[PATCH 0/3] nd/multiple-work-trees updates

These patches are on top of what's in 'pu'. They add
--ignore-other-worktrees and make a note about current submodule
support status. I don't think submodule support is ready yet even
with Max Kirillov's series [1]. His 03/03 is already fixed in 'pu'
though, so only 01/03 and 02/03 are new.

[1] http://thread.gmane.org/gmane.comp.version-control.git/261107

Nguyễn Thái Ngọc Duy (3):
  checkout: pass whole struct to parse_branchname_arg instead of individual flags
  checkout: add --ignore-other-wortrees
  git-checkout.txt: a note about multiple checkout support for submodules

 Documentation/git-checkout.txt |  9 +++++++++
 builtin/checkout.c             | 19 +++++++++++--------
 t/t2025-checkout-to.sh         |  7 +++++++
 3 files changed, 27 insertions(+), 8 deletions(-)

-- 
2.2.0.84.ge9c7a8a

Noticed-by: Mark Levedahl [off-list ref]
Signed-off-by: Nguyễn Thái Ngọc Duy <redacted>
---
 Documentation/git-checkout.txt | 6 ++++++
 builtin/checkout.c             | 6 +++++-
 t/t2025-checkout-to.sh         | 7 +++++++
 3 files changed, 18 insertions(+), 1 deletion(-)

diff --git a/Documentation/git-checkout.txt b/Documentation/git-checkout.txt
index 0c13825..52eaa48 100644
--- a/Documentation/git-checkout.txt
+++ b/Documentation/git-checkout.txt

@@ -232,6 +232,12 @@ section of linkgit:git-add[1] to learn how to operate the `--patch` mode.
 	specific files such as HEAD, index... See "MULTIPLE WORKING
 	TREES" section for more information.
 
+--ignore-other-worktrees::
+	`git checkout` refuses when the wanted ref is already checked
+	out by another worktree. This option makes it check the ref
+	out anyway. In other words, the ref can be held by more than one
+	worktree.
+
 <branch>::
 	Branch to checkout; if it refers to a branch (i.e., a name that,
 	when prepended with "refs/heads/", is a valid ref), then that

diff --git a/builtin/checkout.c b/builtin/checkout.c
index d8717ef..8b2bf20 100644
--- a/builtin/checkout.c
+++ b/builtin/checkout.c

@@ -37,6 +37,7 @@ struct checkout_opts {
 	int writeout_stage;
 	int overwrite_ignore;
 	int ignore_skipworktree;
+	int ignore_other_worktrees;
 
 	const char *new_branch;
 	const char *new_branch_force;

@@ -1210,7 +1211,8 @@ static int parse_branchname_arg(int argc, const char **argv,
 		int flag;
 		char *head_ref = resolve_refdup("HEAD", 0, sha1, &flag);
 		if (head_ref &&
-		    (!(flag & REF_ISSYMREF) || strcmp(head_ref, new->path)))
+		    (!(flag & REF_ISSYMREF) || strcmp(head_ref, new->path)) &&
+		    !opts->ignore_other_worktrees)
 			check_linked_checkouts(new);
 		free(head_ref);
 	}

@@ -1341,6 +1343,8 @@ int cmd_checkout(int argc, const char **argv, const char *prefix)
 				N_("second guess 'git checkout no-such-branch'")),
 		OPT_FILENAME(0, "to", &opts.new_worktree,
 			   N_("check a branch out in a separate working directory")),
+		OPT_BOOL(0, "ignore-other-worktrees", &opts.ignore_other_worktrees,
+			 N_("do not check if another worktree is holding the given ref")),
 		OPT_END(),
 	};
 

diff --git a/t/t2025-checkout-to.sh b/t/t2025-checkout-to.sh
index 915b506..f8e4df4 100755
--- a/t/t2025-checkout-to.sh
+++ b/t/t2025-checkout-to.sh

@@ -79,6 +79,13 @@ test_expect_success 'die the same branch is already checked out' '
 	)
 '
 
+test_expect_success 'not die the same branch is already checked out' '
+	(
+		cd here &&
+		git checkout --ignore-other-worktrees --to anothernewmaster newmaster
+	)
+'
+
 test_expect_success 'not die on re-checking out current branch' '
 	(
 		cd there &&

-- 
2.2.0.84.ge9c7a8a

Signed-off-by: Nguyễn Thái Ngọc Duy <redacted>
---
 builtin/checkout.c | 13 ++++++-------
 1 file changed, 6 insertions(+), 7 deletions(-)

diff --git a/builtin/checkout.c b/builtin/checkout.c
index 953b763..d8717ef 100644
--- a/builtin/checkout.c
+++ b/builtin/checkout.c

@@ -1079,11 +1079,12 @@ static void check_linked_checkouts(struct branch_info *new)
 static int parse_branchname_arg(int argc, const char **argv,
 				int dwim_new_local_branch_ok,
 				struct branch_info *new,
-				struct tree **source_tree,
-				unsigned char rev[20],
-				const char **new_branch,
-				int force_detach)
+				struct checkout_opts *opts,
+				unsigned char rev[20])
 {
+	struct tree **source_tree = &opts->source_tree;
+	const char **new_branch = &opts->new_branch;
+	int force_detach = opts->force_detach;
 	int argcount = 0;
 	unsigned char branch_rev[20];
 	const char *arg;

@@ -1420,9 +1421,7 @@ int cmd_checkout(int argc, const char **argv, const char *prefix)
 			opts.track == BRANCH_TRACK_UNSPECIFIED &&
 			!opts.new_branch;
 		int n = parse_branchname_arg(argc, argv, dwim_ok,
-					     &new, &opts.source_tree,
-					     rev, &opts.new_branch,
-					     opts.force_detach);
+					     &new, &opts, rev);
 		argv += n;
 		argc -= n;
 	}

-- 
2.2.0.84.ge9c7a8a

The goal seems to be using multiple checkouts to reduce disk space.
But we have not reached an agreement how things should be. There are a
couple options.

 - You may want to keep $SUB repos elsewhere (perhaps in a central
   place) outside $SUPER. This is also true for nested submodules
   where a superproject may be a submodule of another superproject.

 - You may want to keep all $SUB repos in $SUPER/modules (or some
   other place in $SUPER)

 - We could even push it further and merge all $SUB repos into $SUPER
   instead of storing them separately. But that would at least require
   ref namespace enabled.

On top of that, git-submodule.sh expects $GIT_DIR/config to be
per-worktree, at least for the submodule.* part. Here I think we have
two options, either update config.c to also read
$GIT_DIR/config.worktree (which is per worktree) in addition to
$GIT_DIR/config (shared) and store worktree-specific vars in the new
place, or update git-submodule.sh to read/write submodule.* directly
from $GIT_DIR/config.submodule (per worktree).

These take time to address properly. Meanwhile, make a note to the
user that they should not use multiple worktrees in submodule context.

Signed-off-by: Nguyễn Thái Ngọc Duy <redacted>
---
 Documentation/git-checkout.txt | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/Documentation/git-checkout.txt b/Documentation/git-checkout.txt
index 52eaa48..72def5b 100644
--- a/Documentation/git-checkout.txt
+++ b/Documentation/git-checkout.txt

@@ -463,6 +463,9 @@ to `/path/main/.git/worktrees/test-next` then a file named
 `test-next` entry from being pruned.  See
 linkgit:gitrepository-layout[5] for details.
 
+Multiple checkout support for submodules is incomplete. It is NOT
+recommended to make multiple checkouts of a superproject.
+
 EXAMPLES
 --------
 

-- 
2.2.0.84.ge9c7a8a

On 01/03/2015 04:41 AM, Nguyễn Thái Ngọc Duy wrote:

The goal seems to be using multiple checkouts to reduce disk space.
But we have not reached an agreement how things should be. There are a
couple options.

  - You may want to keep $SUB repos elsewhere (perhaps in a central
    place) outside $SUPER. This is also true for nested submodules
    where a superproject may be a submodule of another superproject.

This is my preference: I keep a tree of bare git repos outside of all 
work areas, and use new-workdir to create trees of workdirs as needed. I 
explored trying to keep $SUB repos in others (including mods to 
submodule / new-workdir to manage this), found this really leads to too 
much complication compared to just having a set of bare repos elsewhere. 
This bare repo approach also has the advantage that no particular 
workdir is special, all workdirs that point to the same gitdir are equal.

Mark

On 01/03/2015 04:41 AM, Nguyễn Thái Ngọc Duy wrote:

Thanks for adding this, I haven't had a chance to test but by reading 
this solves the problem I raised.

Mark

On Sat, Jan 03, 2015 at 04:41:27PM +0700, Nguyễn Thái Ngọc Duy wrote:

The goal seems to be using multiple checkouts to reduce disk space.
But we have not reached an agreement how things should be. There are a
couple options.

 - You may want to keep $SUB repos elsewhere (perhaps in a central
   place) outside $SUPER. This is also true for nested submodules
   where a superproject may be a submodule of another superproject.

 - You may want to keep all $SUB repos in $SUPER/modules (or some
   other place in $SUPER)

 - We could even push it further and merge all $SUB repos into $SUPER
   instead of storing them separately. But that would at least require
   ref namespace enabled.

On top of that, git-submodule.sh expects $GIT_DIR/config to be
per-worktree, at least for the submodule.* part. Here I think we have
two options, either update config.c to also read
$GIT_DIR/config.worktree (which is per worktree) in addition to
$GIT_DIR/config (shared) and store worktree-specific vars in the new
place, or update git-submodule.sh to read/write submodule.* directly
from $GIT_DIR/config.submodule (per worktree).

These take time to address properly. Meanwhile, make a note to the
user that they should not use multiple worktrees in submodule context.

I'd like to describe though how much the glass is full:

If all submodules exist in same paths in all checked-out
commits, and there is no special settings, only
`submodule.<name>.url` is set, then user can inspect, commit
and update the submodule content. Submodules can be either
independedntly initalized with `submodule update --init`
(being a fully independet clone), or checked-out with
`checkout --to` to the submodule worktree.

It is even verified by tests, so does it worth mentioning in
some manpage or at least commit message?

-- 
Max

Without having looked into this and nd/multiple-work-trees, but with 
"make multiple checkouts aware of each other" in mind: Could this 
mechanism be re-used to make alternates aware of each other, to mitigate 
the dangers of having  git gc  on an alternate remove objects that are 
used by a referencing repository?

Thanks
- Eph

On 03.01.2015 10:41, Nguyễn Thái Ngọc Duy wrote:

These patches are on top of what's in 'pu'. They add
--ignore-other-worktrees and make a note about current submodule
support status. I don't think submodule support is ready yet even
with Max Kirillov's series [1]. His 03/03 is already fixed in 'pu'
though, so only 01/03 and 02/03 are new.

[1] http://thread.gmane.org/gmane.comp.version-control.git/261107

Nguyễn Thái Ngọc Duy (3):
   checkout: pass whole struct to parse_branchname_arg instead of individual flags
   checkout: add --ignore-other-wortrees
   git-checkout.txt: a note about multiple checkout support for submodules

  Documentation/git-checkout.txt |  9 +++++++++
  builtin/checkout.c             | 19 +++++++++++--------
  t/t2025-checkout-to.sh         |  7 +++++++
  3 files changed, 27 insertions(+), 8 deletions(-)

On Wed, Mar 18, 2015 at 3:04 PM, Ephrim Khong [off-list ref] wrote:

Without having looked into this and nd/multiple-work-trees, but with "make
multiple checkouts aware of each other" in mind: Could this mechanism be
re-used to make alternates aware of each other, to mitigate the dangers of
having  git gc  on an alternate remove objects that are used by a
referencing repository?

If we can turn on ref namespace and make $GIT_DIR/config and hooks per
worktree, I think it may have a chance of replacing alternate object
mechanism entirely: one object database, one ref database (but refs of
each worktree is namespaced so no conflicts), multiple worktrees,
multiple config files, multiple hooks.

Because some config keys affect object database, having
multiple/conflicting config keys imply that this worktree may change
object database in a way trhat  impacts performance (not correctness)
of another worktree. Later on when we have multiple ref backends, if
config keys can change ref backend behavior (or even choose the
backend), we may run into other problems. This problem might go away
if we define that those "global" config keys can't be per-worktree..

In short, I am good at confusing people.
-- 
Duy