Re: [PATCH] repack.c: rename and unlink pack file if it exists
- From
Junio C Hamano <gitster@pobox.com>
- Date
- Feb 5, 2014, 20:06 UTC
- Message-ID
- <xmqq1tzh494e.fsf@gitster.dls.corp.google.com>
- In-Reply-To
- <20140205011632.GA3923@sigill.intra.peff.net>
Jeff King <peff@peff.net> writes:
Show 19 quoted lines
> On Tue, Feb 04, 2014 at 03:40:15PM -0800, Junio C Hamano wrote: > >> * Somehow this came to my private mailbox without Cc to list, so I >> am forwarding it. >> >> I think with 1190a1ac (pack-objects: name pack files after >> trailer hash, 2013-12-05), repacking the same set of objects may >> have less chance of producing colliding names, especially if you >> are on a box with more than one core, but it still would be a >> good idea to get this part right in the upcoming release. > > Actually, since 1190a1ac, if you have repacked and gotten the same pack > name, then you do not have to do any rename dance at all; you can throw > away what you just generated because you know that it is byte-for-byte > identical. > > You could collide with a pack created by an older version of git that > used the original scheme, but that is quite unlikely (on the order of > 2^-160).
Yes, so in that sense this is not so urgent, but I'm tempted to split the original patch into two and merge only the first one to 'master' before -rc3 (see below). The renaming of the variables added enough noise to cause me fail to spot a change mixed within.
-- >8 -- From: Torsten Bögershausen <tboegi@web.de> Date: Sun, 2 Feb 2014 16:09:56 +0100
When a repo was fully repacked, and is repacked again, we may run into the situation that "new" packfiles have the same name as already existing ones (traditionally packfiles have been named after the list of names of objects in them, so repacking all the objects in a single pack would have produced a packfile with the same name).
The logic is to rename the existing ones into filename like "old-XXX", create the new ones and then remove the "old-" ones. When something went wrong in the middle, this sequence is rolled back by renaming the "old-" files back.
The renaming into "old-" did not work as intended, because file_exists() was done on "XXX", not "pack-XXX". Also when rolling back the change, the code tried to rename "old-pack-XXX" but the saved ones are named "old-XXX", so this couldn't have worked.
Signed-off-by: Torsten Bögershausen <tboegi@web.de> Signed-off-by: Junio C Hamano <gitster@pobox.com> --- builtin/repack.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/builtin/repack.c b/builtin/repack.c index bca7710..fe31577 100644 --- a/builtin/repack.c +++ b/builtin/repack.c @@ -260,7 +260,7 @@ int cmd_repack(int argc, const char **argv, const char *prefix) for_each_string_list_item(item, &names) { for (ext = 0; ext < 2; ext++) { char *fname, *fname_old; - fname = mkpathdup("%s/%s%s", packdir, + fname = mkpathdup("%s/pack-%s%s", packdir, item->string, exts[ext]); if (!file_exists(fname)) { free(fname); @@ -337,7 +337,7 @@ int cmd_repack(int argc, const char **argv, const char *prefix) for_each_string_list_item(item, &names) { for (ext = 0; ext < 2; ext++) { char *fname; - fname = mkpath("%s/old-pack-%s%s", + fname = mkpath("%s/old-%s%s", packdir, item->string, exts[ext]);
-- 1.9-rc2-217-g24a8b2e