Represent git-submodule as nested projects, take 2

(Previous submission of this change broke Android buildbot due to
 incorrect regular expression for parsing git-config output.  During
 investigation, we also found that Android, which pulls Chromium, has a
 workaround for Chromium's submodules; its manifest includes Chromium's
 submodules.  This new change, in addition to fixing the regex, also
 take this type of workarounds into consideration; it adds a new
 attribute that makes repo not fetch submodules unless submodules have a
 project element defined in the manifest, or this attribute is
 overridden by a parent project element or by the default element.)

We need a representation of git-submodule in repo; otherwise repo will
not sync submodules, and leave workspace in a broken state.  Of course
this will not be a problem if all projects are owned by the owner of the
manifest file, who may simply choose not to use git-submodule in all
projects.  However, this is not possible in practice because manifest
file owner is unlikely to own all upstream projects.

As git submodules are simply git repositories, it is natural to treat
them as plain repo projects that live inside a repo project.  That is,
we could use recursively declared projects to denote the is-submodule
relation of git repositories.

The behavior of repo remains the same to projects that do not have a
sub-project within.  As for parent projects, repo fetches them and their
sub-projects as normal projects, and then checks out subprojects at the
commit specified in parent's commit object.  The sub-project is fetched
at a path relative to parent project's working directory; so the path
specified in manifest file should match that of .gitmodules file.

If a submodule is not registered in repo manifest, repo will derive its
properties from itself and its parent project, which might not always be
correct.  In such cases, the subproject is called a derived subproject.

To a user, a sub-project is merely a git-submodule; so all tips of
working with a git-submodule apply here, too.  For example, you should
not run `repo sync` in a parent repository if its submodule is dirty.

Change-Id: I4b8344c1b9ccad2f58ad304573133e5d52e1faef

command.py

@@ -100,7 +100,33 @@ class Command(object):
     """
     raise NotImplementedError
 
-  def GetProjects(self, args, missing_ok=False):
+  def _ResetPathToProjectMap(self, projects):
+    self._by_path = dict((p.worktree, p) for p in projects)
+
+  def _UpdatePathToProjectMap(self, project):
+    self._by_path[project.worktree] = project
+
+  def _GetProjectByPath(self, path):
+    project = None
+    if os.path.exists(path):
+      oldpath = None
+      while path \
+        and path != oldpath \
+        and path != self.manifest.topdir:
+        try:
+          project = self._by_path[path]
+          break
+        except KeyError:
+          oldpath = path
+          path = os.path.dirname(path)
+    else:
+      try:
+        project = self._by_path[path]
+      except KeyError:
+        pass
+    return project
+
+  def GetProjects(self, args, missing_ok=False, submodules_ok=False):
     """A list of projects that match the arguments.
     """
     all_projects = self.manifest.projects
@@ -114,40 +140,37 @@ class Command(object):
     groups = [x for x in re.split(r'[,\s]+', groups) if x]
 
     if not args:
-      for project in all_projects.values():
+      all_projects_list = all_projects.values()
+      derived_projects = {}
+      for project in all_projects_list:
+        if submodules_ok or project.sync_s:
+          derived_projects.update((p.name, p)
+                                  for p in project.GetDerivedSubprojects())
+      all_projects_list.extend(derived_projects.values())
+      for project in all_projects_list:
         if ((missing_ok or project.Exists) and
             project.MatchesGroups(groups)):
           result.append(project)
     else:
-      by_path = None
+      self._ResetPathToProjectMap(all_projects.values())
 
       for arg in args:
         project = all_projects.get(arg)
 
         if not project:
           path = os.path.abspath(arg).replace('\\', '/')
+          project = self._GetProjectByPath(path)
 
-          if not by_path:
+          # If it's not a derived project, update path->project mapping and
-            by_path = dict()
+          # search again, as arg might actually point to a derived subproject.
-            for p in all_projects.values():
+          if (project and not project.Derived and
-              by_path[p.worktree] = p
+              (submodules_ok or project.sync_s)):
-
+            search_again = False
-          if os.path.exists(path):
+            for subproject in project.GetDerivedSubprojects():
-            oldpath = None
+              self._UpdatePathToProjectMap(subproject)
-            while path \
+              search_again = True
-              and path != oldpath \
+            if search_again:
-              and path != self.manifest.topdir:
+              project = self._GetProjectByPath(path) or project
-              try:
-                project = by_path[path]
-                break
-              except KeyError:
-                oldpath = path
-                path = os.path.dirname(path)
-          else:
-            try:
-              project = by_path[path]
-            except KeyError:
-              pass
 
         if not project:
           raise NoSuchProjectError(arg)

docs/manifest-format.txt

@@ -41,17 +41,20 @@ following DTD:
     <!ATTLIST default revision CDATA #IMPLIED>
     <!ATTLIST default sync-j   CDATA #IMPLIED>
     <!ATTLIST default sync-c   CDATA #IMPLIED>
+    <!ATTLIST default sync-s   CDATA #IMPLIED>
 
     <!ELEMENT manifest-server (EMPTY)>
     <!ATTLIST url              CDATA #REQUIRED>
   
-    <!ELEMENT project (annotation?)>
+    <!ELEMENT project (annotation?,
+                       project*)>
     <!ATTLIST project name     CDATA #REQUIRED>
     <!ATTLIST project path     CDATA #IMPLIED>
     <!ATTLIST project remote   IDREF #IMPLIED>
     <!ATTLIST project revision CDATA #IMPLIED>
     <!ATTLIST project groups   CDATA #IMPLIED>
     <!ATTLIST project sync-c   CDATA #IMPLIED>
+    <!ATTLIST project sync-s   CDATA #IMPLIED>
 
     <!ELEMENT annotation (EMPTY)>
     <!ATTLIST annotation name  CDATA #REQUIRED>
@@ -152,7 +155,10 @@ Element project
 
 One or more project elements may be specified.  Each element
 describes a single Git repository to be cloned into the repo
-client workspace.
+client workspace.  You may specify Git-submodules by creating a
+nested project.  Git-submodules will be automatically
+recognized and inherit their parent's attributes, but those
+may be overridden by an explicitly specified project element.
 
 Attribute `name`: A unique name for this project.  The project's
 name is appended onto its remote's fetch URL to generate the actual
@@ -163,7 +169,8 @@ URL to configure the Git remote with.  The URL gets formed as:
 where ${remote_fetch} is the remote's fetch attribute and
 ${project_name} is the project's name attribute.  The suffix ".git"
 is always appended as repo assumes the upstream is a forest of
-bare Git repositories.
+bare Git repositories.  If the project has a parent element, its
+name will be prefixed by the parent's.
 
 The project name must match the name Gerrit knows, if Gerrit is
 being used for code reviews.
@@ -171,6 +178,8 @@ being used for code reviews.
 Attribute `path`: An optional path relative to the top directory
 of the repo client where the Git working directory for this project
 should be placed.  If not supplied the project name is used.
+If the project has a parent element, its path will be prefixed
+by the parent's.
 
 Attribute `remote`: Name of a previously defined remote element.
 If not supplied the remote given by the default element is used.
@@ -190,6 +199,8 @@ its name:`name` and path:`path`.  E.g. for
 definition is implicitly in the following manifest groups:
 default, name:monkeys, and path:barrel-of.  If you place a project in the
 group "notdefault", it will not be automatically downloaded by repo.
+If the project has a parent element, the `name` and `path` here
+are the prefixed ones.
 
 Element annotation
 ------------------

manifest_xml.py

@@ -40,6 +40,7 @@ class _Default(object):
   remote = None
   sync_j = 1
   sync_c = False
+  sync_s = False
 
 class _XmlRemote(object):
   def __init__(self,
@@ -178,6 +179,9 @@ class XmlManifest(object):
     if d.sync_c:
       have_default = True
       e.setAttribute('sync-c', 'true')
+    if d.sync_s:
+      have_default = True
+      e.setAttribute('sync-s', 'true')
     if have_default:
       root.appendChild(e)
       root.appendChild(doc.createTextNode(''))
@@ -188,20 +192,25 @@ class XmlManifest(object):
       root.appendChild(e)
       root.appendChild(doc.createTextNode(''))
 
-    sort_projects = list(self.projects.keys())
+    def output_projects(parent, parent_node, projects):
-    sort_projects.sort()
+      for p in projects:
-
+        output_project(parent, parent_node, self.projects[p])
-    for p in sort_projects:
-      p = self.projects[p]
 
+    def output_project(parent, parent_node, p):
       if not p.MatchesGroups(groups):
-        continue
+        return
+
+      name = p.name
+      relpath = p.relpath
+      if parent:
+        name = self._UnjoinName(parent.name, name)
+        relpath = self._UnjoinRelpath(parent.relpath, relpath)
 
       e = doc.createElement('project')
-      root.appendChild(e)
+      parent_node.appendChild(e)
-      e.setAttribute('name', p.name)
+      e.setAttribute('name', name)
-      if p.relpath != p.name:
+      if relpath != name:
-        e.setAttribute('path', p.relpath)
+        e.setAttribute('path', relpath)
       if not d.remote or p.remote.name != d.remote.name:
         e.setAttribute('remote', p.remote.name)
       if peg_rev:
@@ -239,6 +248,19 @@ class XmlManifest(object):
       if p.sync_c:
         e.setAttribute('sync-c', 'true')
 
+      if p.sync_s:
+        e.setAttribute('sync-s', 'true')
+
+      if p.subprojects:
+        sort_projects = [subp.name for subp in p.subprojects]
+        sort_projects.sort()
+        output_projects(p, e, sort_projects)
+
+    sort_projects = [key for key in self.projects.keys()
+                     if not self.projects[key].parent]
+    sort_projects.sort()
+    output_projects(None, root, sort_projects)
+
     if self._repo_hooks_project:
       root.appendChild(doc.createTextNode(''))
       e = doc.createElement('repo-hooks')
@@ -409,14 +431,19 @@ class XmlManifest(object):
               (self.manifestFile))
         self._manifest_server = url
 
-    for node in itertools.chain(*node_list):
+    def recursively_add_projects(project):
-      if node.nodeName == 'project':
-        project = self._ParseProject(node)
       if self._projects.get(project.name):
         raise ManifestParseError(
             'duplicate project %s in %s' %
             (project.name, self.manifestFile))
       self._projects[project.name] = project
+      for subproject in project.subprojects:
+        recursively_add_projects(subproject)
+
+    for node in itertools.chain(*node_list):
+      if node.nodeName == 'project':
+        project = self._ParseProject(node)
+        recursively_add_projects(project)
       if node.nodeName == 'repo-hooks':
         # Get the name of the project and the (space-separated) list of enabled.
         repo_hooks_project = self._reqatt(node, 'in-project')
@@ -524,6 +551,12 @@ class XmlManifest(object):
       d.sync_c = False
     else:
       d.sync_c = sync_c.lower() in ("yes", "true", "1")
+
+    sync_s = node.getAttribute('sync-s')
+    if not sync_s:
+      d.sync_s = False
+    else:
+      d.sync_s = sync_s.lower() in ("yes", "true", "1")
     return d
 
   def _ParseNotice(self, node):
@@ -565,11 +598,19 @@ class XmlManifest(object):
 
     return '\n'.join(cleanLines)
 
-  def _ParseProject(self, node):
+  def _JoinName(self, parent_name, name):
+    return os.path.join(parent_name, name)
+
+  def _UnjoinName(self, parent_name, name):
+    return os.path.relpath(name, parent_name)
+
+  def _ParseProject(self, node, parent = None):
     """
     reads a <project> element from the manifest file
     """
     name = self._reqatt(node, 'name')
+    if parent:
+      name = self._JoinName(parent.name, name)
 
     remote = self._get_remote(node)
     if remote is None:
@@ -607,6 +648,12 @@ class XmlManifest(object):
     else:
       sync_c = sync_c.lower() in ("yes", "true", "1")
 
+    sync_s = node.getAttribute('sync-s')
+    if not sync_s:
+      sync_s = self._default.sync_s
+    else:
+      sync_s = sync_s.lower() in ("yes", "true", "1")
+
     upstream = node.getAttribute('upstream')
 
     groups = ''
@@ -614,37 +661,67 @@ class XmlManifest(object):
       groups = node.getAttribute('groups')
     groups = [x for x in re.split(r'[,\s]+', groups) if x]
 
-    default_groups = ['all', 'name:%s' % name, 'path:%s' % path]
+    if parent is None:
-    groups.extend(set(default_groups).difference(groups))
+      relpath, worktree, gitdir = self.GetProjectPaths(name, path)
-
-    if self.IsMirror:
-      worktree = None
-      gitdir = os.path.join(self.topdir, '%s.git' % name)
     else:
-      worktree = os.path.join(self.topdir, path).replace('\\', '/')
+      relpath, worktree, gitdir = self.GetSubprojectPaths(parent, path)
-      gitdir = os.path.join(self.repodir, 'projects/%s.git' % path)
+
+    default_groups = ['all', 'name:%s' % name, 'path:%s' % relpath]
+    groups.extend(set(default_groups).difference(groups))
 
     project = Project(manifest = self,
                       name = name,
                       remote = remote.ToRemoteSpec(name),
                       gitdir = gitdir,
                       worktree = worktree,
-                      relpath = path,
+                      relpath = relpath,
                       revisionExpr = revisionExpr,
                       revisionId = None,
                       rebase = rebase,
                       groups = groups,
                       sync_c = sync_c,
-                      upstream = upstream)
+                      sync_s = sync_s,
+                      upstream = upstream,
+                      parent = parent)
 
     for n in node.childNodes:
       if n.nodeName == 'copyfile':
         self._ParseCopyFile(project, n)
       if n.nodeName == 'annotation':
         self._ParseAnnotation(project, n)
+      if n.nodeName == 'project':
+        project.subprojects.append(self._ParseProject(n, parent = project))
 
     return project
 
+  def GetProjectPaths(self, name, path):
+    relpath = path
+    if self.IsMirror:
+      worktree = None
+      gitdir = os.path.join(self.topdir, '%s.git' % name)
+    else:
+      worktree = os.path.join(self.topdir, path).replace('\\', '/')
+      gitdir = os.path.join(self.repodir, 'projects', '%s.git' % path)
+    return relpath, worktree, gitdir
+
+  def GetSubprojectName(self, parent, submodule_path):
+    return os.path.join(parent.name, submodule_path)
+
+  def _JoinRelpath(self, parent_relpath, relpath):
+    return os.path.join(parent_relpath, relpath)
+
+  def _UnjoinRelpath(self, parent_relpath, relpath):
+    return os.path.relpath(relpath, parent_relpath)
+
+  def GetSubprojectPaths(self, parent, path):
+    relpath = self._JoinRelpath(parent.relpath, path)
+    gitdir = os.path.join(parent.gitdir, 'subprojects', '%s.git' % path)
+    if self.IsMirror:
+      worktree = None
+    else:
+      worktree = os.path.join(parent.worktree, path).replace('\\', '/')
+    return relpath, worktree, gitdir
+
   def _ParseCopyFile(self, project, node):
     src = self._reqatt(node, 'src')
     dest = self._reqatt(node, 'dest')

project.py

@@ -23,6 +23,7 @@ import shutil
 import stat
 import subprocess
 import sys
+import tempfile
 import time
 
 from color import Coloring
@@ -486,7 +487,30 @@ class Project(object):
                rebase = True,
                groups = None,
                sync_c = False,
-               upstream = None):
+               sync_s = False,
+               upstream = None,
+               parent = None,
+               is_derived = False):
+    """Init a Project object.
+
+    Args:
+      manifest: The XmlManifest object.
+      name: The `name` attribute of manifest.xml's project element.
+      remote: RemoteSpec object specifying its remote's properties.
+      gitdir: Absolute path of git directory.
+      worktree: Absolute path of git working tree.
+      relpath: Relative path of git working tree to repo's top directory.
+      revisionExpr: The `revision` attribute of manifest.xml's project element.
+      revisionId: git commit id for checking out.
+      rebase: The `rebase` attribute of manifest.xml's project element.
+      groups: The `groups` attribute of manifest.xml's project element.
+      sync_c: The `sync-c` attribute of manifest.xml's project element.
+      sync_s: The `sync-s` attribute of manifest.xml's project element.
+      upstream: The `upstream` attribute of manifest.xml's project element.
+      parent: The parent Project object.
+      is_derived: False if the project was explicitly defined in the manifest;
+                  True if the project is a discovered submodule.
+    """
     self.manifest = manifest
     self.name = name
     self.remote = remote
@@ -508,7 +532,11 @@ class Project(object):
     self.rebase = rebase
     self.groups = groups
     self.sync_c = sync_c
+    self.sync_s = sync_s
     self.upstream = upstream
+    self.parent = parent
+    self.is_derived = is_derived
+    self.subprojects = []
 
     self.snapshots = {}
     self.copyfiles = []
@@ -528,6 +556,10 @@ class Project(object):
     # project containing repo hooks.
     self.enabled_repo_hooks = []
 
+  @property
+  def Derived(self):
+    return self.is_derived
+
   @property
   def Exists(self):
     return os.path.isdir(self.gitdir)
@@ -1370,6 +1402,149 @@ class Project(object):
     return kept
 
 
+## Submodule Management ##
+
+  def GetRegisteredSubprojects(self):
+    result = []
+    def rec(subprojects):
+      if not subprojects:
+        return
+      result.extend(subprojects)
+      for p in subprojects:
+        rec(p.subprojects)
+    rec(self.subprojects)
+    return result
+
+  def _GetSubmodules(self):
+    # Unfortunately we cannot call `git submodule status --recursive` here
+    # because the working tree might not exist yet, and it cannot be used
+    # without a working tree in its current implementation.
+
+    def get_submodules(gitdir, rev):
+      # Parse .gitmodules for submodule sub_paths and sub_urls
+      sub_paths, sub_urls = parse_gitmodules(gitdir, rev)
+      if not sub_paths:
+        return []
+      # Run `git ls-tree` to read SHAs of submodule object, which happen to be
+      # revision of submodule repository
+      sub_revs = git_ls_tree(gitdir, rev, sub_paths)
+      submodules = []
+      for sub_path, sub_url in zip(sub_paths, sub_urls):
+        try:
+          sub_rev = sub_revs[sub_path]
+        except KeyError:
+          # Ignore non-exist submodules
+          continue
+        submodules.append((sub_rev, sub_path, sub_url))
+      return submodules
+
+    re_path = re.compile(r'^submodule\.([^.]+)\.path=(.*)$')
+    re_url = re.compile(r'^submodule\.([^.]+)\.url=(.*)$')
+    def parse_gitmodules(gitdir, rev):
+      cmd = ['cat-file', 'blob', '%s:.gitmodules' % rev]
+      try:
+        p = GitCommand(None, cmd, capture_stdout = True, capture_stderr = True,
+                       bare = True, gitdir = gitdir)
+      except GitError:
+        return [], []
+      if p.Wait() != 0:
+        return [], []
+
+      gitmodules_lines = []
+      fd, temp_gitmodules_path = tempfile.mkstemp()
+      try:
+        os.write(fd, p.stdout)
+        os.close(fd)
+        cmd = ['config', '--file', temp_gitmodules_path, '--list']
+        p = GitCommand(None, cmd, capture_stdout = True, capture_stderr = True,
+                       bare = True, gitdir = gitdir)
+        if p.Wait() != 0:
+          return [], []
+        gitmodules_lines = p.stdout.split('\n')
+      except GitError:
+        return [], []
+      finally:
+        os.remove(temp_gitmodules_path)
+
+      names = set()
+      paths = {}
+      urls = {}
+      for line in gitmodules_lines:
+        if not line:
+          continue
+        m = re_path.match(line)
+        if m:
+          names.add(m.group(1))
+          paths[m.group(1)] = m.group(2)
+          continue
+        m = re_url.match(line)
+        if m:
+          names.add(m.group(1))
+          urls[m.group(1)] = m.group(2)
+          continue
+      names = sorted(names)
+      return ([paths.get(name, '') for name in names],
+              [urls.get(name, '') for name in names])
+
+    def git_ls_tree(gitdir, rev, paths):
+      cmd = ['ls-tree', rev, '--']
+      cmd.extend(paths)
+      try:
+        p = GitCommand(None, cmd, capture_stdout = True, capture_stderr = True,
+                       bare = True, gitdir = gitdir)
+      except GitError:
+        return []
+      if p.Wait() != 0:
+        return []
+      objects = {}
+      for line in p.stdout.split('\n'):
+        if not line.strip():
+          continue
+        object_rev, object_path = line.split()[2:4]
+        objects[object_path] = object_rev
+      return objects
+
+    try:
+      rev = self.GetRevisionId()
+    except GitError:
+      return []
+    return get_submodules(self.gitdir, rev)
+
+  def GetDerivedSubprojects(self):
+    result = []
+    if not self.Exists:
+      # If git repo does not exist yet, querying its submodules will
+      # mess up its states; so return here.
+      return result
+    for rev, path, url in self._GetSubmodules():
+      name = self.manifest.GetSubprojectName(self, path)
+      project = self.manifest.projects.get(name)
+      if project:
+        result.extend(project.GetDerivedSubprojects())
+        continue
+      relpath, worktree, gitdir = self.manifest.GetSubprojectPaths(self, path)
+      remote = RemoteSpec(self.remote.name,
+                          url = url,
+                          review = self.remote.review)
+      subproject = Project(manifest = self.manifest,
+                           name = name,
+                           remote = remote,
+                           gitdir = gitdir,
+                           worktree = worktree,
+                           relpath = relpath,
+                           revisionExpr = self.revisionExpr,
+                           revisionId = rev,
+                           rebase = self.rebase,
+                           groups = self.groups,
+                           sync_c = self.sync_c,
+                           sync_s = self.sync_s,
+                           parent = self,
+                           is_derived = True)
+      result.append(subproject)
+      result.extend(subproject.GetDerivedSubprojects())
+    return result
+
+
 ## Direct Git Commands ##
 
   def _RemoteFetch(self, name=None,

subcmds/sync.py

@@ -114,6 +114,9 @@ resumeable bundle file on a content delivery network. This
 may be necessary if there are problems with the local Python
 HTTP client or proxy configuration, but the Git binary works.
 
+The --fetch-submodules option enables fetching Git submodules
+of a project from server.
+
 SSH Connections
 ---------------
 
@@ -180,6 +183,9 @@ later is required to fix a server side protocol bug.
     p.add_option('-p', '--manifest-server-password', action='store',
                  dest='manifest_server_password',
                  help='password to authenticate with the manifest server')
+    p.add_option('--fetch-submodules',
+                 dest='fetch_submodules', action='store_true',
+                 help='fetch submodules from server')
     if show_smart:
       p.add_option('-s', '--smart-sync',
                    dest='smart_sync', action='store_true',
@@ -559,7 +565,9 @@ later is required to fix a server side protocol bug.
       self.manifest._Unload()
       if opt.jobs is None:
         self.jobs = self.manifest.default.sync_j
-    all_projects = self.GetProjects(args, missing_ok=True)
+    all_projects = self.GetProjects(args,
+                                    missing_ok=True,
+                                    submodules_ok=opt.fetch_submodules)
 
     self._fetch_times = _FetchTimes(self.manifest)
     if not opt.local_only:
@@ -570,12 +578,33 @@ later is required to fix a server side protocol bug.
       to_fetch.extend(all_projects)
       to_fetch.sort(key=self._fetch_times.Get, reverse=True)
 
-      self._Fetch(to_fetch, opt)
+      fetched = self._Fetch(to_fetch, opt)
       _PostRepoFetch(rp, opt.no_repo_verify)
       if opt.network_only:
         # bail out now; the rest touches the working tree
         return
 
+      # Iteratively fetch missing and/or nested unregistered submodules
+      previously_missing_set = set()
+      while True:
+        self.manifest._Unload()
+        all_projects = self.GetProjects(args,
+                                        missing_ok=True,
+                                        submodules_ok=opt.fetch_submodules)
+        missing = []
+        for project in all_projects:
+          if project.gitdir not in fetched:
+            missing.append(project)
+        if not missing:
+          break
+        # Stop us from non-stopped fetching actually-missing repos: If set of
+        # missing repos has not been changed from last fetch, we break.
+        missing_set = set(p.name for p in missing)
+        if previously_missing_set == missing_set:
+          break
+        previously_missing_set = missing_set
+        fetched.update(self._Fetch(missing, opt))
+
     if self.manifest.IsMirror:
       # bail out now, we have no working tree
       return