sync: Introduce git checkout levels

If a repo manifest is updated so that project B is placed within a
project A, and if project A had content in new B's location in the old
checkout, then repo sync could break depending on checkout order, since
B can't be checked out before A.

This change introduces checkout levels which enforces right sequence of
checkouts while still allowing for parallel checkout. In an example
above, A will always be checked out first before B.

BUG=b:325119758
TEST=./run_tests, manual sync on ChromeOS repository

Change-Id: Ib3b5e4d2639ca56620a1e4c6bf76d7b1ab805250
Reviewed-on: https://gerrit-review.googlesource.com/c/git-repo/+/410421
Tested-by: Josip Sokcevic <sokcevic@google.com>
Reviewed-by: Greg Edelston <gredelston@google.com>
Commit-Queue: Josip Sokcevic <sokcevic@google.com>
Reviewed-by: Gavin Mak <gavinmak@google.com>

git_command.py

@@ -135,6 +135,8 @@ def GetEventTargetPath():
     if retval == 0:
         # Strip trailing carriage-return in path.
         path = p.stdout.rstrip("\n")
+        if path == "":
+            return None
     elif retval != 1:
         # `git config --get` is documented to produce an exit status of `1`
         # if the requested variable is not present in the configuration.

main.py

@@ -270,10 +270,14 @@ class _Repo:
             self._PrintHelp(short=True)
             return 1
 
-        run = lambda: self._RunLong(name, gopts, argv) or 0
+        git_trace2_event_log = EventLog()
+        run = (
+            lambda: self._RunLong(name, gopts, argv, git_trace2_event_log) or 0
+        )
         with Trace(
-            "starting new command: %s",
+            "starting new command: %s [sid=%s]",
             ", ".join([name] + argv),
+            git_trace2_event_log.full_sid,
             first_trace=True,
         ):
             if gopts.trace_python:
@@ -290,12 +294,11 @@ class _Repo:
                 result = run()
         return result
 
-    def _RunLong(self, name, gopts, argv):
+    def _RunLong(self, name, gopts, argv, git_trace2_event_log):
         """Execute the (longer running) requested subcommand."""
         result = 0
         SetDefaultColoring(gopts.color)
 
-        git_trace2_event_log = EventLog()
         outer_client = RepoClient(self.repodir)
         repo_client = outer_client
         if gopts.submanifest_path:

repo

@@ -1,4 +1,4 @@
-#!/usr/bin/env python
+#!/usr/bin/env python3
 #
 # Copyright (C) 2008 The Android Open Source Project
 #
@@ -124,7 +124,7 @@ if not REPO_REV:
 BUG_URL = "https://issues.gerritcodereview.com/issues/new?component=1370071"
 
 # increment this whenever we make important changes to this script
-VERSION = (2, 40)
+VERSION = (2, 42)
 
 # increment this if the MAINTAINER_KEYS block is modified
 KEYRING_VERSION = (2, 3)

subcmds/sync.py

@@ -21,6 +21,7 @@ import multiprocessing
 import netrc
 import optparse
 import os
+from pathlib import Path
 import sys
 import tempfile
 import time
@@ -82,16 +83,50 @@ from wrapper import Wrapper
 
 _ONE_DAY_S = 24 * 60 * 60
 
-# Env var to implicitly turn auto-gc back on.  This was added to allow a user to
-# revert a change in default behavior in v2.29.9.  Remove after 2023-04-01.
-_REPO_AUTO_GC = "REPO_AUTO_GC"
-_AUTO_GC = os.environ.get(_REPO_AUTO_GC) == "1"
-
 _REPO_ALLOW_SHALLOW = os.environ.get("REPO_ALLOW_SHALLOW")
 
 logger = RepoLogger(__file__)
 
 
+def _SafeCheckoutOrder(checkouts: List[Project]) -> List[List[Project]]:
+    """Generate a sequence of checkouts that is safe to perform. The client
+    should checkout everything from n-th index before moving to n+1.
+
+    This is only useful if manifest contains nested projects.
+
+    E.g. if foo, foo/bar and foo/bar/baz are project paths, then foo needs to
+    finish before foo/bar can proceed, and foo/bar needs to finish before
+    foo/bar/baz."""
+    res = [[]]
+    current = res[0]
+
+    # depth_stack contains a current stack of parent paths.
+    depth_stack = []
+    # checkouts are iterated in asc order by relpath. That way, it can easily be
+    # determined if the previous checkout is parent of the current checkout.
+    for checkout in sorted(checkouts, key=lambda x: x.relpath):
+        checkout_path = Path(checkout.relpath)
+        while depth_stack:
+            try:
+                checkout_path.relative_to(depth_stack[-1])
+            except ValueError:
+                # Path.relative_to returns ValueError if paths are not relative.
+                # TODO(sokcevic): Switch to is_relative_to once min supported
+                # version is py3.9.
+                depth_stack.pop()
+            else:
+                if len(depth_stack) >= len(res):
+                    # Another depth created.
+                    res.append([])
+                break
+
+        current = res[len(depth_stack)]
+        current.append(checkout)
+        depth_stack.append(checkout_path)
+
+    return res
+
+
 class _FetchOneResult(NamedTuple):
     """_FetchOne return value.
 
@@ -1040,15 +1075,21 @@ later is required to fix a server side protocol bug.
                 pm.update(msg=project.name)
             return ret
 
-        proc_res = self.ExecuteInParallel(
-            opt.jobs_checkout,
-            functools.partial(
-                self._CheckoutOne, opt.detach_head, opt.force_sync, opt.verbose
-            ),
-            all_projects,
-            callback=_ProcessResults,
-            output=Progress("Checking out", len(all_projects), quiet=opt.quiet),
-        )
+        for projects in _SafeCheckoutOrder(all_projects):
+            proc_res = self.ExecuteInParallel(
+                opt.jobs_checkout,
+                functools.partial(
+                    self._CheckoutOne,
+                    opt.detach_head,
+                    opt.force_sync,
+                    opt.verbose,
+                ),
+                projects,
+                callback=_ProcessResults,
+                output=Progress(
+                    "Checking out", len(all_projects), quiet=opt.quiet
+                ),
+            )
 
         self._local_sync_state.Save()
         return proc_res and not err_results
@@ -1544,9 +1585,7 @@ later is required to fix a server side protocol bug.
                 mp, event_log.TASK_SYNC_LOCAL, start, time.time(), clean
             )
             if not clean:
-                raise UpdateManifestError(
-                    aggregate_errors=errors, project=mp.name
-                )
+                raise UpdateManifestError(aggregate_errors=errors)
             self._ReloadManifest(manifest_name, mp.manifest)
 
     def ValidateOptions(self, opt, args):
@@ -1577,16 +1616,6 @@ later is required to fix a server side protocol bug.
         if opt.prune is None:
             opt.prune = True
 
-        if opt.auto_gc is None and _AUTO_GC:
-            logger.error(
-                "Will run `git gc --auto` because %s is set. %s is deprecated "
-                "and will be removed in a future release.  Use `--auto-gc` "
-                "instead.",
-                _REPO_AUTO_GC,
-                _REPO_AUTO_GC,
-            )
-            opt.auto_gc = True
-
     def _ValidateOptionsWithManifest(self, opt, mp):
         """Like ValidateOptions, but after we've updated the manifest.
 
@@ -1630,7 +1659,7 @@ later is required to fix a server side protocol bug.
         errors = []
         try:
             self._ExecuteHelper(opt, args, errors)
-        except RepoExitError:
+        except (RepoExitError, RepoChangedException):
             raise
         except (KeyboardInterrupt, Exception) as e:
             raise RepoUnhandledExceptionError(e, aggregate_errors=errors)

tests/test_subcmds_sync.py

@@ -304,6 +304,32 @@ class LocalSyncState(unittest.TestCase):
         self.assertEqual(self.state.GetFetchTime(projA), 5)
 
 
+class SafeCheckoutOrder(unittest.TestCase):
+    def test_no_nested(self):
+        p_f = mock.MagicMock(relpath="f")
+        p_foo = mock.MagicMock(relpath="foo")
+        out = sync._SafeCheckoutOrder([p_f, p_foo])
+        self.assertEqual(out, [[p_f, p_foo]])
+
+    def test_basic_nested(self):
+        p_foo = p_foo = mock.MagicMock(relpath="foo")
+        p_foo_bar = mock.MagicMock(relpath="foo/bar")
+        out = sync._SafeCheckoutOrder([p_foo, p_foo_bar])
+        self.assertEqual(out, [[p_foo], [p_foo_bar]])
+
+    def test_complex_nested(self):
+        p_foo = mock.MagicMock(relpath="foo")
+        p_foo_bar = mock.MagicMock(relpath="foo/bar")
+        p_foo_bar_baz_baq = mock.MagicMock(relpath="foo/bar/baz/baq")
+        p_bar = mock.MagicMock(relpath="bar")
+        out = sync._SafeCheckoutOrder(
+            [p_foo_bar_baz_baq, p_foo, p_foo_bar, p_bar]
+        )
+        self.assertEqual(
+            out, [[p_bar, p_foo], [p_foo_bar], [p_foo_bar_baz_baq]]
+        )
+
+
 class GetPreciousObjectsState(unittest.TestCase):
     """Tests for _GetPreciousObjectsState."""