// Copyright 2015 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. // The fiximports command fixes import declarations to use the canonical // import path for packages that have an "import comment" as defined by // https://golang.org/s/go14customimport. // // // Background // // The Go 1 custom import path mechanism lets the maintainer of a // package give it a stable name by which clients may import and "go // get" it, independent of the underlying version control system (such // as Git) or server (such as github.com) that hosts it. Requests for // the custom name are redirected to the underlying name. This allows // packages to be migrated from one underlying server or system to // another without breaking existing clients. // // Because this redirect mechanism creates aliases for existing // packages, it's possible for a single program to import the same // package by its canonical name and by an alias. The resulting // executable will contain two copies of the package, which is wasteful // at best and incorrect at worst. // // To avoid this, "go build" reports an error if it encounters a special // comment like the one below, and if the import path in the comment // does not match the path of the enclosing package relative to // GOPATH/src: // // $ grep ^package $GOPATH/src/github.com/bob/vanity/foo/foo.go // package foo // import "vanity.com/foo" // // The error from "go build" indicates that the package canonically // known as "vanity.com/foo" is locally installed under the // non-canonical name "github.com/bob/vanity/foo". // // // Usage // // When a package that you depend on introduces a custom import comment, // and your workspace imports it by the non-canonical name, your build // will stop working as soon as you update your copy of that package // using "go get -u". // // The purpose of the fiximports tool is to fix up all imports of the // non-canonical path within a Go workspace, replacing them with imports // of the canonical path. Following a run of fiximports, the workspace // will no longer depend on the non-canonical copy of the package, so it // should be safe to delete. It may be necessary to run "go get -u" // again to ensure that the package is locally installed under its // canonical path, if it was not already. // // The fiximports tool operates locally; it does not make HTTP requests // and does not discover new custom import comments. It only operates // on non-canonical packages present in your workspace. // // The -baddomains flag is a list of domain names that should always be // considered non-canonical. You can use this if you wish to make sure // that you no longer have any dependencies on packages from that // domain, even those that do not yet provide a canonical import path // comment. For example, the default value of -baddomains includes the // moribund code hosting site code.google.com, so fiximports will report // an error for each import of a package from this domain remaining // after canonicalization. // // To see the changes fiximports would make without applying them, use // the -n flag. // package main import ( "bytes" "encoding/json" "flag" "fmt" "go/ast" "go/build" "go/format" "go/parser" "go/token" "io" "io/ioutil" "log" "os" "os/exec" "path" "path/filepath" "sort" "strconv" "strings" ) // flags var ( dryrun = flag.Bool("n", false, "dry run: show changes, but don't apply them") badDomains = flag.String("baddomains", "code.google.com", "a comma-separated list of domains from which packages should not be imported") replaceFlag = flag.String("replace", "", "a comma-separated list of noncanonical=canonical pairs of package paths. If both items in a pair end with '...', they are treated as path prefixes.") ) // seams for testing var ( stderr io.Writer = os.Stderr writeFile = ioutil.WriteFile ) const usage = `fiximports: rewrite import paths to use canonical package names. Usage: fiximports [-n] package... The package... arguments specify a list of packages in the style of the go tool; see "go help packages". Hint: use "all" or "..." to match the entire workspace. For details, see http://godoc.org/golang.org/x/tools/cmd/fiximports. Flags: -n: dry run: show changes, but don't apply them -baddomains a comma-separated list of domains from which packages should not be imported ` func main() { flag.Parse() if len(flag.Args()) == 0 { fmt.Fprintf(stderr, usage) os.Exit(1) } if !fiximports(flag.Args()...) { os.Exit(1) } } type canonicalName struct{ path, name string } // fiximports fixes imports in the specified packages. // Invariant: a false result implies an error was already printed. func fiximports(packages ...string) bool { // importedBy is the transpose of the package import graph. importedBy := make(map[string]map[*build.Package]bool) // addEdge adds an edge to the import graph. addEdge := func(from *build.Package, to string) { if to == "C" || to == "unsafe" { return // fake } pkgs := importedBy[to] if pkgs == nil { pkgs = make(map[*build.Package]bool) importedBy[to] = pkgs } pkgs[from] = true } // List metadata for all packages in the workspace. pkgs, err := list("...") if err != nil { fmt.Fprintf(stderr, "importfix: %v\n", err) return false } // packageName maps each package's path to its name. packageName := make(map[string]string) for _, p := range pkgs { packageName[p.ImportPath] = p.Package.Name } // canonical maps each non-canonical package path to // its canonical path and name. // A present nil value indicates that the canonical package // is unknown: hosted on a bad domain with no redirect. canonical := make(map[string]canonicalName) domains := strings.Split(*badDomains, ",") type replaceItem struct { old, new string matchPrefix bool } var replace []replaceItem for _, pair := range strings.Split(*replaceFlag, ",") { if pair == "" { continue } words := strings.Split(pair, "=") if len(words) != 2 { fmt.Fprintf(stderr, "importfix: -replace: %q is not of the form \"canonical=noncanonical\".\n", pair) return false } replace = append(replace, replaceItem{ old: strings.TrimSuffix(words[0], "..."), new: strings.TrimSuffix(words[1], "..."), matchPrefix: strings.HasSuffix(words[0], "...") && strings.HasSuffix(words[1], "..."), }) } // Find non-canonical packages and populate importedBy graph. for _, p := range pkgs { if p.Error != nil { msg := p.Error.Err if strings.Contains(msg, "code in directory") && strings.Contains(msg, "expects import") { // don't show the very errors we're trying to fix } else { fmt.Fprintln(stderr, msg) } } for _, imp := range p.Imports { addEdge(&p.Package, imp) } for _, imp := range p.TestImports { addEdge(&p.Package, imp) } for _, imp := range p.XTestImports { addEdge(&p.Package, imp) } // Does package have an explicit import comment? if p.ImportComment != "" { if p.ImportComment != p.ImportPath { canonical[p.ImportPath] = canonicalName{ path: p.Package.ImportComment, name: p.Package.Name, } } } else { // Is package matched by a -replace item? var newPath string for _, item := range replace { if item.matchPrefix { if strings.HasPrefix(p.ImportPath, item.old) { newPath = item.new + p.ImportPath[len(item.old):] break } } else if p.ImportPath == item.old { newPath = item.new break } } if newPath != "" { newName := packageName[newPath] if newName == "" { newName = filepath.Base(newPath) // a guess } canonical[p.ImportPath] = canonicalName{ path: newPath, name: newName, } continue } // Is package matched by a -baddomains item? for _, domain := range domains { slash := strings.Index(p.ImportPath, "/") if slash < 0 { continue // no slash: standard package } if p.ImportPath[:slash] == domain { // Package comes from bad domain and has no import comment. // Report an error each time this package is imported. canonical[p.ImportPath] = canonicalName{} // TODO(adonovan): should we make an HTTP request to // see if there's an HTTP redirect, a "go-import" meta tag, // or an import comment in the the latest revision? // It would duplicate a lot of logic from "go get". } break } } } // Find all clients (direct importers) of canonical packages. // These are the packages that need fixing up. clients := make(map[*build.Package]bool) for path := range canonical { for client := range importedBy[path] { clients[client] = true } } // Restrict rewrites to the set of packages specified by the user. if len(packages) == 1 && (packages[0] == "all" || packages[0] == "...") { // no restriction } else { pkgs, err := list(packages...) if err != nil { fmt.Fprintf(stderr, "importfix: %v\n", err) return false } seen := make(map[string]bool) for _, p := range pkgs { seen[p.ImportPath] = true } for client := range clients { if !seen[client.ImportPath] { delete(clients, client) } } } // Rewrite selected client packages. ok := true for client := range clients { if !rewritePackage(client, canonical) { ok = false // There were errors. // Show direct and indirect imports of client. seen := make(map[string]bool) var direct, indirect []string for p := range importedBy[client.ImportPath] { direct = append(direct, p.ImportPath) seen[p.ImportPath] = true } var visit func(path string) visit = func(path string) { for q := range importedBy[path] { qpath := q.ImportPath if !seen[qpath] { seen[qpath] = true indirect = append(indirect, qpath) visit(qpath) } } } if direct != nil { fmt.Fprintf(stderr, "\timported directly by:\n") sort.Strings(direct) for _, path := range direct { fmt.Fprintf(stderr, "\t\t%s\n", path) visit(path) } if indirect != nil { fmt.Fprintf(stderr, "\timported indirectly by:\n") sort.Strings(indirect) for _, path := range indirect { fmt.Fprintf(stderr, "\t\t%s\n", path) } } } } } return ok } // Invariant: false result => error already printed. func rewritePackage(client *build.Package, canonical map[string]canonicalName) bool { ok := true used := make(map[string]bool) var filenames []string filenames = append(filenames, client.GoFiles...) filenames = append(filenames, client.TestGoFiles...) filenames = append(filenames, client.XTestGoFiles...) var first bool for _, filename := range filenames { if !first { first = true fmt.Fprintf(stderr, "%s\n", client.ImportPath) } err := rewriteFile(filepath.Join(client.Dir, filename), canonical, used) if err != nil { fmt.Fprintf(stderr, "\tERROR: %v\n", err) ok = false } } // Show which imports were renamed in this package. var keys []string for key := range used { keys = append(keys, key) } sort.Strings(keys) for _, key := range keys { if p := canonical[key]; p.path != "" { fmt.Fprintf(stderr, "\tfixed: %s -> %s\n", key, p.path) } else { fmt.Fprintf(stderr, "\tERROR: %s has no import comment\n", key) ok = false } } return ok } // rewrite reads, modifies, and writes filename, replacing all imports // of packages P in canonical by canonical[P]. // It records in used which canonical packages were imported. // used[P]=="" indicates that P was imported but its canonical path is unknown. func rewriteFile(filename string, canonical map[string]canonicalName, used map[string]bool) error { fset := token.NewFileSet() f, err := parser.ParseFile(fset, filename, nil, parser.ParseComments) if err != nil { return err } var changed bool for _, imp := range f.Imports { impPath, err := strconv.Unquote(imp.Path.Value) if err != nil { log.Printf("%s: bad import spec %q: %v", fset.Position(imp.Pos()), imp.Path.Value, err) continue } canon, ok := canonical[impPath] if !ok { continue // import path is canonical } used[impPath] = true if canon.path == "" { // The canonical path is unknown (a -baddomain). // Show the offending import. // TODO(adonovan): should we show the actual source text? fmt.Fprintf(stderr, "\t%s:%d: import %q\n", shortPath(filename), fset.Position(imp.Pos()).Line, impPath) continue } changed = true imp.Path.Value = strconv.Quote(canon.path) // Add a renaming import if necessary. // // This is a guess at best. We can't see whether a 'go // get' of the canonical import path would have the same // name or not. Assume it's the last segment. newBase := path.Base(canon.path) if imp.Name == nil && newBase != canon.name { imp.Name = &ast.Ident{Name: canon.name} } } if changed && !*dryrun { var buf bytes.Buffer if err := format.Node(&buf, fset, f); err != nil { return fmt.Errorf("%s: couldn't format file: %v", filename, err) } return writeFile(filename, buf.Bytes(), 0644) } return nil } // listPackage is a copy of cmd/go/list.Package. // It has more fields than build.Package and we need some of them. type listPackage struct { build.Package Error *packageError // error loading package } // A packageError describes an error loading information about a package. type packageError struct { ImportStack []string // shortest path from package named on command line to this one Pos string // position of error Err string // the error itself } // list runs 'go list' with the specified arguments and returns the // metadata for matching packages. func list(args ...string) ([]*listPackage, error) { cmd := exec.Command("go", append([]string{"list", "-e", "-json"}, args...)...) cmd.Stdout = new(bytes.Buffer) cmd.Stderr = stderr if err := cmd.Run(); err != nil { return nil, err } dec := json.NewDecoder(cmd.Stdout.(io.Reader)) var pkgs []*listPackage for { var p listPackage if err := dec.Decode(&p); err == io.EOF { break } else if err != nil { return nil, err } pkgs = append(pkgs, &p) } return pkgs, nil } // cwd contains the current working directory of the tool. // // It is initialized directly so that its value will be set for any other // package variables or init functions that depend on it, such as the gopath // variable in main_test.go. var cwd string = func() string { cwd, err := os.Getwd() if err != nil { log.Fatalf("os.Getwd: %v", err) } return cwd }() // shortPath returns an absolute or relative name for path, whatever is shorter. // Plundered from $GOROOT/src/cmd/go/build.go. func shortPath(path string) string { if rel, err := filepath.Rel(cwd, path); err == nil && len(rel) < len(path) { return rel } return path }