Re: [PATCH] fsck.c: fix bogus "empty tree" check
- From
Sergey Vlasov <vsu@altlinux.ru>
- Date
- Mar 4, 2008, 12:26 UTC
- Message-ID
- <20080304152635.40451f7c.vsu@altlinux.ru>
- In-Reply-To
- <7vbq5u91lf.fsf@gitster.siamese.dyndns.org>
On Tue, 04 Mar 2008 03:21:16 -0800 Junio C Hamano wrote:
Show 14 quoted lines
> * I've wasted a few hours tonight hunting for random breakages in "git > push", the symptom of which is "fatal: unresolved deltas left after > unpacking." I was hoping this patch would fix it, but it seems that > the problem is elsewhere. > > I'll revert the following two commits for now: > > d5ef408 (unpack-objects: prevent writing of inconsistent objects) > 28f72a0 (receive-pack: use strict mode for unpacking objects) > > as I have verified that running with receive.fsckobjects set to false > fixes the issues for me, and the repository at the receiving end (both > before and after the push) pass git-fsck without problems. Needless to > say, I am not a happy camper right now.
This part of commit d5ef408 changes is bogus:
Show 7 quoted lines
> @@ -144,9 +205,36 @@ static void added_object(unsigned nr, enum object_type type,
> static void write_object(unsigned nr, enum object_type type,
> void *buf, unsigned long size)
> {
> - if (write_sha1_file(buf, size, typename(type), obj_list[nr].sha1) < 0)
> - die("failed to write object");
> added_object(nr, type, buf, size);The write_sha1_file() call here was calculating obj_list[nr].sha1; now it is removed, but added_object() needs this value:
| static void added_object(unsigned nr, enum object_type type,
| void *data, unsigned long size)
| {
| struct delta_info **p = &delta_list;
| struct delta_info *info;
|
| while ((info = *p) != NULL) {
| if (!hashcmp(info->base_sha1, obj_list[nr].sha1) ||
^^^^^^^^^^^^^^^^^
| info->base_offset == obj_list[nr].offset) {
| *p = info->next;
| p = &delta_list;
| resolve_delta(info->nr, type, data, size,
| info->delta, info->size);
| free(info);
| continue;
| }
| p = &info->next;
| }
| }However, I do not have time to create a proper test case for this.
Show 32 quoted lines
> + if (!strict) {
> + if (write_sha1_file(buf, size, typename(type), obj_list[nr].sha1) < 0)
> + die("failed to write object");
> + free(buf);
> + obj_list[nr].obj = 0;
> + } else if (type == OBJ_BLOB) {
> + struct blob *blob;
> + if (write_sha1_file(buf, size, typename(type), obj_list[nr].sha1) < 0)
> + die("failed to write object");
> + free(buf);
> +
> + blob = lookup_blob(obj_list[nr].sha1);
> + if (blob)
> + blob->object.flags |= FLAG_WRITTEN;
> + else
> + die("invalid blob object");
> + obj_list[nr].obj = 0;
> + } else {
> + struct object *obj;
> + int eaten;
> + hash_sha1_file(buf, size, typename(type), obj_list[nr].sha1);
> + obj = parse_object_buffer(obj_list[nr].sha1, type, size, buf, &eaten);
> + if (!obj)
> + die("invalid %s", typename(type));
> + /* buf is stored via add_object_buffer and in obj, if its a tree or commit */
> + add_object_buffer(obj, buf, size);
> + obj->flags |= FLAG_OPEN;
> + obj_list[nr].obj = obj;
> + }
> }
>
> static void resolve_delta(unsigned nr, enum object_type type,The simplest way to fix this would be to duplicate the added_object() call in all branches; invoking hash_sha1_file() unconditionally will work too, but may be wasteful if we need to call write_sha1_file() afterwards.