Add --no-history option to fetch and gclient for shallow clones.

gclient_scm.py

Index: gclient_scm.py
diff --git a/gclient_scm.py b/gclient_scm.py
index a87032e749902122c676b6621dde4fec33f8f232..df1054f112233032e21cb45a21a50d48af8dbc74 100644
--- a/gclient_scm.py
+++ b/gclient_scm.py
@@ -825,6 +825,25 @@ class GitWrapper(SCMWrapper):
     # create it, so we need to do it manually.
     parent_dir = os.path.dirname(self.checkout_path)
     gclient_utils.safe_makedirs(parent_dir)
+
+    template_dir = None
+    if options.no_history:
+      if gclient_utils.IsGitSha(revision):
+        # In the case of a subproject, the pinned sha is not necessarily the
+        # head of the remote branch (so we can't just use --depth=N). Instead,
+        # we tell git to fetch all the remote objects from SHA..HEAD by means of
+        # a template git dir which has a 'shallow' file pointing to the sha.

[iannucci, 2014/08/02 21:22:15]

sneaky :(. I'm not sure if this is guaranteed to work though... wouldn't git
claim that it already has <SHA> to the remote server?

more specifically, wouldn't this work the same way that `git fetch <sha>` works
(that is: sporadically supported and discouraged?). googlesource.com (AFAIK)
only supports fetching <sha>'s which were (recently) the HEAD of some ref, for
example.

[Primiano Tucci (use gerrit), 2014/08/03 17:43:34]

+torne as he might have more insights on this topic.

On 2014/08/02 21:22:15, iannucci wrote:
Yes, I think this is precisely the way it works.
However, this is different than "git fetch SHA". IIRC "git fetch SHA" is a GoB
invention as is not part of the git wire protocol at all.
The mechanism that shallow uses (--depth/shallow is officially part of git-scm
docs even before 1.9), instead, should just leverage the standard protocol
adopted by fetch-pack.

+        template_dir = tempfile.mkdtemp(
+            prefix='_gclient_gittmp_%s' % os.path.basename(self.checkout_path),
+            dir=parent_dir)
+        self._Run(['init', '--bare', template_dir], options, cwd=self._root_dir)
+        with open(os.path.join(template_dir, 'shallow'), 'w') as template_file:
+          template_file.write(revision)
+        clone_cmd.append('--template=' + template_dir)
+      else:
+        # Otherwise, we're just interested in the HEAD. Just use --depth.
+        clone_cmd.append('--depth=1')
+
     tmp_dir = tempfile.mkdtemp(
         prefix='_gclient_%s_' % os.path.basename(self.checkout_path),
         dir=parent_dir)
@@ -841,6 +860,8 @@ class GitWrapper(SCMWrapper):
       if os.listdir(tmp_dir):
         self.Print('_____ removing non-empty tmp dir %s' % tmp_dir)
       gclient_utils.rmtree(tmp_dir)
+      if template_dir:
+        gclient_utils.rmtree(template_dir)

[wtc, 2014/08/03 04:12:49]

Should we add
  if os.listdir(template_dir):
    self.Print('_____ removing non-empty template dir %s' % template_dir)
?

See lines 860-861.

[Primiano Tucci (use gerrit), 2014/08/03 17:43:34]

Hmm this I think should stay as it is. 
Conversely to tmp_dir, template_dir can be None if the user never specified
--no-history.
On the other side if it is not-None, it necessarily means that mkdtemp has
created a directory and we want to clean that up anyways, empty or not, no?
Agree here instead.: added the print.

     self._UpdateBranchHeads(options, fetch=True)
     self._Checkout(options, revision.replace('refs/heads/', ''), quiet=True)
     if self._GetCurrentBranch() is None: