blob: 8361515519fe05831a17685392d25ea99ebe1fed [file] [log] [blame]
// Copyright 2018 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
// Package gopathwalk is like filepath.Walk but specialized for finding Go
// packages, particularly in $GOPATH and $GOROOT.
package gopathwalk
import (
"bufio"
"bytes"
"io"
"io/fs"
"os"
"path/filepath"
"runtime"
"strings"
"sync"
"time"
)
// Options controls the behavior of a Walk call.
type Options struct {
// If Logf is non-nil, debug logging is enabled through this function.
Logf func(format string, args ...interface{})
// Search module caches. Also disables legacy goimports ignore rules.
ModulesEnabled bool
// Maximum number of concurrent calls to user-provided callbacks,
// or 0 for GOMAXPROCS.
Concurrency int
}
// RootType indicates the type of a Root.
type RootType int
const (
RootUnknown RootType = iota
RootGOROOT
RootGOPATH
RootCurrentModule
RootModuleCache
RootOther
)
// A Root is a starting point for a Walk.
type Root struct {
Path string
Type RootType
}
// Walk concurrently walks Go source directories ($GOROOT, $GOPATH, etc) to find packages.
//
// For each package found, add will be called with the absolute
// paths of the containing source directory and the package directory.
//
// Unlike filepath.WalkDir, Walk follows symbolic links
// (while guarding against cycles).
func Walk(roots []Root, add func(root Root, dir string), opts Options) {
WalkSkip(roots, add, func(Root, string) bool { return false }, opts)
}
// WalkSkip concurrently walks Go source directories ($GOROOT, $GOPATH, etc) to
// find packages.
//
// For each package found, add will be called with the absolute
// paths of the containing source directory and the package directory.
// For each directory that will be scanned, skip will be called
// with the absolute paths of the containing source directory and the directory.
// If skip returns false on a directory it will be processed.
//
// Unlike filepath.WalkDir, WalkSkip follows symbolic links
// (while guarding against cycles).
func WalkSkip(roots []Root, add func(root Root, dir string), skip func(root Root, dir string) bool, opts Options) {
for _, root := range roots {
walkDir(root, add, skip, opts)
}
}
// walkDir creates a walker and starts fastwalk with this walker.
func walkDir(root Root, add func(Root, string), skip func(root Root, dir string) bool, opts Options) {
if opts.Logf == nil {
opts.Logf = func(format string, args ...interface{}) {}
}
if _, err := os.Stat(root.Path); os.IsNotExist(err) {
opts.Logf("skipping nonexistent directory: %v", root.Path)
return
}
start := time.Now()
opts.Logf("scanning %s", root.Path)
concurrency := opts.Concurrency
if concurrency == 0 {
// The walk be either CPU-bound or I/O-bound, depending on what the
// caller-supplied add function does and the details of the user's platform
// and machine. Rather than trying to fine-tune the concurrency level for a
// specific environment, we default to GOMAXPROCS: it is likely to be a good
// choice for a CPU-bound add function, and if it is instead I/O-bound, then
// dealing with I/O saturation is arguably the job of the kernel and/or
// runtime. (Oversaturating I/O seems unlikely to harm performance as badly
// as failing to saturate would.)
concurrency = runtime.GOMAXPROCS(0)
}
w := &walker{
root: root,
add: add,
skip: skip,
opts: opts,
sem: make(chan struct{}, concurrency),
}
w.init()
w.sem <- struct{}{}
path := root.Path
if path == "" {
path = "."
}
if fi, err := os.Lstat(path); err == nil {
w.walk(path, nil, fs.FileInfoToDirEntry(fi))
} else {
w.opts.Logf("scanning directory %v: %v", root.Path, err)
}
<-w.sem
w.walking.Wait()
opts.Logf("scanned %s in %v", root.Path, time.Since(start))
}
// walker is the callback for fastwalk.Walk.
type walker struct {
root Root // The source directory to scan.
add func(Root, string) // The callback that will be invoked for every possible Go package dir.
skip func(Root, string) bool // The callback that will be invoked for every dir. dir is skipped if it returns true.
opts Options // Options passed to Walk by the user.
walking sync.WaitGroup
sem chan struct{} // Channel of semaphore tokens; send to acquire, receive to release.
ignoredDirs []string
added sync.Map // map[string]bool
}
// A symlinkList is a linked list of os.FileInfos for parent directories
// reached via symlinks.
type symlinkList struct {
info os.FileInfo
prev *symlinkList
}
// init initializes the walker based on its Options
func (w *walker) init() {
var ignoredPaths []string
if w.root.Type == RootModuleCache {
ignoredPaths = []string{"cache"}
}
if !w.opts.ModulesEnabled && w.root.Type == RootGOPATH {
ignoredPaths = w.getIgnoredDirs(w.root.Path)
ignoredPaths = append(ignoredPaths, "v", "mod")
}
for _, p := range ignoredPaths {
full := filepath.Join(w.root.Path, p)
w.ignoredDirs = append(w.ignoredDirs, full)
w.opts.Logf("Directory added to ignore list: %s", full)
}
}
// getIgnoredDirs reads an optional config file at <path>/.goimportsignore
// of relative directories to ignore when scanning for go files.
// The provided path is one of the $GOPATH entries with "src" appended.
func (w *walker) getIgnoredDirs(path string) []string {
file := filepath.Join(path, ".goimportsignore")
slurp, err := os.ReadFile(file)
if err != nil {
w.opts.Logf("%v", err)
} else {
w.opts.Logf("Read %s", file)
}
if err != nil {
return nil
}
var ignoredDirs []string
bs := bufio.NewScanner(bytes.NewReader(slurp))
for bs.Scan() {
line := strings.TrimSpace(bs.Text())
if line == "" || strings.HasPrefix(line, "#") {
continue
}
ignoredDirs = append(ignoredDirs, line)
}
return ignoredDirs
}
// shouldSkipDir reports whether the file should be skipped or not.
func (w *walker) shouldSkipDir(dir string) bool {
for _, ignoredDir := range w.ignoredDirs {
if dir == ignoredDir {
return true
}
}
if w.skip != nil {
// Check with the user specified callback.
return w.skip(w.root, dir)
}
return false
}
// walk walks through the given path.
//
// Errors are logged if w.opts.Logf is non-nil, but otherwise ignored.
func (w *walker) walk(path string, pathSymlinks *symlinkList, d fs.DirEntry) {
if d.Type()&os.ModeSymlink != 0 {
// Walk the symlink's target rather than the symlink itself.
//
// (Note that os.Stat, unlike the lower-lever os.Readlink,
// follows arbitrarily many layers of symlinks, so it will eventually
// reach either a non-symlink or a nonexistent target.)
//
// TODO(bcmills): 'go list all' itself ignores symlinks within GOROOT/src
// and GOPATH/src. Do we really need to traverse them here? If so, why?
fi, err := os.Stat(path)
if err != nil {
w.opts.Logf("%v", err)
return
}
// Avoid walking symlink cycles: if we have already followed a symlink to
// this directory as a parent of itself, don't follow it again.
//
// This doesn't catch the first time through a cycle, but it also minimizes
// the number of extra stat calls we make if we *don't* encounter a cycle.
// Since we don't actually expect to encounter symlink cycles in practice,
// this seems like the right tradeoff.
for parent := pathSymlinks; parent != nil; parent = parent.prev {
if os.SameFile(fi, parent.info) {
return
}
}
pathSymlinks = &symlinkList{
info: fi,
prev: pathSymlinks,
}
d = fs.FileInfoToDirEntry(fi)
}
if d.Type().IsRegular() {
if !strings.HasSuffix(path, ".go") {
return
}
dir := filepath.Dir(path)
if dir == w.root.Path && (w.root.Type == RootGOROOT || w.root.Type == RootGOPATH) {
// Doesn't make sense to have regular files
// directly in your $GOPATH/src or $GOROOT/src.
//
// TODO(bcmills): there are many levels of directory within
// RootModuleCache where this also wouldn't make sense,
// Can we generalize this to any directory without a corresponding
// import path?
return
}
if _, dup := w.added.LoadOrStore(dir, true); !dup {
w.add(w.root, dir)
}
}
if !d.IsDir() {
return
}
base := filepath.Base(path)
if base == "" || base[0] == '.' || base[0] == '_' ||
base == "testdata" ||
(w.root.Type == RootGOROOT && w.opts.ModulesEnabled && base == "vendor") ||
(!w.opts.ModulesEnabled && base == "node_modules") ||
w.shouldSkipDir(path) {
return
}
// Read the directory and walk its entries.
f, err := os.Open(path)
if err != nil {
w.opts.Logf("%v", err)
return
}
defer f.Close()
for {
// We impose an arbitrary limit on the number of ReadDir results per
// directory to limit the amount of memory consumed for stale or upcoming
// directory entries. The limit trades off CPU (number of syscalls to read
// the whole directory) against RAM (reachable directory entries other than
// the one currently being processed).
//
// Since we process the directories recursively, we will end up maintaining
// a slice of entries for each level of the directory tree.
// (Compare https://go.dev/issue/36197.)
ents, err := f.ReadDir(1024)
if err != nil {
if err != io.EOF {
w.opts.Logf("%v", err)
}
break
}
for _, d := range ents {
nextPath := filepath.Join(path, d.Name())
if d.IsDir() {
select {
case w.sem <- struct{}{}:
// Got a new semaphore token, so we can traverse the directory concurrently.
d := d
w.walking.Add(1)
go func() {
defer func() {
<-w.sem
w.walking.Done()
}()
w.walk(nextPath, pathSymlinks, d)
}()
continue
default:
// No tokens available, so traverse serially.
}
}
w.walk(nextPath, pathSymlinks, d)
}
}
}