threads / discuss / 22101

checkout -m dumping core

Subject: checkout -m dumping core

## tl;dr

3 messages between Jan 5, 2010 and Jan 7, 2010.

replies: 2people: 3as markdown or json

Tomas Carnecky· Jan 5, 2010, 18:15 UTC · lore
git version 1.6.6.78.gbd757c

HEAD points to a non-existent branch refs/heads/master. Normal checkout origin fails with: error: Entry '.cvsignore' would be overwritten by merge. Cannot merge. (the working tree does indeed contain this file). So I tried checkout -m and git crashed. Workaround for me was reset --hard origin; checkout origin. I have the repository backed up in case someone wants to try themselves.

$ gdb `which git` GNU gdb 6.8 Copyright (C) 2008 Free Software Foundation, Inc. License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html> This is free software: you are free to change and redistribute it. There is NO WARRANTY, to the extent permitted by law. Type "show copying" and "show warranty" for details. This GDB was configured as "i386-pc-solaris2.11"... (gdb) run checkout -m origin Starting program: /export/home/tomc/local/git/bin/git checkout -m origin warning: Lowest section in /lib/libpthread.so.1 is .dynamic at 00000074

Program received signal SIGSEGV, Segmentation fault.
0x080788fa in cmd_checkout (argc=0, argv=0x8047538, prefix=0x0) at 
builtin-checkout.c:450
450                             merge_trees(&o, new->commit->tree, work,
(gdb) list
445                             ret = reset_tree(new->commit->tree, 
opts, 1);
446                             if (ret)
447                                     return ret;
448                             o.branch1 = new->name;
449                             o.branch2 = "local";
450                             merge_trees(&o, new->commit->tree, work,
451                                     old->commit->tree, &result);
452                             ret = reset_tree(new->commit->tree, 
opts, 0);
453                             if (ret)
454                                     return ret;
(gdb) p o
$1 = {branch1 = 0x8047650 "origin", branch2 = 0x0, subtree_merge = 0, 
buffer_output = 1, verbosity = 0, diff_rename_limit = -1, 
merge_rename_limit = -1, call_depth = 0, obuf = {alloc = 0, len = 0, buf 
= 0x81643ac ""}, current_file_set = {
     items = 0x0, nr = 0, alloc = 0, strdup_strings = 1}, 
current_directory_set = {items = 0x0, nr = 0, alloc = 0, strdup_strings 
= 1}}
(gdb) p new
$2 = {name = 0x8047650 "origin", path = 0x8166438 "refs/heads/origin", 
commit = 0x8168f48}
(gdb) p work
$3 = (struct tree *) 0x8174f90
(gdb) p old
No symbol "old" in current context.
(gdb) p result
$4 = (struct tree *) 0xfefc81be
(gdb)
Daniel· Jan 6, 2010, 08:11 UTC · re: Tomas Carnecky · lore

Re: checkout -m dumping core

Tomas Carnecky <tom@dbservice.com> wrote:
Show 64 quoted lines
> git version 1.6.6.78.gbd757c
> 
> HEAD points to a non-existent branch refs/heads/master. Normal checkout 
> origin fails with:
> error: Entry '.cvsignore' would be overwritten by merge. Cannot merge.
> (the working tree does indeed contain this file). So I tried checkout -m 
> and git crashed. Workaround for me was reset --hard origin; checkout 
> origin. I have the repository backed up in case someone wants to try 
> themselves.
> 
> $ gdb `which git`
> GNU gdb 6.8
> Copyright (C) 2008 Free Software Foundation, Inc.
> License GPLv3+: GNU GPL version 3 or later 
> <http://gnu.org/licenses/gpl.html>
> This is free software: you are free to change and redistribute it.
> There is NO WARRANTY, to the extent permitted by law.  Type "show copying"
> and "show warranty" for details.
> This GDB was configured as "i386-pc-solaris2.11"...
> (gdb) run checkout -m origin
> Starting program: /export/home/tomc/local/git/bin/git checkout -m origin
> warning: Lowest section in /lib/libpthread.so.1 is .dynamic at 00000074
> 
> Program received signal SIGSEGV, Segmentation fault.
> 0x080788fa in cmd_checkout (argc=0, argv=0x8047538, prefix=0x0) at 
> builtin-checkout.c:450
> 450                             merge_trees(&o, new->commit->tree, work,
> (gdb) list
> 445                             ret = reset_tree(new->commit->tree, 
> opts, 1);
> 446                             if (ret)
> 447                                     return ret;
> 448                             o.branch1 = new->name;
> 449                             o.branch2 = "local";
> 450                             merge_trees(&o, new->commit->tree, work,
> 451                                     old->commit->tree, &result);
> 452                             ret = reset_tree(new->commit->tree, 
> opts, 0);
> 453                             if (ret)
> 454                                     return ret;
> (gdb) p o
> $1 = {branch1 = 0x8047650 "origin", branch2 = 0x0, subtree_merge = 0, 
> buffer_output = 1, verbosity = 0, diff_rename_limit = -1, 
> merge_rename_limit = -1, call_depth = 0, obuf = {alloc = 0, len = 0, buf 
> = 0x81643ac ""}, current_file_set = {
>      items = 0x0, nr = 0, alloc = 0, strdup_strings = 1}, 
> current_directory_set = {items = 0x0, nr = 0, alloc = 0, strdup_strings 
> = 1}}
> (gdb) p new
> $2 = {name = 0x8047650 "origin", path = 0x8166438 "refs/heads/origin", 
> commit = 0x8168f48}
> (gdb) p work
> $3 = (struct tree *) 0x8174f90
> (gdb) p old
> No symbol "old" in current context.
> (gdb) p result
> $4 = (struct tree *) 0xfefc81be
> (gdb)
> 
> --
> To unsubscribe from this list: send the line "unsubscribe git" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 
Does this patch help?
---
From b2203bded22db1a496ee3c9f6f5f4a384a8ccefa Mon Sep 17 00:00:00 2001
From: Daniel Baranski <mjucde@o2.pl>
Date: Wed, 6 Jan 2010 08:58:21 +0100
Subject: [PATCH] checkout -m: Fix SEGFAULT if HEAD is not valid.
Signed-off-by: Daniel Barański <mjucde@o2.pl>
---
 builtin-checkout.c |    3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)
diff --git a/builtin-checkout.c b/builtin-checkout.c
index 64f3a11..0ab59b2 100644
--- a/builtin-checkout.c
+++ b/builtin-checkout.c
@@ -422,7 +422,8 @@ static int merge_working_tree(struct checkout_opts *opts,
                        struct merge_options o;
                        if (!opts->merge)
                                return 1;
-                       parse_commit(old->commit);
+                       if (!parse_commit(old->commit))
+                               die("Couldn't parse commit '%s'", old->path);

                        /* Do more real merge */
-- 
1.6.5.6
Junio C Hamano· Jan 7, 2010, 05:07 UTC · re: Daniel · lore

Re: checkout -m dumping core

Daniel <mjucde@o2.pl> writes:
> Does this patch help?

There actually is no point running parse_commit() at that point, as the code obtained old->commit from lookup_commit_reference_gently() in switch_branches() which must have already parsed it. We do use it to try switching the branches before falling back to the --merge mode (or we notice old->commit does not exist and switch from an empty tree).

Instead of silently dying without diagnosing what paths are problematic like your patch does, we can unsquelch the error from the unpack_tree() we run before falling back to the --merge mode, and allow it to notice and report the paths that will be overwritten (since you do not have HEAD, everything in the work tree will be overwritten). Perhaps like this.

 builtin-checkout.c |   10 ++++++++--
 1 files changed, 8 insertions(+), 2 deletions(-)
diff --git a/builtin-checkout.c b/builtin-checkout.c
index 64f3a11..2708669 100644
--- a/builtin-checkout.c
+++ b/builtin-checkout.c
@@ -397,7 +397,7 @@ static int merge_working_tree(struct checkout_opts *opts,
 		topts.initial_checkout = is_cache_unborn();
 		topts.update = 1;
 		topts.merge = 1;
-		topts.gently = opts->merge;
+		topts.gently = opts->merge && old->commit;
 		topts.verbose_update = !opts->quiet;
 		topts.fn = twoway_merge;
 		topts.dir = xcalloc(1, sizeof(*topts.dir));
@@ -422,7 +422,13 @@ static int merge_working_tree(struct checkout_opts *opts,
 			struct merge_options o;
 			if (!opts->merge)
 				return 1;
-			parse_commit(old->commit);
+
+			/*
+			 * Without old->commit, the below is the same as
+			 * the two-tree unpack we already tried and failed.
+			 */
+			if (!old->commit)
+				return 1;
 
 			/* Do more real merge */
 

← back to recent threads