git@vger.kernel.org mailing list mirror (one of many)
 help / color / mirror / code / Atom feed
From: Junio C Hamano <gitster@pobox.com>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Paolo Bonzini <bonzini@gnu.org>,
	Miklos Vajna <vmiklos@frugalware.org>, Jeff King <peff@peff.net>,
	git@vger.kernel.org
Subject: [RFH] two and half potential fixlets to the in-core index handling
Date: Sat, 23 Aug 2008 03:41:16 -0700	[thread overview]
Message-ID: <7vbpzkypwz.fsf_-_@gitster.siamese.dyndns.org> (raw)
In-Reply-To: <7v7ia8126z.fsf@gitster.siamese.dyndns.org> (Junio C. Hamano's message of "Sat, 23 Aug 2008 03:00:04 -0700")

While debugging merge-in-C with Miklos, I noticed that relatively recent
codepaths that deal with the in-core index are subtly broken when a single
process calls read_cache() more than once.

A historical invariant in the API is "if you have read the index or
populated it, then the next read_cache() is a no-op and you need to run
discard_cache() beforehand if you want to read a new on-disk index."  I
think this was broken by relatively recent unpack-trees rewrite, namely
34110cd (Make 'unpack_trees()' have a separate source and destination
index, 2008-03-06) and also 7a51ed6 (Make on-disk index representation
separate from in-core one, 2008-01-14).

With the current code, unpack_trees() builds its merge into a cleaned
o->result (an index_state structure), and it is copied to o->dst_index,
which often is the_index.  But the logic in read_from_index() that
implements the historical invariant checks if istate->alloc is NULL (in
which case the istate is empty -- just after startup, or just after
discard_index()).  The "result" index_state does not have alloc because it
is built from scratch and never allocated anything.

This wouldn't work very well.

A hacky solution I have in the attached patch is to waste an xmalloc(1)
and store it there when o->result is created, and also make
read_from_index() pay attention to the cache_nr and the cache_changed
bit. I think it is the safest and minimum fix.

Another independent issue is that istate has name_hash_initialized bit
that records the validity of the name_hash.  discard_index() frees the
name_hash, but it never resets the bit to zero.  I am not sure what the
ramifications of not doing so, but it certainly feels wrong.

Also, discard_index() did not free the array of pointers istate->cache[];
I think it should.

 read-cache.c   |   11 ++++-------
 unpack-trees.c |    2 ++
 2 files changed, 6 insertions(+), 7 deletions(-)

diff --git c/read-cache.c w/read-cache.c
index 2c03ec3..b29f263 100644
--- c/read-cache.c
+++ w/read-cache.c
@@ -1144,7 +1144,7 @@ static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entrie
 	return ondisk_size + entries*per_entry;
 }
 
-/* remember to discard_cache() before reading a different cache! */
+/* remember to discard_index() before reading a different cache! */
 int read_index_from(struct index_state *istate, const char *path)
 {
 	int fd, i;
@@ -1155,7 +1155,7 @@ int read_index_from(struct index_state *istate, const char *path)
 	size_t mmap_size;
 
 	errno = EBUSY;
-	if (istate->alloc)
+	if (istate->alloc || istate->cache_nr || istate->cache_changed)
 		return istate->cache_nr;
 
 	errno = ENOENT;
@@ -1240,15 +1240,12 @@ unmap:
 
 int discard_index(struct index_state *istate)
 {
-	istate->cache_nr = 0;
-	istate->cache_changed = 0;
-	istate->timestamp = 0;
 	free_hash(&istate->name_hash);
 	cache_tree_free(&(istate->cache_tree));
 	free(istate->alloc);
-	istate->alloc = NULL;
+	free(istate->cache);
 
-	/* no need to throw away allocated active_cache */
+	memset(istate, 0, sizeof(*istate));
 	return 0;
 }
 
diff --git c/unpack-trees.c w/unpack-trees.c
index cba0aca..016fd46 100644
--- c/unpack-trees.c
+++ w/unpack-trees.c
@@ -378,6 +378,8 @@ int unpack_trees(unsigned len, struct tree_desc *t, struct unpack_trees_options
 	memset(&o->result, 0, sizeof(o->result));
 	if (o->src_index)
 		o->result.timestamp = o->src_index->timestamp;
+	if (o->dst_index)
+		o->result.alloc = xmalloc(1);
 	o->merge_size = len;
 
 	if (!dfc)

  reply	other threads:[~2008-08-23 10:42 UTC|newest]

Thread overview: 22+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2008-08-22  6:36 [bug] git `next' does not do trivial merges Paolo Bonzini
2008-08-22 19:31 ` Jeff King
2008-08-23  6:08   ` Miklos Vajna
2008-08-23  8:14     ` [PATCH] Fix in-index merge Miklos Vajna
2008-08-23  8:17       ` [PATCH] builtin-merge: fail properly when we are in the middle of a conflicted merge Miklos Vajna
2008-08-23  9:01         ` Junio C Hamano
2008-08-23 10:57           ` Miklos Vajna
2008-08-23 19:55             ` Junio C Hamano
2008-08-23 19:56               ` [PATCH 1/2] merge: fix numerus bugs around "trivial merge" area Junio C Hamano
2008-08-24  1:58                 ` Junio C Hamano
2008-08-28 13:43                   ` [PATCH] builtin-merge: avoid run_command_v_opt() for recursive and subtree Miklos Vajna
2008-08-23 19:57               ` [PATCH 2/2] unpack_trees(): protect the handcrafted in-core index from read_cache() Junio C Hamano
2008-08-23  9:50         ` [PATCH] builtin-merge: fail properly when we are in the middle of a conflicted merge Junio C Hamano
2008-08-23  8:50       ` [PATCH] Fix in-index merge Junio C Hamano
2008-08-23 10:41         ` Miklos Vajna
2008-08-23  9:19       ` Paolo Bonzini
2008-08-23  9:55         ` Junio C Hamano
2008-08-23 10:00           ` Junio C Hamano
2008-08-23 10:41             ` Junio C Hamano [this message]
2008-08-23 18:13               ` [RFH] two and half potential fixlets to the in-core index handling Linus Torvalds
2008-08-23 19:14                 ` Junio C Hamano
2008-08-23 19:21                   ` Junio C Hamano

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

  List information: http://vger.kernel.org/majordomo-info.html

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=7vbpzkypwz.fsf_-_@gitster.siamese.dyndns.org \
    --to=gitster@pobox.com \
    --cc=bonzini@gnu.org \
    --cc=git@vger.kernel.org \
    --cc=peff@peff.net \
    --cc=torvalds@linux-foundation.org \
    --cc=vmiklos@frugalware.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
Code repositories for project(s) associated with this public inbox

	https://80x24.org/mirrors/git.git

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).