[PATCH] submodule: no [--merge|--rebase] when newly cloned

[PATCH] submodule: no [--merge|--rebase] when newly cloned

[Spencer E. Olson]

Previously, when a submodule was cloned in the same command execution, --rebase
or --merge would attempt to run (instead of and) before a checkout, resulting in
an empty working directory.  This patch ignores --rebase or --merge for a
particular submodule when that submodule is newly cloned and instead simply
checks out the appropriate revision.

Signed-off-by: Spencer E. Olson <olsonse@umich.edu>
---
 git-submodule.sh |   28 ++++++++++++++++++++++++++++
 1 files changed, 28 insertions(+), 0 deletions(-)

diff --git a/git-submodule.sh b/git-submodule.sh
index 8b90589..44d8741 100755
--- a/git-submodule.sh
+++ b/git-submodule.sh
@@ -365,6 +365,27 @@ cmd_init()
 }
 
 #
+# Test whether an element of the $3(=IFS) separated list $2 matches $1
+#
+str_contains()
+{
+	if test -n "$3"
+	then
+		local IFS="$3"
+	fi
+	for i in $2
+	do
+		if test "$i" = "$1"
+		then
+			echo "yes"
+			return
+		fi
+	done
+	echo "no"
+	return
+}
+
+#
 # Update each submodule path to correct revision, using clone and checkout as needed
 #
 # $@ = requested paths (default to all)
@@ -423,6 +444,7 @@ cmd_update()
 		cmd_init "--" "$@" || return
 	fi
 
+	cloned_modules=
 	module_list "$@" |
 	while read mode sha1 stage path
 	do
@@ -442,6 +464,7 @@ cmd_update()
 		if ! test -d "$path"/.git -o -f "$path"/.git
 		then
 			module_clone "$path" "$url" "$reference"|| exit
+			cloned_modules="$cloned_modules;$name"
 			subsha1=
 		else
 			subsha1=$(clear_local_git_env; cd "$path" &&
@@ -469,6 +492,11 @@ cmd_update()
 				die "Unable to fetch in submodule path '$path'"
 			fi
 
+			if test "$(str_contains "$name" "$cloned_modules" '\;')" = "yes"
+			then
+				update_module=
+			fi
+
 			case "$update_module" in
 			rebase)
 				command="git rebase"
-- 
1.7.4.1.42.g43f9f

Re: [PATCH] submodule: no [--merge|--rebase] when newly cloned

[Junio C Hamano]

"Spencer E. Olson" <olsonse@umich.edu> writes:

> Previously, when a submodule was cloned in the same command execution, --rebase
> or --merge would attempt to run (instead of and) before a checkout, resulting in
> an empty working directory.  This patch ignores --rebase or --merge for a
> particular submodule when that submodule is newly cloned and instead simply
> checks out the appropriate revision.

Sorry, but I cannot parse the problem description, "(instead of and)" part.

Why is it a better thing to do to ignore these options, instead of
detecting the situation and error it out, saying "you are initially
cloning, don't say --rebase"?

> +# Test whether an element of the $3(=IFS) separated list $2 matches $1
> +#
> +str_contains()
> +{
> +	if test -n "$3"
> +	then
> +		local IFS="$3"
> +	fi
> +	for i in $2
> +	do
> +		if test "$i" = "$1"
> +		then
> +			echo "yes"
> +			return
> +		fi
> +	done
> +	echo "no"
> +	return
> +}

We don't allow bash-ism outside POSIX shell in our primary Porcelain set
(the rule is looser for stuff in contrib/), so "local" is out.

Here is how to write the above more concisely, efficiently and portably.

	case "$2;" in
        *";$1;"*)
        	echo yes ;;
        *)
        	echo no ;;
	esac

The trailing ';' takes care of the case where cloned_modules has only one
element, in which case you have ";name" in "$2".  No need for a loop.

> +	cloned_modules=
> ...
>  	while read mode sha1 stage path
>  	do
> @@ -442,6 +464,7 @@ cmd_update()
>  		if ! test -d "$path"/.git -o -f "$path"/.git
>  		then
>  			module_clone "$path" "$url" "$reference"|| exit
> +			cloned_modules="$cloned_modules;$name"
>  			subsha1=

Re: [PATCH] submodule: no [--merge|--rebase] when newly cloned

[Jens Lehmann]

Am 16.02.2011 20:51, schrieb Junio C Hamano:
> "Spencer E. Olson" <olsonse@umich.edu> writes:
> 
>> Previously, when a submodule was cloned in the same command execution, --rebase
>> or --merge would attempt to run (instead of and) before a checkout, resulting in
>> an empty working directory.  This patch ignores --rebase or --merge for a
>> particular submodule when that submodule is newly cloned and instead simply
>> checks out the appropriate revision.
> 
> Sorry, but I cannot parse the problem description, "(instead of and)" part.

I think what he wanted to say was that after a new submodule is cloned
by running "git submodule update" it will attempt to do a rebase or
merge without having checked out the submodule at all, which obviously
can't work. A plain checkout looks like the right thing to do, as there
aren't any local commits to rebase or merge yet.

> Why is it a better thing to do to ignore these options, instead of
> detecting the situation and error it out, saying "you are initially
> cloning, don't say --rebase"?

It should be fine to use these options when new submodules appear.
(And even without explicitly specifing these command line options
this bug can also be triggered by having the "submodule.<name>.update"
option set in .gitmodules to either "rebase" or "merge", which is easy
to miss)

So this looks like a worthwhile fix. The commit message and POSIX issue
need to be addressed, tests would be a good thing to add too, but apart
from that it looks sane.

Re: [PATCH] submodule: no [--merge|--rebase] when newly cloned

[Junio C Hamano]

Jens Lehmann <Jens.Lehmann@web.de> writes:

> So this looks like a worthwhile fix. The commit message and POSIX issue
> need to be addressed, tests would be a good thing to add too, but apart
> from that it looks sane.

Thanks for a translation ;-).  I agree everthing you said above.

Re: [PATCH] submodule: no [--merge|--rebase] when newly cloned

[Spencer E. Olson]

I'll fix the POSIX issue and commit message and re-submit.  (I wasn't aware 
that for loops were from a bash-ism).

On Wednesday 16 February 2011 14:41, Junio C Hamano wrote:
> Jens Lehmann <Jens.Lehmann@web.de> writes:
> > So this looks like a worthwhile fix. The commit message and POSIX issue
> > need to be addressed, tests would be a good thing to add too, but apart
> > from that it looks sane.
>
> Thanks for a translation ;-).  I agree everthing you said above.

Re: [PATCH] submodule: no [--merge|--rebase] when newly cloned

[Johannes Sixt]

Am 2/16/2011 20:51, schrieb Junio C Hamano:
> Here is how to write the above more concisely, efficiently and portably.
> 
> 	case "$2;" in
>         *";$1;"*)
>         	echo yes ;;
>         *)
>         	echo no ;;
> 	esac
> 
> The trailing ';' takes care of the case where cloned_modules has only one
> element, in which case you have ";name" in "$2".  No need for a loop.

And while you are here, you could make this:

list_contains()
{
	case "$2;" in
	*";$1;"*)
		: yes ;;
	*)
		! : no ;;
	esac
}

and test for the exit code of this function rather than its output at the
call site.

-- Hannes

Re: [PATCH] submodule: no [--merge|--rebase] when newly cloned

[Martin von Zweigbergk]

On Thu, 17 Feb 2011, Johannes Sixt wrote:

> Am 2/16/2011 20:51, schrieb Junio C Hamano:
> > Here is how to write the above more concisely, efficiently and portably.
> > 
> > 	case "$2;" in
> >         *";$1;"*)
> >         	echo yes ;;
> >         *)
> >         	echo no ;;
> > 	esac
> > 
> > The trailing ';' takes care of the case where cloned_modules has only one
> > element, in which case you have ";name" in "$2".  No need for a loop.
> 
> And while you are here, you could make this:
> 
> list_contains()
> {
> 	case "$2;" in
> 	*";$1;"*)
> 		: yes ;;
> 	*)
> 		! : no ;;
> 	esac
> }
> 
> and test for the exit code of this function rather than its output at the
> call site.

According to Brandon Casey: "Some platforms (IRIX 6.5, Solaris 7) do
not provide the 'yes' utility." See 8648732 (t/test-lib.sh: provide a
shell implementation of the 'yes' utility, 2009-08-28).


/Martin

Re: [PATCH] submodule: no [--merge|--rebase] when newly cloned

[Spencer E. Olson]

We are not actually using the yes utility (think y\ny\ny\ny\ny\ny\n...)

As far as I understand, in this particular use, "yes" and "no" are just 
comments.  We could also write
: I like banannas ;;
and
! : I do not like anchovies ;;

On Thursday 17 February 2011 17:34, Martin von Zweigbergk wrote:
> On Thu, 17 Feb 2011, Johannes Sixt wrote:
> > Am 2/16/2011 20:51, schrieb Junio C Hamano:
> > > Here is how to write the above more concisely, efficiently and
> > > portably.
> > >
> > > 	case "$2;" in
> > >         *";$1;"*)
> > >         	echo yes ;;
> > >         *)
> > >         	echo no ;;
> > > 	esac
> > >
> > > The trailing ';' takes care of the case where cloned_modules has only
> > > one element, in which case you have ";name" in "$2".  No need for a
> > > loop.
> >
> > And while you are here, you could make this:
> >
> > list_contains()
> > {
> > 	case "$2;" in
> > 	*";$1;"*)
> >
> > 		: yes ;;
> >
> > 	*)
> > 		! : no ;;
> > 	esac
> > }
> >
> > and test for the exit code of this function rather than its output at the
> > call site.
>
> According to Brandon Casey: "Some platforms (IRIX 6.5, Solaris 7) do
> not provide the 'yes' utility." See 8648732 (t/test-lib.sh: provide a
> shell implementation of the 'yes' utility, 2009-08-28).
>
>
> /Martin

Re: [PATCH] submodule: no [--merge|--rebase] when newly cloned

[Martin von Zweigbergk]

On Thu, 17 Feb 2011, Spencer E. Olson wrote:

> We are not actually using the yes utility (think y\ny\ny\ny\ny\ny\n...)
> 
> As far as I understand, in this particular use, "yes" and "no" are just 
> comments.  We could also write
> : I like banannas ;;
> and
> ! : I do not like anchovies ;;

Ah, I'm sure you're right. Need to read up on that colon operator. Thanks.

> 
> On Thursday 17 February 2011 17:34, Martin von Zweigbergk wrote:
> > On Thu, 17 Feb 2011, Johannes Sixt wrote:
> > > Am 2/16/2011 20:51, schrieb Junio C Hamano:
> > > > Here is how to write the above more concisely, efficiently and
> > > > portably.
> > > >
> > > > 	case "$2;" in
> > > >         *";$1;"*)
> > > >         	echo yes ;;
> > > >         *)
> > > >         	echo no ;;
> > > > 	esac
> > > >
> > > > The trailing ';' takes care of the case where cloned_modules has only
> > > > one element, in which case you have ";name" in "$2".  No need for a
> > > > loop.
> > >
> > > And while you are here, you could make this:
> > >
> > > list_contains()
> > > {
> > > 	case "$2;" in
> > > 	*";$1;"*)
> > >
> > > 		: yes ;;
> > >
> > > 	*)
> > > 		! : no ;;
> > > 	esac
> > > }
> > >
> > > and test for the exit code of this function rather than its output at the
> > > call site.
> >
> > According to Brandon Casey: "Some platforms (IRIX 6.5, Solaris 7) do
> > not provide the 'yes' utility." See 8648732 (t/test-lib.sh: provide a
> > shell implementation of the 'yes' utility, 2009-08-28).
> >
> >
> > /Martin
>