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

View as plain text