/go/issues/32363
msg = msg[1:]
}
var pos token.Position
if e.Pos == "" {
// Under certain conditions (malformed package
// declarations, multiple packages in the same
// directory), go list emits an error on stderr
// instead of JSON. Those errors do not have
// associated position information in
// go/packages.Error, even though the output on
// stderr may contain it.
if p, n, err := parsePos(msg); err == nil {
if abs, err := filepath.Abs(p.Filename); err == nil {
p.Filename = abs
}
pos = p
msg = msg[n+2:]
}
} else {
var err error
pos, _, err = parsePos(e.Pos)
if err != nil {
panic(fmt.Sprintf("internal error: %s", e))
}
}
p := Problem{
Pos: pos,
Message: msg,
Severity: Error,
Check: "compile",
}
pkg.problems = append(pkg.problems, p)
case scanner.ErrorList:
for _, e := range e {
p := Problem{
Pos: e.Pos,
Message: e.Msg,
Severity: Error,
Check: "compile",
}
pkg.problems = append(pkg.problems, p)
}
case error:
p := Problem{
Pos: token.Position{},
Message: e.Error(),
Severity: Error,
Check: "compile",
}
pkg.problems = append(pkg.problems, p)
}
}
}
atomic.StoreUint32(&r.stats.State, StateCumulative)
for _, cum := range l.CumulativeCheckers {
for _, res := range cum.Result() {
pkg := tpkgToPkg[res.Pkg()]
if pkg == nil {
panic(fmt.Sprintf("analyzer %s flagged object %s in package %s, a package that we aren't tracking", cum.Analyzer(), res, res.Pkg()))
}
allowedChecks := FilterChecks(allowedAnalyzers, pkg.cfg.Merge(l.Config).Checks)
if allowedChecks[cum.Analyzer().Name] {
pos := DisplayPosition(pkg.Fset, res.Pos())
// FIXME(dh): why are we ignoring generated files
// here? Surely this is specific to 'unused', not all
// cumulative checkers
if _, ok := pkg.gen[pos.Filename]; ok {
continue
}
p := cum.ProblemObject(pkg.Fset, res)
pkg.problems = append(pkg.problems, p)
}
}
}
for _, pkg := range pkgs {
if !pkg.fromSource {
// Don't cache packages that we loaded from the cache
continue
}
cpkg := cachedPackage{
Problems: pkg.problems,
Ignores: pkg.ignores,
Config: pkg.cfg,
}
buf := &bytes.Buffer{}
if err := gob.NewEncoder(buf).Encode(cpkg); err != nil {
return nil, err
}
id := cache.Subkey(pkg.actionID, "data "+r.problemsCacheKey)
if err := r.cache.PutBytes(id, buf.Bytes()); err != nil {
return nil, err
}
}
var problems []Problem
// Deduplicate line ignores. When U1000 processes a package and
// its test variant, it will only emit a single problem for an
// unused object, not two problems. We will, however, have two
// line ignores, one per package. Without deduplication, one line
// ignore will be marked as matched, while the other one won't,
// subsequently reporting a "this linter directive didn't match
// anything" error.
ignores := map[token.Position]Ignore{}
for _, pkg := range pkgs {
for _, ig := range pkg.ignores {
if lig, ok := ig.(*LineIgnore); ok {
ig = ignores[lig.Pos]
if ig == nil {
ignores[lig.Pos] = lig
ig = lig
}
}
for i := range pkg.problems {
p := &pkg.problems[i]
if ig.Match(*p) {
p.Severity = Ignored
}
}
}
if pkg.cfg == nil {
// The package failed to load, otherwise we would have a
// valid config. Pass through all errors.
problems = append(problems, pkg.problems...)
} else {
for _, p := range pkg.problems {
allowedChecks := FilterChecks(allowedAnalyzers, pkg.cfg.Merge(l.Config).Checks)
allowedChecks["compile"] = true
if allowedChecks[p.Check] {
problems = append(problems, p)
}
}
}
for _, ig := range pkg.ignores {
ig, ok := ig.(*LineIgnore)
if !ok {
continue
}
ig = ignores[ig.Pos].(*LineIgnore)
if ig.Matched {
continue
}
couldveMatched := false
allowedChecks := FilterChecks(allowedAnalyzers, pkg.cfg.Merge(l.Config).Checks)
for _, c := range ig.Checks {
if !allowedChecks[c] {
continue
}
couldveMatched = true
break
}
if !couldveMatched {
// The ignored checks were disabled for the containing package.
// Don't flag the ignore for not having matched.
continue
}
p := Problem{
Pos: ig.Pos,
Message: "this linter directive didn't match anything; should it be removed?",
Check: "",
}
problems = append(problems, p)
}
}
if len(problems) == 0 {
return nil, nil
}
sort.Slice(problems, func(i, j int) bool {
pi := problems[i].Pos
pj := problems[j].Pos
if pi.Filename != pj.Filename {
return pi.Filename < pj.Filename
}
if pi.Line != pj.Line {
return pi.Line < pj.Line
}
if pi.Column != pj.Column {
return pi.Column < pj.Column
}
return problems[i].Message < problems[j].Message
})
var out []Problem
out = append(out, problems[0])
for i, p := range problems[1:] {
// We may encounter duplicate problems because one file
// can be part of many packages.
if !problems[i].Equal(p) {
out = append(out, p)
}
}
return out, nil
}
func FilterChecks(allChecks []*analysis.Analyzer, checks []string) map[string]bool {
// OPT(dh): this entire computation could be cached per package
allowedChecks := map[string]bool{}
for _, check := range checks {
b := true
if len(check) > 1 && check[0] == '-' {
b = false
check = check[1:]
}
if check == "*" || check == "all" {
// Match all
for _, c := range allChecks {
allowedChecks[c.Name] = b
}
} else if strings.HasSuffix(check, "*") {
// Glob
prefix := check[:len(check)-1]
isCat := strings.IndexFunc(prefix, func(r rune) bool { return unicode.IsNumber(r) }) == -1
for _, c := range allChecks {
idx := strings.IndexFunc(c.Name, func(r rune) bool { return unicode.IsNumber(r) })
if isCat {
// Glob is S*, which should match S1000 but not SA1000
cat := c.Name[:idx]
if prefix == cat {
allowedChecks[c.Name] = b
}
} else {
// Glob is S1*
if strings.HasPrefix(c.Name, prefix) {
allowedChecks[c.Name] = b
}
}
}
} else {
// Literal check name
allowedChecks[check] = b
}
}
return allowedChecks
}
func DisplayPosition(fset *token.FileSet, p token.Pos) token.Position {
if p == token.NoPos {
return token.Position{}
}
// Only use the adjusted position if it points to another Go file.
// This means we'll point to the original file for cgo files, but
// we won't point to a YACC grammar file.
pos := fset.PositionFor(p, false)
adjPos := fset.PositionFor(p, true)
if filepath.Ext(adjPos.Filename) == ".go" {
return adjPos
}
return pos
}
var bufferPool = &sync.Pool{
New: func() interface{} {
buf := bytes.NewBuffer(nil)
buf.Grow(64)
return buf
},
}
func FuncName(f *types.Func) string {
buf := bufferPool.Get().(*bytes.Buffer)
buf.Reset()
if f.Type() != nil {
sig := f.Type().(*types.Signature)
if recv := sig.Recv(); recv != nil {
buf.WriteByte('(')
if _, ok := recv.Type().(*types.Interface); ok {
// gcimporter creates abstract methods of
// named interfaces using the interface type
// (not the named type) as the receiver.
// Don't print it in full.
buf.WriteString("interface")
} else {
types.WriteType(buf, recv.Type(), nil)
}
buf.WriteByte(')')
buf.WriteByte('.')
} else if f.Pkg() != nil {
writePackage(buf, f.Pkg())
}
}
buf.WriteString(f.Name())
s := buf.String()
bufferPool.Put(buf)
return s
}
func writePackage(buf *bytes.Buffer, pkg *types.Package) {
if pkg == nil {
return
}
s := pkg.Path()
if s != "" {
buf.WriteString(s)
buf.WriteByte('.')
}
}