Re: [PATCH] branch: update all per-worktree HEADs when renaming a branch

[Kazuki Yamaguchi]

On Tue, Mar 22, 2016 at 07:49:00AM +0700, Duy Nguyen wrote:
> On Tue, Mar 22, 2016 at 12:41 AM, Eric Sunshine  
> wrote:
> >> diff --git a/worktree.c b/worktree.c
> >> @@ -217,3 +217,41 @@ char *find_shared_symref(const char *symref, const 
> >> char *target)
> >> +int update_worktrees_head_symref(const char *oldref, const char *newref)
> >> +{
> >> +   int error = 0;
> >> +   struct strbuf path = STRBUF_INIT;
> >> +   struct strbuf origref = STRBUF_INIT;
> >> +   int i;
> >> +   struct worktree **worktrees = get_worktrees();
> >> +
> >> +   for (i = 0; worktrees[i]; i++) {
> >> +   if (worktrees[i]->is_detached)
> >> +   continue;
> >> +
> >> +   strbuf_reset();
> >> +   strbuf_reset();
> >> +   strbuf_addf(, "%s/HEAD", worktrees[i]->git_dir);
> >> +
> >> +   if (parse_ref(path.buf, , NULL))
> >> +   continue;
> >> +
> >> +   if (!strcmp(origref.buf, oldref)) {
> >> +   int prefix_len = 
> >> strlen(absolute_path(get_git_common_dir())) + 1;
> >> +   const char *symref = path.buf + prefix_len;
> >> +
> >> +   /* no need to pass logmsg here as HEAD didn't 
> >> really move */
> >> +   if (create_symref(symref, newref, NULL)) {
> >> +   error = -1;
> >> +   break;
> >
> > Is aborting upon the first error desired behavior? (Genuine question.)
> > Would it make more sense to continue attempting the rename for the
> > remaining worktrees (and remember that an error was encountered)?
> 
> Since all these HEADs stay at the same (or close) location, if one
> fails, I think the rest will fail too. Which leads to a series of
> warnings if we continue anyway. A more interesting approach is update
> HEADs in a transaction, so we successfully update all or we update
> none. But I do not know if ref transactions can be used for HEAD,
> especially worktree HEADs. I'm ok with either abort here or continue
> anyway, though.
> -- 
> Duy

Thanks for suggestion, but it looks like ref_transaction can be used
only for updating non-symbolic references. Extending it only for this
purpose seems too much...
--
To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH] branch: update all per-worktree HEADs when renaming a branch

[Kazuki Yamaguchi]

Hello,

On 03/22/2016 03:41 AM, Eric Sunshine  wrote:
>> diff --git a/t/t3200-branch.sh b/t/t3200-branch.sh
>> @@ -126,7 +126,19 @@ test_expect_success 'git branch -M foo bar should fail 
>> when bar is checked out'
>>  test_expect_success 'git branch -M baz bam should succeed when baz is 
>> checked out' '
>> +test_expect_success 'git branch -M baz bam should succeed when baz is 
>> checked out as linked working tree' '
>> +   git checkout master &&
>> +   git branch &&
>> +   git worktree add -b baz bazdir &&
>> +   git branch -M baz bam &&
>> +   (
>> +   cd bazdir &&
>> +   test $(git rev-parse --abbrev-ref HEAD) = bam
>> +   )
>>  '
> 
> This can be done more easily without the subshell:
> 
> test $(git -C bazdir rev-parse ...) = bam

Thank you for reviewing. And sorry for late response.
I didn't know -C option, thanks.

> 
> Can you also expand the test so that it verifies that the rename works
> as expected when the branch is checked out in multiple worktrees,
> rather than just one. Likewise, it would be nice to check branch
> rename from within a worktree in which the branch is checked out (in
> addition to the test above which does the rename from outside such a
> worktree).
> 

I'll add them.
And I noticed my patch is broken in the latter case (rename in a linked
working tree).
Since create_symref() calls resolve_ref_unsafe() and it uses $GIT_DIR
for worktree-specific refs thus my patch fails to update main tree's
HEAD when we are in a linked working tree.
I'm thinking about adding new flag to resolve_ref_unsafe(), to force
using $GIT_COMMON_DIR. This will at the same time allows to remove
parse_ref() in worktree.c.

> 
>> diff --git a/worktree.c b/worktree.c
>> @@ -217,3 +217,41 @@ char *find_shared_symref(const char *symref, const char 
>> *target)
>> +int update_worktrees_head_symref(const char *oldref, const char *newref)
>> +{
>> +   int error = 0;
>> +   struct strbuf path = STRBUF_INIT;
>> +   struct strbuf origref = STRBUF_INIT;
>> +   int i;
>> +   struct worktree **worktrees = get_worktrees();
>> +
>> +   for (i = 0; worktrees[i]; i++) {
>> +   if (worktrees[i]->is_detached)
>> +   continue;
>> +
>> +   strbuf_reset();
>> +   strbuf_reset();
>> +   strbuf_addf(, "%s/HEAD", worktrees[i]->git_dir);
>> +
>> +   if (parse_ref(path.buf, , NULL))
>> +   continue;
>> +
>> +   if (!strcmp(origref.buf, oldref)) {
>> +   int prefix_len = 
>> strlen(absolute_path(get_git_common_dir())) + 1;
>> +   const char *symref = path.buf + prefix_len;
>> +
>> +   /* no need to pass logmsg here as HEAD didn't really 
>> move */
>> +   if (create_symref(symref, newref, NULL)) {
>> +   error = -1;
>> +   break;
> 
> Is aborting upon the first error desired behavior? (Genuine question.)
> Would it make more sense to continue attempting the rename for the
> remaining worktrees (and remember that an error was encountered)?
> Related: Since you're now dealing with multiple worktrees, you can do
> a better job of letting the user know in which worktree something went
> wrong rather than merely emitting the relatively generic "Branch
> renamed to %s, but HEAD is not updated!".

I think both is ok.
But continuing shouldn't be harm, so continuing might be better in terms
of that it can tell the user what files need to be fixed manually.
I'll try it.

>> +}
>> diff --git a/worktree.h b/worktree.h
>> @@ -35,4 +35,11 @@ extern void free_worktrees(struct worktree **);
>> +/*
>> + * Update all per-worktree HEADs pointing the old ref to point the new ref.
>> + * This will be used when renaming a branch. Returns 0 if successful,
>> + * non-zero otherwise.
>> + */
>> +extern int update_worktrees_head_symref(const char *, const char *);
> 
> I guess I can understand the desire to libify this functionality,
> however, it feels as if it is a feature of "branch" rather than
> "worktree", hence perhaps it should reside in top-level branch.[hc]?

I agree, I'll move it.
I chose worktree.c just because it has parse_ref().
--
To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH] branch: update all per-worktree HEADs when renaming a branch

[Duy Nguyen]

On Tue, Mar 22, 2016 at 12:41 AM, Eric Sunshine  wrote:
>> diff --git a/worktree.c b/worktree.c
>> @@ -217,3 +217,41 @@ char *find_shared_symref(const char *symref, const char 
>> *target)
>> +int update_worktrees_head_symref(const char *oldref, const char *newref)
>> +{
>> +   int error = 0;
>> +   struct strbuf path = STRBUF_INIT;
>> +   struct strbuf origref = STRBUF_INIT;
>> +   int i;
>> +   struct worktree **worktrees = get_worktrees();
>> +
>> +   for (i = 0; worktrees[i]; i++) {
>> +   if (worktrees[i]->is_detached)
>> +   continue;
>> +
>> +   strbuf_reset();
>> +   strbuf_reset();
>> +   strbuf_addf(, "%s/HEAD", worktrees[i]->git_dir);
>> +
>> +   if (parse_ref(path.buf, , NULL))
>> +   continue;
>> +
>> +   if (!strcmp(origref.buf, oldref)) {
>> +   int prefix_len = 
>> strlen(absolute_path(get_git_common_dir())) + 1;
>> +   const char *symref = path.buf + prefix_len;
>> +
>> +   /* no need to pass logmsg here as HEAD didn't really 
>> move */
>> +   if (create_symref(symref, newref, NULL)) {
>> +   error = -1;
>> +   break;
>
> Is aborting upon the first error desired behavior? (Genuine question.)
> Would it make more sense to continue attempting the rename for the
> remaining worktrees (and remember that an error was encountered)?

Since all these HEADs stay at the same (or close) location, if one
fails, I think the rest will fail too. Which leads to a series of
warnings if we continue anyway. A more interesting approach is update
HEADs in a transaction, so we successfully update all or we update
none. But I do not know if ref transactions can be used for HEAD,
especially worktree HEADs. I'm ok with either abort here or continue
anyway, though.
-- 
Duy
--
To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH] branch: update all per-worktree HEADs when renaming a branch

[Eric Sunshine]

On Mon, Mar 21, 2016 at 5:50 AM, Kazuki Yamaguchi  wrote:
> When renaming a branch, the current code only updates the current
> working tree's HEAD, but it should update .git/HEAD of all checked out
> working trees.
>
> This is the current behavior, /path/to/wt's HEAD is not updated:
> [...]
> This patch fixes this issue by updating all relevant worktree HEADs
> when renaming a branch.

Makes sense; seems like a genuine problem. Some comment below...

> Signed-off-by: Kazuki Yamaguchi 
> ---
> diff --git a/t/t3200-branch.sh b/t/t3200-branch.sh
> @@ -126,7 +126,19 @@ test_expect_success 'git branch -M foo bar should fail 
> when bar is checked out'
>  test_expect_success 'git branch -M baz bam should succeed when baz is 
> checked out' '
> +test_expect_success 'git branch -M baz bam should succeed when baz is 
> checked out as linked working tree' '
> +   git checkout master &&
> +   git branch &&
> +   git worktree add -b baz bazdir &&
> +   git branch -M baz bam &&
> +   (
> +   cd bazdir &&
> +   test $(git rev-parse --abbrev-ref HEAD) = bam
> +   )
>  '

This can be done more easily without the subshell:

test $(git -C bazdir rev-parse ...) = bam

Can you also expand the test so that it verifies that the rename works
as expected when the branch is checked out in multiple worktrees,
rather than just one. Likewise, it would be nice to check branch
rename from within a worktree in which the branch is checked out (in
addition to the test above which does the rename from outside such a
worktree).

More below...

> diff --git a/worktree.c b/worktree.c
> @@ -217,3 +217,41 @@ char *find_shared_symref(const char *symref, const char 
> *target)
> +int update_worktrees_head_symref(const char *oldref, const char *newref)
> +{
> +   int error = 0;
> +   struct strbuf path = STRBUF_INIT;
> +   struct strbuf origref = STRBUF_INIT;
> +   int i;
> +   struct worktree **worktrees = get_worktrees();
> +
> +   for (i = 0; worktrees[i]; i++) {
> +   if (worktrees[i]->is_detached)
> +   continue;
> +
> +   strbuf_reset();
> +   strbuf_reset();
> +   strbuf_addf(, "%s/HEAD", worktrees[i]->git_dir);
> +
> +   if (parse_ref(path.buf, , NULL))
> +   continue;
> +
> +   if (!strcmp(origref.buf, oldref)) {
> +   int prefix_len = 
> strlen(absolute_path(get_git_common_dir())) + 1;
> +   const char *symref = path.buf + prefix_len;
> +
> +   /* no need to pass logmsg here as HEAD didn't really 
> move */
> +   if (create_symref(symref, newref, NULL)) {
> +   error = -1;
> +   break;

Is aborting upon the first error desired behavior? (Genuine question.)
Would it make more sense to continue attempting the rename for the
remaining worktrees (and remember that an error was encountered)?
Related: Since you're now dealing with multiple worktrees, you can do
a better job of letting the user know in which worktree something went
wrong rather than merely emitting the relatively generic "Branch
renamed to %s, but HEAD is not updated!".

More below...

> +   }
> +   }
> +   }
> +
> +   strbuf_release();
> +   strbuf_release();
> +   free_worktrees(worktrees);
> +
> +   return error;
> +}
> diff --git a/worktree.h b/worktree.h
> @@ -35,4 +35,11 @@ extern void free_worktrees(struct worktree **);
> +/*
> + * Update all per-worktree HEADs pointing the old ref to point the new ref.
> + * This will be used when renaming a branch. Returns 0 if successful,
> + * non-zero otherwise.
> + */
> +extern int update_worktrees_head_symref(const char *, const char *);

I guess I can understand the desire to libify this functionality,
however, it feels as if it is a feature of "branch" rather than
"worktree", hence perhaps it should reside in top-level branch.[hc]?
--
To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH] branch: update all per-worktree HEADs when renaming a branch

[Kazuki Yamaguchi]

When renaming a branch, the current code only updates the current
working tree's HEAD, but it should update .git/HEAD of all checked out
working trees.

This is the current behavior, /path/to/wt's HEAD is not updated:

  % git worktree list
  /path/to 2c3c5f2 [master]
  /path/to/wt  2c3c5f2 [oldname]
  % git branch -m master master2
  % git worktree list
  /path/to 2c3c5f2 [master2]
  /path/to/wt  2c3c5f2 [oldname]
  % git branch -m oldname newname
  % git worktree list
  /path/to 2c3c5f2 [master2]
  /path/to/wt  000 [oldname]

This patch fixes this issue by updating all relevant worktree HEADs
when renaming a branch.

Signed-off-by: Kazuki Yamaguchi 
---
 builtin/branch.c  |  4 ++--
 t/t3200-branch.sh | 14 +-
 worktree.c| 38 ++
 worktree.h|  7 +++
 4 files changed, 60 insertions(+), 3 deletions(-)

diff --git a/builtin/branch.c b/builtin/branch.c
index 7b45b6b..596fb5f 100644
--- a/builtin/branch.c
+++ b/builtin/branch.c
@@ -20,6 +20,7 @@
 #include "utf8.h"
 #include "wt-status.h"
 #include "ref-filter.h"
+#include "worktree.h"
 
 static const char * const builtin_branch_usage[] = {
N_("git branch [] [-r | -a] [--merged | --no-merged]"),
@@ -552,8 +553,7 @@ static void rename_branch(const char *oldname, const char 
*newname, int force)
if (recovery)
warning(_("Renamed a misnamed branch '%s' away"), oldref.buf + 
11);
 
-   /* no need to pass logmsg here as HEAD didn't really move */
-   if (!strcmp(oldname, head) && create_symref("HEAD", newref.buf, NULL))
+   if (update_worktrees_head_symref(oldref.buf, newref.buf))
die(_("Branch renamed to %s, but HEAD is not updated!"), 
newname);
 
strbuf_addf(, "branch.%s", oldref.buf + 11);
diff --git a/t/t3200-branch.sh b/t/t3200-branch.sh
index a897248..da107d0 100755
--- a/t/t3200-branch.sh
+++ b/t/t3200-branch.sh
@@ -126,7 +126,19 @@ test_expect_success 'git branch -M foo bar should fail 
when bar is checked out'
 test_expect_success 'git branch -M baz bam should succeed when baz is checked 
out' '
git checkout -b baz &&
git branch bam &&
-   git branch -M baz bam
+   git branch -M baz bam &&
+   test $(git rev-parse --abbrev-ref HEAD) = bam
+'
+
+test_expect_success 'git branch -M baz bam should succeed when baz is checked 
out as linked working tree' '
+   git checkout master &&
+   git branch &&
+   git worktree add -b baz bazdir &&
+   git branch -M baz bam &&
+   (
+   cd bazdir &&
+   test $(git rev-parse --abbrev-ref HEAD) = bam
+   )
 '
 
 test_expect_success 'git branch -M master should work when master is checked 
out' '
diff --git a/worktree.c b/worktree.c
index 6181a66..9e7d0f3 100644
--- a/worktree.c
+++ b/worktree.c
@@ -217,3 +217,41 @@ char *find_shared_symref(const char *symref, const char 
*target)
 
return existing;
 }
+
+int update_worktrees_head_symref(const char *oldref, const char *newref)
+{
+   int error = 0;
+   struct strbuf path = STRBUF_INIT;
+   struct strbuf origref = STRBUF_INIT;
+   int i;
+   struct worktree **worktrees = get_worktrees();
+
+   for (i = 0; worktrees[i]; i++) {
+   if (worktrees[i]->is_detached)
+   continue;
+
+   strbuf_reset();
+   strbuf_reset();
+   strbuf_addf(, "%s/HEAD", worktrees[i]->git_dir);
+
+   if (parse_ref(path.buf, , NULL))
+   continue;
+
+   if (!strcmp(origref.buf, oldref)) {
+   int prefix_len = 
strlen(absolute_path(get_git_common_dir())) + 1;
+   const char *symref = path.buf + prefix_len;
+
+   /* no need to pass logmsg here as HEAD didn't really 
move */
+   if (create_symref(symref, newref, NULL)) {
+   error = -1;
+   break;
+   }
+   }
+   }
+
+   strbuf_release();
+   strbuf_release();
+   free_worktrees(worktrees);
+
+   return error;
+}
diff --git a/worktree.h b/worktree.h
index b4b3dda..0d15d11 100644
--- a/worktree.h
+++ b/worktree.h
@@ -35,4 +35,11 @@ extern void free_worktrees(struct worktree **);
  */
 extern char *find_shared_symref(const char *symref, const char *target);
 
+/*
+ * Update all per-worktree HEADs pointing the old ref to point the new ref.
+ * This will be used when renaming a branch. Returns 0 if successful,
+ * non-zero otherwise.
+ */
+extern int update_worktrees_head_symref(const char *, const char *);
+
 #endif
-- 
2.8.0.rc3.13.gcd7ec22

--
To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html