...
Run Format

Source file src/cmd/go/internal/get/vcs.go

Documentation: cmd/go/internal/get

     1  // Copyright 2012 The Go Authors. All rights reserved.
     2  // Use of this source code is governed by a BSD-style
     3  // license that can be found in the LICENSE file.
     4  
     5  package get
     6  
     7  import (
     8  	"encoding/json"
     9  	"errors"
    10  	"fmt"
    11  	"internal/singleflight"
    12  	"log"
    13  	"net/url"
    14  	"os"
    15  	"os/exec"
    16  	"path/filepath"
    17  	"regexp"
    18  	"strings"
    19  	"sync"
    20  
    21  	"cmd/go/internal/base"
    22  	"cmd/go/internal/cfg"
    23  	"cmd/go/internal/web"
    24  )
    25  
    26  // A vcsCmd describes how to use a version control system
    27  // like Mercurial, Git, or Subversion.
    28  type vcsCmd struct {
    29  	name string
    30  	cmd  string // name of binary to invoke command
    31  
    32  	createCmd   []string // commands to download a fresh copy of a repository
    33  	downloadCmd []string // commands to download updates into an existing repository
    34  
    35  	tagCmd         []tagCmd // commands to list tags
    36  	tagLookupCmd   []tagCmd // commands to lookup tags before running tagSyncCmd
    37  	tagSyncCmd     []string // commands to sync to specific tag
    38  	tagSyncDefault []string // commands to sync to default tag
    39  
    40  	scheme  []string
    41  	pingCmd string
    42  
    43  	remoteRepo  func(v *vcsCmd, rootDir string) (remoteRepo string, err error)
    44  	resolveRepo func(v *vcsCmd, rootDir, remoteRepo string) (realRepo string, err error)
    45  }
    46  
    47  var defaultSecureScheme = map[string]bool{
    48  	"https":   true,
    49  	"git+ssh": true,
    50  	"bzr+ssh": true,
    51  	"svn+ssh": true,
    52  	"ssh":     true,
    53  }
    54  
    55  func (v *vcsCmd) isSecure(repo string) bool {
    56  	u, err := url.Parse(repo)
    57  	if err != nil {
    58  		// If repo is not a URL, it's not secure.
    59  		return false
    60  	}
    61  	return v.isSecureScheme(u.Scheme)
    62  }
    63  
    64  func (v *vcsCmd) isSecureScheme(scheme string) bool {
    65  	switch v.cmd {
    66  	case "git":
    67  		// GIT_ALLOW_PROTOCOL is an environment variable defined by Git. It is a
    68  		// colon-separated list of schemes that are allowed to be used with git
    69  		// fetch/clone. Any scheme not mentioned will be considered insecure.
    70  		if allow := os.Getenv("GIT_ALLOW_PROTOCOL"); allow != "" {
    71  			for _, s := range strings.Split(allow, ":") {
    72  				if s == scheme {
    73  					return true
    74  				}
    75  			}
    76  			return false
    77  		}
    78  	}
    79  	return defaultSecureScheme[scheme]
    80  }
    81  
    82  // A tagCmd describes a command to list available tags
    83  // that can be passed to tagSyncCmd.
    84  type tagCmd struct {
    85  	cmd     string // command to list tags
    86  	pattern string // regexp to extract tags from list
    87  }
    88  
    89  // vcsList lists the known version control systems
    90  var vcsList = []*vcsCmd{
    91  	vcsHg,
    92  	vcsGit,
    93  	vcsSvn,
    94  	vcsBzr,
    95  	vcsFossil,
    96  }
    97  
    98  // vcsByCmd returns the version control system for the given
    99  // command name (hg, git, svn, bzr).
   100  func vcsByCmd(cmd string) *vcsCmd {
   101  	for _, vcs := range vcsList {
   102  		if vcs.cmd == cmd {
   103  			return vcs
   104  		}
   105  	}
   106  	return nil
   107  }
   108  
   109  // vcsHg describes how to use Mercurial.
   110  var vcsHg = &vcsCmd{
   111  	name: "Mercurial",
   112  	cmd:  "hg",
   113  
   114  	createCmd:   []string{"clone -U {repo} {dir}"},
   115  	downloadCmd: []string{"pull"},
   116  
   117  	// We allow both tag and branch names as 'tags'
   118  	// for selecting a version. This lets people have
   119  	// a go.release.r60 branch and a go1 branch
   120  	// and make changes in both, without constantly
   121  	// editing .hgtags.
   122  	tagCmd: []tagCmd{
   123  		{"tags", `^(\S+)`},
   124  		{"branches", `^(\S+)`},
   125  	},
   126  	tagSyncCmd:     []string{"update -r {tag}"},
   127  	tagSyncDefault: []string{"update default"},
   128  
   129  	scheme:     []string{"https", "http", "ssh"},
   130  	pingCmd:    "identify {scheme}://{repo}",
   131  	remoteRepo: hgRemoteRepo,
   132  }
   133  
   134  func hgRemoteRepo(vcsHg *vcsCmd, rootDir string) (remoteRepo string, err error) {
   135  	out, err := vcsHg.runOutput(rootDir, "paths default")
   136  	if err != nil {
   137  		return "", err
   138  	}
   139  	return strings.TrimSpace(string(out)), nil
   140  }
   141  
   142  // vcsGit describes how to use Git.
   143  var vcsGit = &vcsCmd{
   144  	name: "Git",
   145  	cmd:  "git",
   146  
   147  	createCmd:   []string{"clone {repo} {dir}", "-go-internal-cd {dir} submodule update --init --recursive"},
   148  	downloadCmd: []string{"pull --ff-only", "submodule update --init --recursive"},
   149  
   150  	tagCmd: []tagCmd{
   151  		// tags/xxx matches a git tag named xxx
   152  		// origin/xxx matches a git branch named xxx on the default remote repository
   153  		{"show-ref", `(?:tags|origin)/(\S+)$`},
   154  	},
   155  	tagLookupCmd: []tagCmd{
   156  		{"show-ref tags/{tag} origin/{tag}", `((?:tags|origin)/\S+)$`},
   157  	},
   158  	tagSyncCmd: []string{"checkout {tag}", "submodule update --init --recursive"},
   159  	// both createCmd and downloadCmd update the working dir.
   160  	// No need to do more here. We used to 'checkout master'
   161  	// but that doesn't work if the default branch is not named master.
   162  	// DO NOT add 'checkout master' here.
   163  	// See golang.org/issue/9032.
   164  	tagSyncDefault: []string{"submodule update --init --recursive"},
   165  
   166  	scheme:     []string{"git", "https", "http", "git+ssh", "ssh"},
   167  	pingCmd:    "ls-remote {scheme}://{repo}",
   168  	remoteRepo: gitRemoteRepo,
   169  }
   170  
   171  // scpSyntaxRe matches the SCP-like addresses used by Git to access
   172  // repositories by SSH.
   173  var scpSyntaxRe = regexp.MustCompile(`^([a-zA-Z0-9_]+)@([a-zA-Z0-9._-]+):(.*)$`)
   174  
   175  func gitRemoteRepo(vcsGit *vcsCmd, rootDir string) (remoteRepo string, err error) {
   176  	cmd := "config remote.origin.url"
   177  	errParse := errors.New("unable to parse output of git " + cmd)
   178  	errRemoteOriginNotFound := errors.New("remote origin not found")
   179  	outb, err := vcsGit.run1(rootDir, cmd, nil, false)
   180  	if err != nil {
   181  		// if it doesn't output any message, it means the config argument is correct,
   182  		// but the config value itself doesn't exist
   183  		if outb != nil && len(outb) == 0 {
   184  			return "", errRemoteOriginNotFound
   185  		}
   186  		return "", err
   187  	}
   188  	out := strings.TrimSpace(string(outb))
   189  
   190  	var repoURL *url.URL
   191  	if m := scpSyntaxRe.FindStringSubmatch(out); m != nil {
   192  		// Match SCP-like syntax and convert it to a URL.
   193  		// Eg, "git@github.com:user/repo" becomes
   194  		// "ssh://git@github.com/user/repo".
   195  		repoURL = &url.URL{
   196  			Scheme: "ssh",
   197  			User:   url.User(m[1]),
   198  			Host:   m[2],
   199  			Path:   m[3],
   200  		}
   201  	} else {
   202  		repoURL, err = url.Parse(out)
   203  		if err != nil {
   204  			return "", err
   205  		}
   206  	}
   207  
   208  	// Iterate over insecure schemes too, because this function simply
   209  	// reports the state of the repo. If we can't see insecure schemes then
   210  	// we can't report the actual repo URL.
   211  	for _, s := range vcsGit.scheme {
   212  		if repoURL.Scheme == s {
   213  			return repoURL.String(), nil
   214  		}
   215  	}
   216  	return "", errParse
   217  }
   218  
   219  // vcsBzr describes how to use Bazaar.
   220  var vcsBzr = &vcsCmd{
   221  	name: "Bazaar",
   222  	cmd:  "bzr",
   223  
   224  	createCmd: []string{"branch {repo} {dir}"},
   225  
   226  	// Without --overwrite bzr will not pull tags that changed.
   227  	// Replace by --overwrite-tags after http://pad.lv/681792 goes in.
   228  	downloadCmd: []string{"pull --overwrite"},
   229  
   230  	tagCmd:         []tagCmd{{"tags", `^(\S+)`}},
   231  	tagSyncCmd:     []string{"update -r {tag}"},
   232  	tagSyncDefault: []string{"update -r revno:-1"},
   233  
   234  	scheme:      []string{"https", "http", "bzr", "bzr+ssh"},
   235  	pingCmd:     "info {scheme}://{repo}",
   236  	remoteRepo:  bzrRemoteRepo,
   237  	resolveRepo: bzrResolveRepo,
   238  }
   239  
   240  func bzrRemoteRepo(vcsBzr *vcsCmd, rootDir string) (remoteRepo string, err error) {
   241  	outb, err := vcsBzr.runOutput(rootDir, "config parent_location")
   242  	if err != nil {
   243  		return "", err
   244  	}
   245  	return strings.TrimSpace(string(outb)), nil
   246  }
   247  
   248  func bzrResolveRepo(vcsBzr *vcsCmd, rootDir, remoteRepo string) (realRepo string, err error) {
   249  	outb, err := vcsBzr.runOutput(rootDir, "info "+remoteRepo)
   250  	if err != nil {
   251  		return "", err
   252  	}
   253  	out := string(outb)
   254  
   255  	// Expect:
   256  	// ...
   257  	//   (branch root|repository branch): <URL>
   258  	// ...
   259  
   260  	found := false
   261  	for _, prefix := range []string{"\n  branch root: ", "\n  repository branch: "} {
   262  		i := strings.Index(out, prefix)
   263  		if i >= 0 {
   264  			out = out[i+len(prefix):]
   265  			found = true
   266  			break
   267  		}
   268  	}
   269  	if !found {
   270  		return "", fmt.Errorf("unable to parse output of bzr info")
   271  	}
   272  
   273  	i := strings.Index(out, "\n")
   274  	if i < 0 {
   275  		return "", fmt.Errorf("unable to parse output of bzr info")
   276  	}
   277  	out = out[:i]
   278  	return strings.TrimSpace(out), nil
   279  }
   280  
   281  // vcsSvn describes how to use Subversion.
   282  var vcsSvn = &vcsCmd{
   283  	name: "Subversion",
   284  	cmd:  "svn",
   285  
   286  	createCmd:   []string{"checkout {repo} {dir}"},
   287  	downloadCmd: []string{"update"},
   288  
   289  	// There is no tag command in subversion.
   290  	// The branch information is all in the path names.
   291  
   292  	scheme:     []string{"https", "http", "svn", "svn+ssh"},
   293  	pingCmd:    "info {scheme}://{repo}",
   294  	remoteRepo: svnRemoteRepo,
   295  }
   296  
   297  func svnRemoteRepo(vcsSvn *vcsCmd, rootDir string) (remoteRepo string, err error) {
   298  	outb, err := vcsSvn.runOutput(rootDir, "info")
   299  	if err != nil {
   300  		return "", err
   301  	}
   302  	out := string(outb)
   303  
   304  	// Expect:
   305  	//
   306  	//	 ...
   307  	// 	URL: <URL>
   308  	// 	...
   309  	//
   310  	// Note that we're not using the Repository Root line,
   311  	// because svn allows checking out subtrees.
   312  	// The URL will be the URL of the subtree (what we used with 'svn co')
   313  	// while the Repository Root may be a much higher parent.
   314  	i := strings.Index(out, "\nURL: ")
   315  	if i < 0 {
   316  		return "", fmt.Errorf("unable to parse output of svn info")
   317  	}
   318  	out = out[i+len("\nURL: "):]
   319  	i = strings.Index(out, "\n")
   320  	if i < 0 {
   321  		return "", fmt.Errorf("unable to parse output of svn info")
   322  	}
   323  	out = out[:i]
   324  	return strings.TrimSpace(out), nil
   325  }
   326  
   327  // fossilRepoName is the name go get associates with a fossil repository. In the
   328  // real world the file can be named anything.
   329  const fossilRepoName = ".fossil"
   330  
   331  // vcsFossil describes how to use Fossil (fossil-scm.org)
   332  var vcsFossil = &vcsCmd{
   333  	name: "Fossil",
   334  	cmd:  "fossil",
   335  
   336  	createCmd:   []string{"-go-internal-mkdir {dir} clone {repo} " + filepath.Join("{dir}", fossilRepoName), "-go-internal-cd {dir} open .fossil"},
   337  	downloadCmd: []string{"up"},
   338  
   339  	tagCmd:         []tagCmd{{"tag ls", `(.*)`}},
   340  	tagSyncCmd:     []string{"up tag:{tag}"},
   341  	tagSyncDefault: []string{"up trunk"},
   342  
   343  	scheme:     []string{"https", "http"},
   344  	remoteRepo: fossilRemoteRepo,
   345  }
   346  
   347  func fossilRemoteRepo(vcsFossil *vcsCmd, rootDir string) (remoteRepo string, err error) {
   348  	out, err := vcsFossil.runOutput(rootDir, "remote-url")
   349  	if err != nil {
   350  		return "", err
   351  	}
   352  	return strings.TrimSpace(string(out)), nil
   353  }
   354  
   355  func (v *vcsCmd) String() string {
   356  	return v.name
   357  }
   358  
   359  // run runs the command line cmd in the given directory.
   360  // keyval is a list of key, value pairs. run expands
   361  // instances of {key} in cmd into value, but only after
   362  // splitting cmd into individual arguments.
   363  // If an error occurs, run prints the command line and the
   364  // command's combined stdout+stderr to standard error.
   365  // Otherwise run discards the command's output.
   366  func (v *vcsCmd) run(dir string, cmd string, keyval ...string) error {
   367  	_, err := v.run1(dir, cmd, keyval, true)
   368  	return err
   369  }
   370  
   371  // runVerboseOnly is like run but only generates error output to standard error in verbose mode.
   372  func (v *vcsCmd) runVerboseOnly(dir string, cmd string, keyval ...string) error {
   373  	_, err := v.run1(dir, cmd, keyval, false)
   374  	return err
   375  }
   376  
   377  // runOutput is like run but returns the output of the command.
   378  func (v *vcsCmd) runOutput(dir string, cmd string, keyval ...string) ([]byte, error) {
   379  	return v.run1(dir, cmd, keyval, true)
   380  }
   381  
   382  // run1 is the generalized implementation of run and runOutput.
   383  func (v *vcsCmd) run1(dir string, cmdline string, keyval []string, verbose bool) ([]byte, error) {
   384  	m := make(map[string]string)
   385  	for i := 0; i < len(keyval); i += 2 {
   386  		m[keyval[i]] = keyval[i+1]
   387  	}
   388  	args := strings.Fields(cmdline)
   389  	for i, arg := range args {
   390  		args[i] = expand(m, arg)
   391  	}
   392  
   393  	if len(args) >= 2 && args[0] == "-go-internal-mkdir" {
   394  		var err error
   395  		if filepath.IsAbs(args[1]) {
   396  			err = os.Mkdir(args[1], os.ModePerm)
   397  		} else {
   398  			err = os.Mkdir(filepath.Join(dir, args[1]), os.ModePerm)
   399  		}
   400  		if err != nil {
   401  			return nil, err
   402  		}
   403  		args = args[2:]
   404  	}
   405  
   406  	if len(args) >= 2 && args[0] == "-go-internal-cd" {
   407  		if filepath.IsAbs(args[1]) {
   408  			dir = args[1]
   409  		} else {
   410  			dir = filepath.Join(dir, args[1])
   411  		}
   412  		args = args[2:]
   413  	}
   414  
   415  	_, err := exec.LookPath(v.cmd)
   416  	if err != nil {
   417  		fmt.Fprintf(os.Stderr,
   418  			"go: missing %s command. See https://golang.org/s/gogetcmd\n",
   419  			v.name)
   420  		return nil, err
   421  	}
   422  
   423  	cmd := exec.Command(v.cmd, args...)
   424  	cmd.Dir = dir
   425  	cmd.Env = base.EnvForDir(cmd.Dir, os.Environ())
   426  	if cfg.BuildX {
   427  		fmt.Printf("cd %s\n", dir)
   428  		fmt.Printf("%s %s\n", v.cmd, strings.Join(args, " "))
   429  	}
   430  	out, err := cmd.Output()
   431  	if err != nil {
   432  		if verbose || cfg.BuildV {
   433  			fmt.Fprintf(os.Stderr, "# cd %s; %s %s\n", dir, v.cmd, strings.Join(args, " "))
   434  			if ee, ok := err.(*exec.ExitError); ok && len(ee.Stderr) > 0 {
   435  				os.Stderr.Write(ee.Stderr)
   436  			} else {
   437  				fmt.Fprintf(os.Stderr, err.Error())
   438  			}
   439  		}
   440  	}
   441  	return out, err
   442  }
   443  
   444  // ping pings to determine scheme to use.
   445  func (v *vcsCmd) ping(scheme, repo string) error {
   446  	return v.runVerboseOnly(".", v.pingCmd, "scheme", scheme, "repo", repo)
   447  }
   448  
   449  // create creates a new copy of repo in dir.
   450  // The parent of dir must exist; dir must not.
   451  func (v *vcsCmd) create(dir, repo string) error {
   452  	for _, cmd := range v.createCmd {
   453  		if err := v.run(".", cmd, "dir", dir, "repo", repo); err != nil {
   454  			return err
   455  		}
   456  	}
   457  	return nil
   458  }
   459  
   460  // download downloads any new changes for the repo in dir.
   461  func (v *vcsCmd) download(dir string) error {
   462  	for _, cmd := range v.downloadCmd {
   463  		if err := v.run(dir, cmd); err != nil {
   464  			return err
   465  		}
   466  	}
   467  	return nil
   468  }
   469  
   470  // tags returns the list of available tags for the repo in dir.
   471  func (v *vcsCmd) tags(dir string) ([]string, error) {
   472  	var tags []string
   473  	for _, tc := range v.tagCmd {
   474  		out, err := v.runOutput(dir, tc.cmd)
   475  		if err != nil {
   476  			return nil, err
   477  		}
   478  		re := regexp.MustCompile(`(?m-s)` + tc.pattern)
   479  		for _, m := range re.FindAllStringSubmatch(string(out), -1) {
   480  			tags = append(tags, m[1])
   481  		}
   482  	}
   483  	return tags, nil
   484  }
   485  
   486  // tagSync syncs the repo in dir to the named tag,
   487  // which either is a tag returned by tags or is v.tagDefault.
   488  func (v *vcsCmd) tagSync(dir, tag string) error {
   489  	if v.tagSyncCmd == nil {
   490  		return nil
   491  	}
   492  	if tag != "" {
   493  		for _, tc := range v.tagLookupCmd {
   494  			out, err := v.runOutput(dir, tc.cmd, "tag", tag)
   495  			if err != nil {
   496  				return err
   497  			}
   498  			re := regexp.MustCompile(`(?m-s)` + tc.pattern)
   499  			m := re.FindStringSubmatch(string(out))
   500  			if len(m) > 1 {
   501  				tag = m[1]
   502  				break
   503  			}
   504  		}
   505  	}
   506  
   507  	if tag == "" && v.tagSyncDefault != nil {
   508  		for _, cmd := range v.tagSyncDefault {
   509  			if err := v.run(dir, cmd); err != nil {
   510  				return err
   511  			}
   512  		}
   513  		return nil
   514  	}
   515  
   516  	for _, cmd := range v.tagSyncCmd {
   517  		if err := v.run(dir, cmd, "tag", tag); err != nil {
   518  			return err
   519  		}
   520  	}
   521  	return nil
   522  }
   523  
   524  // A vcsPath describes how to convert an import path into a
   525  // version control system and repository name.
   526  type vcsPath struct {
   527  	prefix string                              // prefix this description applies to
   528  	re     string                              // pattern for import path
   529  	repo   string                              // repository to use (expand with match of re)
   530  	vcs    string                              // version control system to use (expand with match of re)
   531  	check  func(match map[string]string) error // additional checks
   532  	ping   bool                                // ping for scheme to use to download repo
   533  
   534  	regexp *regexp.Regexp // cached compiled form of re
   535  }
   536  
   537  // vcsFromDir inspects dir and its parents to determine the
   538  // version control system and code repository to use.
   539  // On return, root is the import path
   540  // corresponding to the root of the repository.
   541  func vcsFromDir(dir, srcRoot string) (vcs *vcsCmd, root string, err error) {
   542  	// Clean and double-check that dir is in (a subdirectory of) srcRoot.
   543  	dir = filepath.Clean(dir)
   544  	srcRoot = filepath.Clean(srcRoot)
   545  	if len(dir) <= len(srcRoot) || dir[len(srcRoot)] != filepath.Separator {
   546  		return nil, "", fmt.Errorf("directory %q is outside source root %q", dir, srcRoot)
   547  	}
   548  
   549  	var vcsRet *vcsCmd
   550  	var rootRet string
   551  
   552  	origDir := dir
   553  	for len(dir) > len(srcRoot) {
   554  		for _, vcs := range vcsList {
   555  			if _, err := os.Stat(filepath.Join(dir, "."+vcs.cmd)); err == nil {
   556  				root := filepath.ToSlash(dir[len(srcRoot)+1:])
   557  				// Record first VCS we find, but keep looking,
   558  				// to detect mistakes like one kind of VCS inside another.
   559  				if vcsRet == nil {
   560  					vcsRet = vcs
   561  					rootRet = root
   562  					continue
   563  				}
   564  				// Allow .git inside .git, which can arise due to submodules.
   565  				if vcsRet == vcs && vcs.cmd == "git" {
   566  					continue
   567  				}
   568  				// Otherwise, we have one VCS inside a different VCS.
   569  				return nil, "", fmt.Errorf("directory %q uses %s, but parent %q uses %s",
   570  					filepath.Join(srcRoot, rootRet), vcsRet.cmd, filepath.Join(srcRoot, root), vcs.cmd)
   571  			}
   572  		}
   573  
   574  		// Move to parent.
   575  		ndir := filepath.Dir(dir)
   576  		if len(ndir) >= len(dir) {
   577  			// Shouldn't happen, but just in case, stop.
   578  			break
   579  		}
   580  		dir = ndir
   581  	}
   582  
   583  	if vcsRet != nil {
   584  		return vcsRet, rootRet, nil
   585  	}
   586  
   587  	return nil, "", fmt.Errorf("directory %q is not using a known version control system", origDir)
   588  }
   589  
   590  // checkNestedVCS checks for an incorrectly-nested VCS-inside-VCS
   591  // situation for dir, checking parents up until srcRoot.
   592  func checkNestedVCS(vcs *vcsCmd, dir, srcRoot string) error {
   593  	if len(dir) <= len(srcRoot) || dir[len(srcRoot)] != filepath.Separator {
   594  		return fmt.Errorf("directory %q is outside source root %q", dir, srcRoot)
   595  	}
   596  
   597  	otherDir := dir
   598  	for len(otherDir) > len(srcRoot) {
   599  		for _, otherVCS := range vcsList {
   600  			if _, err := os.Stat(filepath.Join(otherDir, "."+otherVCS.cmd)); err == nil {
   601  				// Allow expected vcs in original dir.
   602  				if otherDir == dir && otherVCS == vcs {
   603  					continue
   604  				}
   605  				// Allow .git inside .git, which can arise due to submodules.
   606  				if otherVCS == vcs && vcs.cmd == "git" {
   607  					continue
   608  				}
   609  				// Otherwise, we have one VCS inside a different VCS.
   610  				return fmt.Errorf("directory %q uses %s, but parent %q uses %s", dir, vcs.cmd, otherDir, otherVCS.cmd)
   611  			}
   612  		}
   613  		// Move to parent.
   614  		newDir := filepath.Dir(otherDir)
   615  		if len(newDir) >= len(otherDir) {
   616  			// Shouldn't happen, but just in case, stop.
   617  			break
   618  		}
   619  		otherDir = newDir
   620  	}
   621  
   622  	return nil
   623  }
   624  
   625  // RepoRoot describes the repository root for a tree of source code.
   626  type RepoRoot struct {
   627  	Repo     string // repository URL, including scheme
   628  	Root     string // import path corresponding to root of repo
   629  	IsCustom bool   // defined by served <meta> tags (as opposed to hard-coded pattern)
   630  	VCS      string // vcs type ("mod", "git", ...)
   631  
   632  	vcs *vcsCmd // internal: vcs command access
   633  }
   634  
   635  var httpPrefixRE = regexp.MustCompile(`^https?:`)
   636  
   637  // ModuleMode specifies whether to prefer modules when looking up code sources.
   638  type ModuleMode int
   639  
   640  const (
   641  	IgnoreMod ModuleMode = iota
   642  	PreferMod
   643  )
   644  
   645  // RepoRootForImportPath analyzes importPath to determine the
   646  // version control system, and code repository to use.
   647  func RepoRootForImportPath(importPath string, mod ModuleMode, security web.SecurityMode) (*RepoRoot, error) {
   648  	rr, err := repoRootFromVCSPaths(importPath, "", security, vcsPaths)
   649  	if err == errUnknownSite {
   650  		rr, err = repoRootForImportDynamic(importPath, mod, security)
   651  		if err != nil {
   652  			err = fmt.Errorf("unrecognized import path %q (%v)", importPath, err)
   653  		}
   654  	}
   655  	if err != nil {
   656  		rr1, err1 := repoRootFromVCSPaths(importPath, "", security, vcsPathsAfterDynamic)
   657  		if err1 == nil {
   658  			rr = rr1
   659  			err = nil
   660  		}
   661  	}
   662  
   663  	// Should have been taken care of above, but make sure.
   664  	if err == nil && strings.Contains(importPath, "...") && strings.Contains(rr.Root, "...") {
   665  		// Do not allow wildcards in the repo root.
   666  		rr = nil
   667  		err = fmt.Errorf("cannot expand ... in %q", importPath)
   668  	}
   669  	return rr, err
   670  }
   671  
   672  var errUnknownSite = errors.New("dynamic lookup required to find mapping")
   673  
   674  // repoRootFromVCSPaths attempts to map importPath to a repoRoot
   675  // using the mappings defined in vcsPaths.
   676  // If scheme is non-empty, that scheme is forced.
   677  func repoRootFromVCSPaths(importPath, scheme string, security web.SecurityMode, vcsPaths []*vcsPath) (*RepoRoot, error) {
   678  	// A common error is to use https://packagepath because that's what
   679  	// hg and git require. Diagnose this helpfully.
   680  	if loc := httpPrefixRE.FindStringIndex(importPath); loc != nil {
   681  		// The importPath has been cleaned, so has only one slash. The pattern
   682  		// ignores the slashes; the error message puts them back on the RHS at least.
   683  		return nil, fmt.Errorf("%q not allowed in import path", importPath[loc[0]:loc[1]]+"//")
   684  	}
   685  	for _, srv := range vcsPaths {
   686  		if !strings.HasPrefix(importPath, srv.prefix) {
   687  			continue
   688  		}
   689  		m := srv.regexp.FindStringSubmatch(importPath)
   690  		if m == nil {
   691  			if srv.prefix != "" {
   692  				return nil, fmt.Errorf("invalid %s import path %q", srv.prefix, importPath)
   693  			}
   694  			continue
   695  		}
   696  
   697  		// Build map of named subexpression matches for expand.
   698  		match := map[string]string{
   699  			"prefix": srv.prefix,
   700  			"import": importPath,
   701  		}
   702  		for i, name := range srv.regexp.SubexpNames() {
   703  			if name != "" && match[name] == "" {
   704  				match[name] = m[i]
   705  			}
   706  		}
   707  		if srv.vcs != "" {
   708  			match["vcs"] = expand(match, srv.vcs)
   709  		}
   710  		if srv.repo != "" {
   711  			match["repo"] = expand(match, srv.repo)
   712  		}
   713  		if srv.check != nil {
   714  			if err := srv.check(match); err != nil {
   715  				return nil, err
   716  			}
   717  		}
   718  		vcs := vcsByCmd(match["vcs"])
   719  		if vcs == nil {
   720  			return nil, fmt.Errorf("unknown version control system %q", match["vcs"])
   721  		}
   722  		if srv.ping {
   723  			if scheme != "" {
   724  				match["repo"] = scheme + "://" + match["repo"]
   725  			} else {
   726  				for _, scheme := range vcs.scheme {
   727  					if security == web.Secure && !vcs.isSecureScheme(scheme) {
   728  						continue
   729  					}
   730  					if vcs.pingCmd != "" && vcs.ping(scheme, match["repo"]) == nil {
   731  						match["repo"] = scheme + "://" + match["repo"]
   732  						goto Found
   733  					}
   734  				}
   735  				// No scheme found. Fall back to the first one.
   736  				match["repo"] = vcs.scheme[0] + "://" + match["repo"]
   737  			Found:
   738  			}
   739  		}
   740  		rr := &RepoRoot{
   741  			Repo: match["repo"],
   742  			Root: match["root"],
   743  			VCS:  vcs.cmd,
   744  			vcs:  vcs,
   745  		}
   746  		return rr, nil
   747  	}
   748  	return nil, errUnknownSite
   749  }
   750  
   751  // repoRootForImportDynamic finds a *RepoRoot for a custom domain that's not
   752  // statically known by repoRootForImportPathStatic.
   753  //
   754  // This handles custom import paths like "name.tld/pkg/foo" or just "name.tld".
   755  func repoRootForImportDynamic(importPath string, mod ModuleMode, security web.SecurityMode) (*RepoRoot, error) {
   756  	slash := strings.Index(importPath, "/")
   757  	if slash < 0 {
   758  		slash = len(importPath)
   759  	}
   760  	host := importPath[:slash]
   761  	if !strings.Contains(host, ".") {
   762  		return nil, errors.New("import path does not begin with hostname")
   763  	}
   764  	urlStr, body, err := web.GetMaybeInsecure(importPath, security)
   765  	if err != nil {
   766  		msg := "https fetch: %v"
   767  		if security == web.Insecure {
   768  			msg = "http/" + msg
   769  		}
   770  		return nil, fmt.Errorf(msg, err)
   771  	}
   772  	defer body.Close()
   773  	imports, err := parseMetaGoImports(body, mod)
   774  	if err != nil {
   775  		return nil, fmt.Errorf("parsing %s: %v", importPath, err)
   776  	}
   777  	// Find the matched meta import.
   778  	mmi, err := matchGoImport(imports, importPath)
   779  	if err != nil {
   780  		if _, ok := err.(ImportMismatchError); !ok {
   781  			return nil, fmt.Errorf("parse %s: %v", urlStr, err)
   782  		}
   783  		return nil, fmt.Errorf("parse %s: no go-import meta tags (%s)", urlStr, err)
   784  	}
   785  	if cfg.BuildV {
   786  		log.Printf("get %q: found meta tag %#v at %s", importPath, mmi, urlStr)
   787  	}
   788  	// If the import was "uni.edu/bob/project", which said the
   789  	// prefix was "uni.edu" and the RepoRoot was "evilroot.com",
   790  	// make sure we don't trust Bob and check out evilroot.com to
   791  	// "uni.edu" yet (possibly overwriting/preempting another
   792  	// non-evil student). Instead, first verify the root and see
   793  	// if it matches Bob's claim.
   794  	if mmi.Prefix != importPath {
   795  		if cfg.BuildV {
   796  			log.Printf("get %q: verifying non-authoritative meta tag", importPath)
   797  		}
   798  		urlStr0 := urlStr
   799  		var imports []metaImport
   800  		urlStr, imports, err = metaImportsForPrefix(mmi.Prefix, mod, security)
   801  		if err != nil {
   802  			return nil, err
   803  		}
   804  		metaImport2, err := matchGoImport(imports, importPath)
   805  		if err != nil || mmi != metaImport2 {
   806  			return nil, fmt.Errorf("%s and %s disagree about go-import for %s", urlStr0, urlStr, mmi.Prefix)
   807  		}
   808  	}
   809  
   810  	if err := validateRepoRoot(mmi.RepoRoot); err != nil {
   811  		return nil, fmt.Errorf("%s: invalid repo root %q: %v", urlStr, mmi.RepoRoot, err)
   812  	}
   813  	vcs := vcsByCmd(mmi.VCS)
   814  	if vcs == nil && mmi.VCS != "mod" {
   815  		return nil, fmt.Errorf("%s: unknown vcs %q", urlStr, mmi.VCS)
   816  	}
   817  
   818  	rr := &RepoRoot{
   819  		Repo:     mmi.RepoRoot,
   820  		Root:     mmi.Prefix,
   821  		IsCustom: true,
   822  		VCS:      mmi.VCS,
   823  		vcs:      vcs,
   824  	}
   825  	return rr, nil
   826  }
   827  
   828  // validateRepoRoot returns an error if repoRoot does not seem to be
   829  // a valid URL with scheme.
   830  func validateRepoRoot(repoRoot string) error {
   831  	url, err := url.Parse(repoRoot)
   832  	if err != nil {
   833  		return err
   834  	}
   835  	if url.Scheme == "" {
   836  		return errors.New("no scheme")
   837  	}
   838  	return nil
   839  }
   840  
   841  var fetchGroup singleflight.Group
   842  var (
   843  	fetchCacheMu sync.Mutex
   844  	fetchCache   = map[string]fetchResult{} // key is metaImportsForPrefix's importPrefix
   845  )
   846  
   847  // metaImportsForPrefix takes a package's root import path as declared in a <meta> tag
   848  // and returns its HTML discovery URL and the parsed metaImport lines
   849  // found on the page.
   850  //
   851  // The importPath is of the form "golang.org/x/tools".
   852  // It is an error if no imports are found.
   853  // urlStr will still be valid if err != nil.
   854  // The returned urlStr will be of the form "https://golang.org/x/tools?go-get=1"
   855  func metaImportsForPrefix(importPrefix string, mod ModuleMode, security web.SecurityMode) (urlStr string, imports []metaImport, err error) {
   856  	setCache := func(res fetchResult) (fetchResult, error) {
   857  		fetchCacheMu.Lock()
   858  		defer fetchCacheMu.Unlock()
   859  		fetchCache[importPrefix] = res
   860  		return res, nil
   861  	}
   862  
   863  	resi, _, _ := fetchGroup.Do(importPrefix, func() (resi interface{}, err error) {
   864  		fetchCacheMu.Lock()
   865  		if res, ok := fetchCache[importPrefix]; ok {
   866  			fetchCacheMu.Unlock()
   867  			return res, nil
   868  		}
   869  		fetchCacheMu.Unlock()
   870  
   871  		urlStr, body, err := web.GetMaybeInsecure(importPrefix, security)
   872  		if err != nil {
   873  			return setCache(fetchResult{urlStr: urlStr, err: fmt.Errorf("fetch %s: %v", urlStr, err)})
   874  		}
   875  		imports, err := parseMetaGoImports(body, mod)
   876  		if err != nil {
   877  			return setCache(fetchResult{urlStr: urlStr, err: fmt.Errorf("parsing %s: %v", urlStr, err)})
   878  		}
   879  		if len(imports) == 0 {
   880  			err = fmt.Errorf("fetch %s: no go-import meta tag", urlStr)
   881  		}
   882  		return setCache(fetchResult{urlStr: urlStr, imports: imports, err: err})
   883  	})
   884  	res := resi.(fetchResult)
   885  	return res.urlStr, res.imports, res.err
   886  }
   887  
   888  type fetchResult struct {
   889  	urlStr  string // e.g. "https://foo.com/x/bar?go-get=1"
   890  	imports []metaImport
   891  	err     error
   892  }
   893  
   894  // metaImport represents the parsed <meta name="go-import"
   895  // content="prefix vcs reporoot" /> tags from HTML files.
   896  type metaImport struct {
   897  	Prefix, VCS, RepoRoot string
   898  }
   899  
   900  func splitPathHasPrefix(path, prefix []string) bool {
   901  	if len(path) < len(prefix) {
   902  		return false
   903  	}
   904  	for i, p := range prefix {
   905  		if path[i] != p {
   906  			return false
   907  		}
   908  	}
   909  	return true
   910  }
   911  
   912  // A ImportMismatchError is returned where metaImport/s are present
   913  // but none match our import path.
   914  type ImportMismatchError struct {
   915  	importPath string
   916  	mismatches []string // the meta imports that were discarded for not matching our importPath
   917  }
   918  
   919  func (m ImportMismatchError) Error() string {
   920  	formattedStrings := make([]string, len(m.mismatches))
   921  	for i, pre := range m.mismatches {
   922  		formattedStrings[i] = fmt.Sprintf("meta tag %s did not match import path %s", pre, m.importPath)
   923  	}
   924  	return strings.Join(formattedStrings, ", ")
   925  }
   926  
   927  // matchGoImport returns the metaImport from imports matching importPath.
   928  // An error is returned if there are multiple matches.
   929  // errNoMatch is returned if none match.
   930  func matchGoImport(imports []metaImport, importPath string) (metaImport, error) {
   931  	match := -1
   932  	imp := strings.Split(importPath, "/")
   933  
   934  	errImportMismatch := ImportMismatchError{importPath: importPath}
   935  	for i, im := range imports {
   936  		pre := strings.Split(im.Prefix, "/")
   937  
   938  		if !splitPathHasPrefix(imp, pre) {
   939  			errImportMismatch.mismatches = append(errImportMismatch.mismatches, im.Prefix)
   940  			continue
   941  		}
   942  
   943  		if match >= 0 {
   944  			if imports[match].VCS == "mod" && im.VCS != "mod" {
   945  				// All the mod entries precede all the non-mod entries.
   946  				// We have a mod entry and don't care about the rest,
   947  				// matching or not.
   948  				break
   949  			}
   950  			return metaImport{}, fmt.Errorf("multiple meta tags match import path %q", importPath)
   951  		}
   952  		match = i
   953  	}
   954  
   955  	if match == -1 {
   956  		return metaImport{}, errImportMismatch
   957  	}
   958  	return imports[match], nil
   959  }
   960  
   961  // expand rewrites s to replace {k} with match[k] for each key k in match.
   962  func expand(match map[string]string, s string) string {
   963  	// We want to replace each match exactly once, and the result of expansion
   964  	// must not depend on the iteration order through the map.
   965  	// A strings.Replacer has exactly the properties we're looking for.
   966  	oldNew := make([]string, 0, 2*len(match))
   967  	for k, v := range match {
   968  		oldNew = append(oldNew, "{"+k+"}", v)
   969  	}
   970  	return strings.NewReplacer(oldNew...).Replace(s)
   971  }
   972  
   973  // vcsPaths defines the meaning of import paths referring to
   974  // commonly-used VCS hosting sites (github.com/user/dir)
   975  // and import paths referring to a fully-qualified importPath
   976  // containing a VCS type (foo.com/repo.git/dir)
   977  var vcsPaths = []*vcsPath{
   978  	// Github
   979  	{
   980  		prefix: "github.com/",
   981  		re:     `^(?P<root>github\.com/[A-Za-z0-9_.\-]+/[A-Za-z0-9_.\-]+)(/[\p{L}0-9_.\-]+)*$`,
   982  		vcs:    "git",
   983  		repo:   "https://{root}",
   984  		check:  noVCSSuffix,
   985  	},
   986  
   987  	// Bitbucket
   988  	{
   989  		prefix: "bitbucket.org/",
   990  		re:     `^(?P<root>bitbucket\.org/(?P<bitname>[A-Za-z0-9_.\-]+/[A-Za-z0-9_.\-]+))(/[A-Za-z0-9_.\-]+)*$`,
   991  		repo:   "https://{root}",
   992  		check:  bitbucketVCS,
   993  	},
   994  
   995  	// IBM DevOps Services (JazzHub)
   996  	{
   997  		prefix: "hub.jazz.net/git/",
   998  		re:     `^(?P<root>hub\.jazz\.net/git/[a-z0-9]+/[A-Za-z0-9_.\-]+)(/[A-Za-z0-9_.\-]+)*$`,
   999  		vcs:    "git",
  1000  		repo:   "https://{root}",
  1001  		check:  noVCSSuffix,
  1002  	},
  1003  
  1004  	// Git at Apache
  1005  	{
  1006  		prefix: "git.apache.org/",
  1007  		re:     `^(?P<root>git\.apache\.org/[a-z0-9_.\-]+\.git)(/[A-Za-z0-9_.\-]+)*$`,
  1008  		vcs:    "git",
  1009  		repo:   "https://{root}",
  1010  	},
  1011  
  1012  	// Git at OpenStack
  1013  	{
  1014  		prefix: "git.openstack.org/",
  1015  		re:     `^(?P<root>git\.openstack\.org/[A-Za-z0-9_.\-]+/[A-Za-z0-9_.\-]+)(\.git)?(/[A-Za-z0-9_.\-]+)*$`,
  1016  		vcs:    "git",
  1017  		repo:   "https://{root}",
  1018  	},
  1019  
  1020  	// chiselapp.com for fossil
  1021  	{
  1022  		prefix: "chiselapp.com/",
  1023  		re:     `^(?P<root>chiselapp\.com/user/[A-Za-z0-9]+/repository/[A-Za-z0-9_.\-]+)$`,
  1024  		vcs:    "fossil",
  1025  		repo:   "https://{root}",
  1026  	},
  1027  
  1028  	// General syntax for any server.
  1029  	// Must be last.
  1030  	{
  1031  		re:   `^(?P<root>(?P<repo>([a-z0-9.\-]+\.)+[a-z0-9.\-]+(:[0-9]+)?(/~?[A-Za-z0-9_.\-]+)+?)\.(?P<vcs>bzr|fossil|git|hg|svn))(/~?[A-Za-z0-9_.\-]+)*$`,
  1032  		ping: true,
  1033  	},
  1034  }
  1035  
  1036  // vcsPathsAfterDynamic gives additional vcsPaths entries
  1037  // to try after the dynamic HTML check.
  1038  // This gives those sites a chance to introduce <meta> tags
  1039  // as part of a graceful transition away from the hard-coded logic.
  1040  var vcsPathsAfterDynamic = []*vcsPath{
  1041  	// Launchpad. See golang.org/issue/11436.
  1042  	{
  1043  		prefix: "launchpad.net/",
  1044  		re:     `^(?P<root>launchpad\.net/((?P<project>[A-Za-z0-9_.\-]+)(?P<series>/[A-Za-z0-9_.\-]+)?|~[A-Za-z0-9_.\-]+/(\+junk|[A-Za-z0-9_.\-]+)/[A-Za-z0-9_.\-]+))(/[A-Za-z0-9_.\-]+)*$`,
  1045  		vcs:    "bzr",
  1046  		repo:   "https://{root}",
  1047  		check:  launchpadVCS,
  1048  	},
  1049  }
  1050  
  1051  func init() {
  1052  	// fill in cached regexps.
  1053  	// Doing this eagerly discovers invalid regexp syntax
  1054  	// without having to run a command that needs that regexp.
  1055  	for _, srv := range vcsPaths {
  1056  		srv.regexp = regexp.MustCompile(srv.re)
  1057  	}
  1058  	for _, srv := range vcsPathsAfterDynamic {
  1059  		srv.regexp = regexp.MustCompile(srv.re)
  1060  	}
  1061  }
  1062  
  1063  // noVCSSuffix checks that the repository name does not
  1064  // end in .foo for any version control system foo.
  1065  // The usual culprit is ".git".
  1066  func noVCSSuffix(match map[string]string) error {
  1067  	repo := match["repo"]
  1068  	for _, vcs := range vcsList {
  1069  		if strings.HasSuffix(repo, "."+vcs.cmd) {
  1070  			return fmt.Errorf("invalid version control suffix in %s path", match["prefix"])
  1071  		}
  1072  	}
  1073  	return nil
  1074  }
  1075  
  1076  // bitbucketVCS determines the version control system for a
  1077  // Bitbucket repository, by using the Bitbucket API.
  1078  func bitbucketVCS(match map[string]string) error {
  1079  	if err := noVCSSuffix(match); err != nil {
  1080  		return err
  1081  	}
  1082  
  1083  	var resp struct {
  1084  		SCM string `json:"scm"`
  1085  	}
  1086  	url := expand(match, "https://api.bitbucket.org/2.0/repositories/{bitname}?fields=scm")
  1087  	data, err := web.Get(url)
  1088  	if err != nil {
  1089  		if httpErr, ok := err.(*web.HTTPError); ok && httpErr.StatusCode == 403 {
  1090  			// this may be a private repository. If so, attempt to determine which
  1091  			// VCS it uses. See issue 5375.
  1092  			root := match["root"]
  1093  			for _, vcs := range []string{"git", "hg"} {
  1094  				if vcsByCmd(vcs).ping("https", root) == nil {
  1095  					resp.SCM = vcs
  1096  					break
  1097  				}
  1098  			}
  1099  		}
  1100  
  1101  		if resp.SCM == "" {
  1102  			return err
  1103  		}
  1104  	} else {
  1105  		if err := json.Unmarshal(data, &resp); err != nil {
  1106  			return fmt.Errorf("decoding %s: %v", url, err)
  1107  		}
  1108  	}
  1109  
  1110  	if vcsByCmd(resp.SCM) != nil {
  1111  		match["vcs"] = resp.SCM
  1112  		if resp.SCM == "git" {
  1113  			match["repo"] += ".git"
  1114  		}
  1115  		return nil
  1116  	}
  1117  
  1118  	return fmt.Errorf("unable to detect version control system for bitbucket.org/ path")
  1119  }
  1120  
  1121  // launchpadVCS solves the ambiguity for "lp.net/project/foo". In this case,
  1122  // "foo" could be a series name registered in Launchpad with its own branch,
  1123  // and it could also be the name of a directory within the main project
  1124  // branch one level up.
  1125  func launchpadVCS(match map[string]string) error {
  1126  	if match["project"] == "" || match["series"] == "" {
  1127  		return nil
  1128  	}
  1129  	_, err := web.Get(expand(match, "https://code.launchpad.net/{project}{series}/.bzr/branch-format"))
  1130  	if err != nil {
  1131  		match["root"] = expand(match, "launchpad.net/{project}")
  1132  		match["repo"] = expand(match, "https://{root}")
  1133  	}
  1134  	return nil
  1135  }
  1136  

View as plain text