git/list[1] front-page[2] threads[3] people[4] search[5] about
 

[PATCH v4 23/23] files_transaction_commit(): clean up empty directories

From
Michael Haggerty <mhagger@alum.mit.edu>
Date
Jan 6, 2017, 16:22 UTC
Message-ID
<dc7ea83d82cbdfe2693d5dc2eb3d5275a96a665f.1483719289.git.mhagger@alum.mit.edu>
In-Reply-To
<cover.1483719289.git.mhagger@alum.mit.edu>

When deleting/pruning references, remove any directories that are made empty by the deletion of loose references or of reflogs. Otherwise such empty directories can survive forever and accumulate over time. (Even 'pack-refs', which is smart enough to remove the parent directories of loose references that it prunes, leaves directories that were already empty.)

And now that files_transaction_commit() takes care of deleting the parent directories of loose references that it prunes, we don't have to do that in prune_ref() anymore.

This change would be unwise if the *creation* of these directories could race with our deletion of them. But the earlier changes in this patch series made the creation paths robust against races, so now it is safe to tidy them up more aggressively.

Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
---
 refs/files-backend.c  | 34 ++++++++++++++++++++++++++++------
 refs/refs-internal.h  | 11 +++++++++--
 t/t1400-update-ref.sh | 27 +++++++++++++++++++++++++++
 3 files changed, 64 insertions(+), 8 deletions(-)
diff --git a/refs/files-backend.c b/refs/files-backend.c
index bce0022..c426575 100644
--- a/refs/files-backend.c
+++ b/refs/files-backend.c
@@ -2346,7 +2346,6 @@ static void prune_ref(struct ref_to_prune *r)
 	}
 	ref_transaction_free(transaction);
 	strbuf_release(&err);
-	try_remove_empty_parents(r->name, REMOVE_EMPTY_PARENTS_REF);
 }
 
 static void prune_refs(struct ref_to_prune *r)
@@ -3794,6 +3793,7 @@ static int files_transaction_commit(struct ref_store *ref_store,
 					ret = TRANSACTION_GENERIC_ERROR;
 					goto cleanup;
 				}
+				update->flags |= REF_DELETED_LOOSE;
 			}
 
 			if (!(update->flags & REF_ISPRUNING))
@@ -3806,16 +3806,38 @@ static int files_transaction_commit(struct ref_store *ref_store,
 		ret = TRANSACTION_GENERIC_ERROR;
 		goto cleanup;
 	}
-	for_each_string_list_item(ref_to_delete, &refs_to_delete)
-		unlink_or_warn(git_path("logs/%s", ref_to_delete->string));
+
+	/* Delete the reflogs of any references that were deleted: */
+	for_each_string_list_item(ref_to_delete, &refs_to_delete) {
+		if (!unlink_or_warn(git_path("logs/%s", ref_to_delete->string)))
+			try_remove_empty_parents(ref_to_delete->string,
+						 REMOVE_EMPTY_PARENTS_REFLOG);
+	}
+
 	clear_loose_ref_cache(refs);
 
 cleanup:
 	transaction->state = REF_TRANSACTION_CLOSED;
 
-	for (i = 0; i < transaction->nr; i++)
-		if (transaction->updates[i]->backend_data)
-			unlock_ref(transaction->updates[i]->backend_data);
+	for (i = 0; i < transaction->nr; i++) {
+		struct ref_update *update = transaction->updates[i];
+		struct ref_lock *lock = update->backend_data;
+
+		if (lock)
+			unlock_ref(lock);
+
+		if (update->flags & REF_DELETED_LOOSE) {
+			/*
+			 * The loose reference was deleted. Delete any
+			 * empty parent directories. (Note that this
+			 * can only work because we have already
+			 * removed the lockfile.)
+			 */
+			try_remove_empty_parents(update->refname,
+						 REMOVE_EMPTY_PARENTS_REF);
+		}
+	}
+
 	string_list_clear(&refs_to_delete, 0);
 	free(head_ref);
 	string_list_clear(&affected_refnames, 0);
diff --git a/refs/refs-internal.h b/refs/refs-internal.h
index dc81acc..15d5a1e 100644
--- a/refs/refs-internal.h
+++ b/refs/refs-internal.h
@@ -56,6 +56,12 @@
 #define REF_UPDATE_VIA_HEAD 0x100
 
 /*
+ * Used as a flag in ref_update::flags when the loose reference has
+ * been deleted.
+ */
+#define REF_DELETED_LOOSE 0x200
+
+/*
  * Return true iff refname is minimally safe. "Safe" here means that
  * deleting a loose reference by this name will not do any damage, for
  * example by causing a file that is not a reference to be deleted.
@@ -158,8 +164,9 @@ struct ref_update {
 
 	/*
 	 * One or more of REF_HAVE_NEW, REF_HAVE_OLD, REF_NODEREF,
-	 * REF_DELETING, REF_ISPRUNING, REF_LOG_ONLY, and
-	 * REF_UPDATE_VIA_HEAD:
+	 * REF_DELETING, REF_ISPRUNING, REF_LOG_ONLY,
+	 * REF_UPDATE_VIA_HEAD, REF_NEEDS_COMMIT, and
+	 * REF_DELETED_LOOSE:
 	 */
 	unsigned int flags;
 
diff --git a/t/t1400-update-ref.sh b/t/t1400-update-ref.sh
index d4fb977..97d8793 100755
--- a/t/t1400-update-ref.sh
+++ b/t/t1400-update-ref.sh
@@ -191,6 +191,33 @@ test_expect_success \
 	 git update-ref HEAD'" $A &&
 	 test $A"' = $(cat .git/'"$m"')'
 
+test_expect_success "empty directory removal" '
+	git branch d1/d2/r1 HEAD &&
+	git branch d1/r2 HEAD &&
+	test -f .git/refs/heads/d1/d2/r1 &&
+	test -f .git/logs/refs/heads/d1/d2/r1 &&
+	git branch -d d1/d2/r1 &&
+	! test -e .git/refs/heads/d1/d2 &&
+	! test -e .git/logs/refs/heads/d1/d2 &&
+	test -f .git/refs/heads/d1/r2 &&
+	test -f .git/logs/refs/heads/d1/r2
+'
+
+test_expect_success "symref empty directory removal" '
+	git branch e1/e2/r1 HEAD &&
+	git branch e1/r2 HEAD &&
+	git checkout e1/e2/r1 &&
+	test_when_finished "git checkout master" &&
+	test -f .git/refs/heads/e1/e2/r1 &&
+	test -f .git/logs/refs/heads/e1/e2/r1 &&
+	git update-ref -d HEAD &&
+	! test -e .git/refs/heads/e1/e2 &&
+	! test -e .git/logs/refs/heads/e1/e2 &&
+	test -f .git/refs/heads/e1/r2 &&
+	test -f .git/logs/refs/heads/e1/r2 &&
+	test -f .git/logs/HEAD
+'
+
 cat >expect <<EOF
 $Z $A $GIT_COMMITTER_NAME <$GIT_COMMITTER_EMAIL> 1117150200 +0000	Initial Creation
 $A $B $GIT_COMMITTER_NAME <$GIT_COMMITTER_EMAIL> 1117150260 +0000	Switch

-- 
2.9.3
Previous: Michael HaggertyNext: Michael Haggerty
Message 14 of 25 in “Delete directories left empty after ref deletion”
  1. 00/23 Delete directories left empty after ref deletionMichael Haggerty, Jan 6, 2017
  2. 01/23 files_rename_ref(): tidy up whitespaceMichael Haggerty, Jan 6, 2017
  3. 03/23 t5505: use "for-each-ref" to test for the non-existence of referencesMichael Haggerty, Jan 6, 2017
  4. 06/23 raceproof_create_file(): new functionMichael Haggerty, Jan 6, 2017
  5. 02/23 refname_is_safe(): correct docstringMichael Haggerty, Jan 6, 2017
  6. 05/23 safe_create_leading_directories(): set errno on SCLD_EXISTSMichael Haggerty, Jan 6, 2017
  7. 10/23 rename_tmp_log(): improve error reportingMichael Haggerty, Jan 6, 2017
  8. 09/23 rename_tmp_log(): use raceproof_create_file()Michael Haggerty, Jan 6, 2017
  9. 04/23 safe_create_leading_directories_const(): preserve errnoMichael Haggerty, Jan 6, 2017
  10. 07/23 lock_ref_sha1_basic(): inline constantMichael Haggerty, Jan 6, 2017
  11. 13/23 log_ref_setup(): improve robustness against racesMichael Haggerty, Jan 6, 2017
  12. 08/23 lock_ref_sha1_basic(): use raceproof_create_file()Michael Haggerty, Jan 6, 2017
  13. 11/23 log_ref_write(): inline functionMichael Haggerty, Jan 6, 2017
  14. 23/23 files_transaction_commit(): clean up empty directoriesMichael Haggerty, Jan 6, 2017
  15. 18/23 delete_ref_loose(): derive loose reference path from lockMichael Haggerty, Jan 6, 2017
  16. 19/23 delete_ref_loose(): inline functionMichael Haggerty, Jan 6, 2017
  17. 22/23 try_remove_empty_parents(): teach to remove parents of reflogs, tooMichael Haggerty, Jan 6, 2017
  18. 21/23 try_remove_empty_parents(): don't trash argument contentsMichael Haggerty, Jan 6, 2017
  19. 20/23 try_remove_empty_parents(): rename parameter "name" -> "refname"Michael Haggerty, Jan 6, 2017
  20. 16/23 log_ref_setup(): manage the name of the reflog file internallyMichael Haggerty, Jan 6, 2017
  21. 12/23 log_ref_setup(): separate code for create vs non-createMichael Haggerty, Jan 6, 2017
  22. 15/23 log_ref_write_1(): don't depend on logfile argumentMichael Haggerty, Jan 6, 2017
  23. 17/23 log_ref_write_1(): inline functionMichael Haggerty, Jan 6, 2017
  24. 14/23 log_ref_setup(): pass the open file descriptor back to the callerMichael Haggerty, Jan 6, 2017
  25. Jeff KingJan 6, 2017

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.