filter-repo: rewrite the stash too

While filter-repo by default will abort if there are any stashes in the
git repository (because that makes it look like not a clean clone), it
is not uncommon for people to want to work on repositories that are not
clean clones.  In such cases, rather than simply dropping all their
stashes, it is nicer if we can "rewrite the stashes" too.

To do so, we:
  * get the revision for each stash (getting the merge commit ID is
    sufficient since revision walking then includes all the parents)
  * get a copy of the contents of the .git/log/refs/stash file
  * provide the revisions for each stash to fast-export so that it
    iterates over the stashes too
  * between the reflog expiration and repacking steps, write out the
    contents of the .git/log/refs/stash file after first substituting
    for all the rewritten commits

Interestingly, the issue was caused by running
  git reflog expire --expire=now --all && git gc --prune=now --aggressive
upon completion of the filtering; it is the reflog expiration step in
particular that wipes out all the stashes.  So, any page that recommends
running both of those steps or any tool that recommends running those is
putting their users' stashes at risk.  As far as I can tell, that makes
git-filter-repo (after this commit) the first repository filtering tool
to not have the stash deletion problem.

Signed-off-by: Elijah Newren <[email protected]>

contrib/filter-repo-demos/bfg-ish

@@ -452,7 +452,7 @@ class BFG_ish:
     if not os.path.isabs(os.fsdecode(bfg_args.repo)):
       bfg_args.repo = os.fsencode(os.path.join(work_dir, os.fsdecode(bfg_args.repo)))
 
-    fr.RepoFilter.cleanup(bfg_args.repo, repack=True, reset=need_another_reset)
+    self.filter.cleanup(bfg_args.repo, repack=True, reset=need_another_reset)
 
 if __name__ == '__main__':
   bfg = BFG_ish()

git-filter-repo

@@ -2853,12 +2853,14 @@ class RepoFilter(object):
     self._finalize_handled = False
     self._orig_refs = None
     self._newnames = {}
+    self._stash = None
 
     # Cache a few message translations for performance reasons
     self._parsed_message = _("Parsed %d commits")
 
     # Compile some regexes and cache those
     self._hash_re = re.compile(br'(\b[0-9a-f]{7,40}\b)')
+    self._full_hash_re = re.compile(br'(\b[0-9a-f]{40}\b)')
 
   def _handle_arg_callbacks(self):
     def make_callback(argname, str):
@@ -3036,8 +3038,8 @@ class RepoFilter(object):
       if len(output.splitlines()) > 1:
         abort(_('you have multiple worktrees'))
 
-  @staticmethod
-  def cleanup(repo, repack, reset, run_quietly=False, show_debuginfo=False):
+  def cleanup(self, repo, repack, reset,
+              run_quietly=False, show_debuginfo=False):
     ''' cleanup repo; if repack then expire reflogs and do a gc --prune=now.
         if reset then do a reset --hard.  Optionally also curb output if
         run_quietly is True, or go the opposite direction and show extra
@@ -3058,6 +3060,8 @@ class RepoFilter(object):
       if show_debuginfo:
         print("[DEBUG] Running{}: {}".format(location_info, ' '.join(cmd)))
       subproc.call(cmd, cwd=repo)
+      if cmd[0:3] == 'git reflog expire'.split():
+        self._write_stash()
 
   def _get_rename(self, old_hash):
     # If we already know the rename, just return it
@@ -3115,6 +3119,13 @@ class RepoFilter(object):
     assert new_hash is not None
     return new_hash[0:orig_len]
 
+  def _translate_full_commit_hash(self, matchobj):
+    old_hash = matchobj.group(1)
+    new_hash = self._get_rename(old_hash)
+    if new_hash is None:
+      return old_hash
+    return new_hash
+
   def _trim_extra_parents(self, orig_parents, parents):
     '''Due to pruning of empty commits, some parents could be non-existent
        (None) or otherwise redundant.  Remove the non-existent parents, and
@@ -3685,12 +3696,37 @@ class RepoFilter(object):
     # Handle sanity checks, though currently none needed for export-only cases
     self._run_sanity_checks()
 
+  def _read_stash(self):
+    if self._orig_refs and b'refs/stash' in self._orig_refs and \
+       self._args.refs == ['--all']:
+      repo_working_dir = self._args.source or b'.'
+      git_dir = GitUtils.determine_git_dir(repo_working_dir)
+      stash = os.path.join(git_dir, b'logs', b'refs', b'stash')
+      if os.path.exists(stash):
+        with open(stash, 'br') as f:
+          self._stash = f.read()
+        out = subproc.check_output('git rev-list -g refs/stash'.split(),
+                                   cwd=repo_working_dir)
+        self._args.refs.extend(decode(out.strip()).split())
+
+  def _write_stash(self):
+    if self._stash:
+      target_working_dir = self._args.target or b'.'
+      git_dir = GitUtils.determine_git_dir(target_working_dir)
+      stash = os.path.join(git_dir, b'logs', b'refs', b'stash')
+      with open(stash, 'bw') as f:
+        self._stash = self._full_hash_re.sub(self._translate_full_commit_hash,
+                                             self._stash)
+        f.write(self._stash)
+      print(_("Rewrote the stash."))
+
   def _setup_input(self, use_done_feature):
     if self._args.stdin:
       self._input = sys.stdin.detach()
       sys.stdin = None # Make sure no one tries to accidentally use it
       self._fe_orig = None
     else:
+      self._read_stash()
       skip_blobs = (self._blob_callback is None and
                     self._args.replace_text is None and
                     self._args.source == self._args.target)
@@ -4038,9 +4074,9 @@ class RepoFilter(object):
     #   If we need a repack, then nuke the reflogs and repack.
     #   If we need a reset, do a reset --hard
     reset = not GitUtils.is_repository_bare(target_working_dir)
-    RepoFilter.cleanup(target_working_dir, self._args.repack, reset,
-                       run_quietly=self._args.quiet,
-                       show_debuginfo=self._args.debug)
+    self.cleanup(target_working_dir, self._args.repack, reset,
+                 run_quietly=self._args.quiet,
+                 show_debuginfo=self._args.debug)
 
     # Let user know how long it took
     print(_("Completely finished after {:.2f} seconds.")

t/t9390-filter-repo.sh

@@ -1918,4 +1918,26 @@ test_expect_success 'empty author ident' '
 	)
 '
 
+test_expect_success 'rewrite stash' '
+	test_create_repo rewrite_stash &&
+	(
+		cd rewrite_stash &&
+
+		git init &&
+		test_write_lines 1 2 3 4 5 6 7 8 9 10 >numbers &&
+		git add numbers &&
+		git commit -qm initial &&
+
+		echo 11 >>numbers &&
+		git stash push -m "add eleven" &&
+		echo foobar >>numbers &&
+		git stash push -m "add foobar" &&
+
+		git filter-repo --force --path-rename numbers:values &&
+
+		git stash list >output &&
+		test 2 -eq $(cat output | wc -l)
+	)
+'
+
 test_done