git.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Junio C Hamano <gitster@pobox.com>
To: Miklos Vajna <vmiklos@frugalware.org>
Cc: git@vger.kernel.org, Stephan Beyer <s-beyer@gmx.net>
Subject: Re: What's cooking in git.git (Aug 2008, #05; Tue, 19)
Date: Tue, 02 Sep 2008 13:02:31 -0700	[thread overview]
Message-ID: <7v1w02l3jc.fsf@gitster.siamese.dyndns.org> (raw)
In-Reply-To: <20080901010612.GC16514@genesis.frugalware.org> (Miklos Vajna's message of "Mon, 1 Sep 2008 03:06:12 +0200")

Miklos Vajna <vmiklos@frugalware.org> writes:

> I tried to let cmd_revert() use merge_trees() only and not
> merge_recursive_generic(), but something is fishy with it.

> t3501-revert-cherry-pick passes fine, but t3404-rebase-interactive fail,
> becase once we have a conflict, git diff --name-status says 'M' for the
> given file and not 'U', which is obviously wrong.

I think this is because you are forgotting to write the index file out.
You are calling "git commit -n" at the end if a cherry-pick/revert is
clean, or giving the user back an unmerged state in the index otherwise.
In either case, after merge_trees() gives you a potentially unmerged index
back in-core, you need to write it to the $GIT_INDEX_FILE for later users.

Also notice that your output is awfully silent.  You are forgetting to
flush the buffered output that is kept in "obuf".

Please check what other extra things merge_recursive() function does that
merge_trees() doesn't; the above two are what I've spotted by code
inspection, but there may be others.  It could be an indication of
incorrect interface layering when the builtin was refactored.

I found it a bit disturbing that "index_only" and "call_depth" were not
part of merge_options structure.  The machinery is not yet meant to be
reentrant, so we may rely the previous call to the function (either
merge_recursive() or merge_trees() revert them to a sane initial value,
but it is a bit unnerving, having to depend on that assumption.

I think your code happily dereferences NULL when picking the root commit
is involved, although I do not think the failure in the test suite you saw
is related to this omission.

Here is a partial fix to address the above issues I noticed on
top of your version; untested.

 builtin-revert.c |   46 ++++++++++++++++++++++++++++++++++++----------
 1 files changed, 36 insertions(+), 10 deletions(-)

diff --git i/builtin-revert.c w/builtin-revert.c
index 3071518..41f3ca2 100644
--- i/builtin-revert.c
+++ w/builtin-revert.c
@@ -234,16 +234,27 @@ static int index_is_dirty(void)
 	return !!DIFF_OPT_TST(&rev.diffopt, HAS_CHANGES);
 }
 
+static struct tree *empty_tree(void)
+{
+	struct tree *tree = xcalloc(1, sizeof(struct tree));
+
+	tree->object.parsed = 1;
+	tree->object.type = OBJ_TREE;
+	pretend_sha1_file(NULL, 0, OBJ_TREE, tree->object.sha1);
+	return tree;
+}
+
 static int revert_or_cherry_pick(int argc, const char **argv)
 {
 	unsigned char head[20];
 	struct commit *base, *next, *parent;
-	int i;
+	int i, index_fd, clean;
 	char *oneline, *reencoded_message = NULL;
 	const char *message, *encoding;
 	const char *defmsg = xstrdup(git_path("MERGE_MSG"));
 	struct merge_options o;
-	struct tree *result;
+	struct tree *result, *next_tree, *base_tree, *head_tree;
+	static struct lock_file index_lock;
 
 	git_config(git_default_config, NULL);
 	me = action == REVERT ? "revert" : "cherry-pick";
@@ -254,6 +265,10 @@ static int revert_or_cherry_pick(int argc, const char **argv)
 	if (action == REVERT && !no_replay)
 		die("revert is incompatible with replay");
 
+	index_fd = hold_locked_index(&index_lock, 1);
+
+	if (read_cache() < 0)
+		die("git %s: failed to read the index", me);
 	if (no_commit) {
 		/*
 		 * We do not intend to commit immediately.  We just want to
@@ -266,12 +281,10 @@ static int revert_or_cherry_pick(int argc, const char **argv)
 	} else {
 		if (get_sha1("HEAD", head))
 			die ("You do not have a valid HEAD");
-		if (read_cache() < 0)
-			die("could not read the index");
 		if (index_is_dirty())
 			die ("Dirty index: cannot %s", me);
-		discard_cache();
 	}
+	discard_cache();
 
 	if (!commit->parents) {
 		if (action == REVERT)
@@ -305,6 +318,10 @@ static int revert_or_cherry_pick(int argc, const char **argv)
 		die ("Cannot get commit message for %s",
 				sha1_to_hex(commit->object.sha1));
 
+	if (parent && parse_commit(parent) < 0)
+		die("%s: cannot parse parent commit %s",
+		    me, sha1_to_hex(parent->object.sha1));
+
 	/*
 	 * "commit" is an existing commit.  We would want to apply
 	 * the difference it introduces since its first parent "prev"
@@ -351,11 +368,20 @@ static int revert_or_cherry_pick(int argc, const char **argv)
 	init_merge_options(&o);
 	o.branch1 = "HEAD";
 	o.branch2 = oneline;
-	parse_commit(next);
-	parse_commit(base);
-	if (!merge_trees(&o, lookup_commit_reference_gently(head, 0)->tree,
-				next->tree, base->tree, &result) ||
-		write_cache_as_tree(head, 0, NULL)) {
+
+	head_tree = parse_tree_indirect(head);
+	next_tree = next ? next->tree : empty_tree();
+	base_tree = base ? base->tree : empty_tree();
+
+	clean = merge_trees(&o,
+			    head_tree,
+			    next_tree, base_tree, &result);
+
+	if (write_cache(index_fd, active_cache, active_nr) ||
+	    commit_locked_index(&index_lock))
+		die("%s: Unable to write new index file", me);
+
+	if (!clean) {
 		add_to_msg("\nConflicts:\n\n");
 		for (i = 0; i < active_nr;) {
 			struct cache_entry *ce = active_cache[i++];

  parent reply	other threads:[~2008-09-02 20:03 UTC|newest]

Thread overview: 35+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2008-08-19  9:05 What's cooking in git.git (Aug 2008, #05; Tue, 19) Junio C Hamano
2008-08-19 11:02 ` Johannes Sixt
2008-08-19 12:35   ` Andreas Färber
2008-08-19 12:54 ` Miklos Vajna
2008-08-19 19:19   ` Junio C Hamano
2008-08-19 20:59     ` Miklos Vajna
2008-08-19 22:00       ` Junio C Hamano
2008-08-20 22:42         ` Miklos Vajna
2008-08-25  1:44         ` [PATCH] merge-recursive: introduce merge_options Miklos Vajna
2008-08-25  6:06           ` Junio C Hamano
2008-08-25 14:25             ` Miklos Vajna
2008-08-28  4:50               ` Junio C Hamano
2008-08-30 15:42               ` [PATCH] merge-recursive: fix subtree merge Miklos Vajna
2008-08-30 16:39                 ` Junio C Hamano
2008-08-30 17:55                 ` Junio C Hamano
2008-08-31 23:49                   ` Miklos Vajna
2008-09-01  1:06         ` What's cooking in git.git (Aug 2008, #05; Tue, 19) Miklos Vajna
2008-09-01  1:09           ` [PATCH] builtin-revert: use merge_recursive_generic() Miklos Vajna
2008-09-02 20:02           ` Junio C Hamano [this message]
2008-09-02 20:43             ` What's cooking in git.git (Aug 2008, #05; Tue, 19) Junio C Hamano
2008-09-02 22:05             ` [PATCH 0/2] Move call_depth and index_only to struct merge_options Miklos Vajna
2008-09-02 22:05               ` [PATCH 1/2] merge-recursive: move call_depth " Miklos Vajna
2008-09-02 22:05                 ` [PATCH 2/2] merge-recursive: move index_only " Miklos Vajna
2008-09-02 22:13                 ` [PATCH] Makefile: add merge_recursive.h to LIB_H Miklos Vajna
2008-09-02 22:39                   ` Junio C Hamano
2008-09-02 23:49                     ` Miklos Vajna
2008-09-02 22:39               ` [PATCH 0/2] Move call_depth and index_only to struct merge_options Junio C Hamano
2008-09-03  0:16                 ` [PATCH] merge-recursive: get rid of the index_only global variable Miklos Vajna
2008-09-03  0:39                 ` [PATCH] merge-recursive: move the global obuf to struct merge_options Miklos Vajna
2008-09-03 17:34                   ` Miklos Vajna
2008-09-03 17:34                     ` [PATCH 1/2] merge-recursive: move current_{file,directory}_set " Miklos Vajna
2008-09-03 17:34                       ` [PATCH 2/2] merge-recursive: move make_virtual_commit()'s virtual_id to merge_options Miklos Vajna
2008-09-04 19:03                         ` Junio C Hamano
2008-09-05 17:26                           ` [PATCH] merge-recursive: get rid of virtual_id Miklos Vajna
2008-09-04 19:05                     ` [PATCH] merge-recursive: move the global obuf to struct merge_options 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

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

  git send-email \
    --in-reply-to=7v1w02l3jc.fsf@gitster.siamese.dyndns.org \
    --to=gitster@pobox.com \
    --cc=git@vger.kernel.org \
    --cc=s-beyer@gmx.net \
    --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.
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 NNTP newsgroup(s).