[RFC] submodule: fix handling of supermodules with relative origin URLs.

[RFC] submodule: fix handling of supermodules with relative origin URLs.

[Jon Seymour]

Prior to this change, operations such as git submodule sync produces
the wrong result when the origin URL of the super module
is itself a relative URL.

The issue arises because in this case, the origin URL of the supermodule
needs to be prepended with a prefix that navigates from the submodule to
the supermodule.

This change adds that prefix.

Signed-off-by: Jon Seymour <jon.seymour@gmail.com>
---
 git-submodule.sh | 39 +++++++++++++++++++++++----------------
 1 file changed, 23 insertions(+), 16 deletions(-)

If people are ok with the fix, I'll roll this as a patch together
with some tests.

diff --git a/git-submodule.sh b/git-submodule.sh
index 64a70d6..5008867 100755
--- a/git-submodule.sh
+++ b/git-submodule.sh
@@ -37,7 +37,8 @@ resolve_relative_url ()
 	remoteurl=$(git config "remote.$remote.url") ||
 		remoteurl=$(pwd) # the repository is its own authoritative upstream
 	url="$1"
-	remoteurl=${remoteurl%/}
+	up_path="$(echo "$2" | sed "s/[^/]*/../g")"
+	remoteurl=${remoteurl%/*}
 	sep=/
 	while test -n "$url"
 	do
@@ -45,6 +46,9 @@ resolve_relative_url ()
 		../*)
 			url="${url#../}"
 			case "$remoteurl" in
+			.*/*)
+				remoteurl="${up_path%/}/${remoteurl%/*}"
+				;;
 			*/*)
 				remoteurl="${remoteurl%/*}"
 				;;
@@ -235,11 +239,24 @@ cmd_add()
 		usage
 	fi
 
+	# normalize path:
+	# multiple //; leading ./; /./; /../; trailing /
+	sm_path=$(printf '%s/\n' "$sm_path" |
+		sed -e '
+			s|//*|/|g
+			s|^\(\./\)*||
+			s|/\./|/|g
+			:start
+			s|\([^/]*\)/\.\./||
+			tstart
+			s|/*$||
+		')
+
 	# assure repo is absolute or relative to parent
 	case "$repo" in
 	./*|../*)
 		# dereference source url relative to parent's url
-		realrepo=$(resolve_relative_url "$repo") || exit
+		realrepo=$(resolve_relative_url "$repo" "$sm_path") || exit
 		;;
 	*:*|/*)
 		# absolute url
@@ -250,18 +267,6 @@ cmd_add()
 	;;
 	esac
 
-	# normalize path:
-	# multiple //; leading ./; /./; /../; trailing /
-	sm_path=$(printf '%s/\n' "$sm_path" |
-		sed -e '
-			s|//*|/|g
-			s|^\(\./\)*||
-			s|/\./|/|g
-			:start
-			s|\([^/]*\)/\.\./||
-			tstart
-			s|/*$||
-		')
 	git ls-files --error-unmatch "$sm_path" > /dev/null 2>&1 &&
 	die "$(eval_gettext "'\$sm_path' already exists in the index")"
 
@@ -401,13 +406,14 @@ cmd_init()
 		if test -z "$(git config "submodule.$name.url")"
 		then
 			url=$(git config -f .gitmodules submodule."$name".url)
+			sm_path=$(git config -f .gitmodules submodule."$name".path)
 			test -z "$url" &&
 			die "$(eval_gettext "No url found for submodule path '\$sm_path' in .gitmodules")"
 
 			# Possibly a url relative to parent
 			case "$url" in
 			./*|../*)
-				url=$(resolve_relative_url "$url") || exit
+				url=$(resolve_relative_url "$url" "$sm_path") || exit
 				;;
 			esac
 			git config submodule."$name".url "$url" ||
@@ -960,11 +966,12 @@ cmd_sync()
 	do
 		name=$(module_name "$sm_path")
 		url=$(git config -f .gitmodules --get submodule."$name".url)
+		sm_path=$(git config -f .gitmodules --get submodule."$name".path)
 
 		# Possibly a url relative to parent
 		case "$url" in
 		./*|../*)
-			url=$(resolve_relative_url "$url") || exit
+			url=$(resolve_relative_url "$url" "$sm_path") || exit
 			;;
 		esac
 
-- 
1.7.10.2.647.g63fe035.dirty

Re: [RFC] submodule: fix handling of supermodules with relative origin URLs.

[Jon Seymour]

On Fri, May 18, 2012 at 10:13 PM, Jon Seymour <jon.seymour@gmail.com> wrote:
> -       remoteurl=${remoteurl%/}
> +       up_path="$(echo "$2" | sed "s/[^/]*/../g")"
> +       remoteurl=${remoteurl%/*}

Oops - didn't mean to change remoteurl here.


> +                               remoteurl="${up_path%/}/${remoteurl%/*}"

Meant up_path%/ to be up_path%/*

jon

Re: [RFC] submodule: fix handling of supermodules with relative origin URLs.

[Jens Lehmann]

Am 18.05.2012 14:13, schrieb Jon Seymour:
> Prior to this change, operations such as git submodule sync produces
> the wrong result when the origin URL of the super module
> is itself a relative URL.
> 
> The issue arises because in this case, the origin URL of the supermodule
> needs to be prepended with a prefix that navigates from the submodule to
> the supermodule.
> 
> This change adds that prefix.

Thanks, sounds sane. Please see my comments below.

> Signed-off-by: Jon Seymour <jon.seymour@gmail.com>
> ---
>  git-submodule.sh | 39 +++++++++++++++++++++++----------------
>  1 file changed, 23 insertions(+), 16 deletions(-)
> 
> If people are ok with the fix, I'll roll this as a patch together
> with some tests.

Yeah, tests would be great (and while at it please drop the trailing
'.' from the subject ;-).

This version of the patch does break some existing tests, but your
follow up suggests you already found that out yourself ;-)

> diff --git a/git-submodule.sh b/git-submodule.sh
> index 64a70d6..5008867 100755
> --- a/git-submodule.sh
> +++ b/git-submodule.sh
> @@ -37,7 +37,8 @@ resolve_relative_url ()
>  	remoteurl=$(git config "remote.$remote.url") ||
>  		remoteurl=$(pwd) # the repository is its own authoritative upstream
>  	url="$1"
> -	remoteurl=${remoteurl%/}
> +	up_path="$(echo "$2" | sed "s/[^/]*/../g")"

Me thinks up_path should be set in the case below, which is the only
place where it is used.

> +	remoteurl=${remoteurl%/*}

As you mentioned in your follow up this change is not correct.

>  	sep=/
>  	while test -n "$url"
>  	do
> @@ -45,6 +46,9 @@ resolve_relative_url ()
>  		../*)
>  			url="${url#../}"
>  			case "$remoteurl" in
> +			.*/*)

up_path should be set here.

> +				remoteurl="${up_path%/}/${remoteurl%/*}"
> +				;;
>  			*/*)
>  				remoteurl="${remoteurl%/*}"
>  				;;
> @@ -235,11 +239,24 @@ cmd_add()
>  		usage
>  	fi
>  
> +	# normalize path:
> +	# multiple //; leading ./; /./; /../; trailing /
> +	sm_path=$(printf '%s/\n' "$sm_path" |
> +		sed -e '
> +			s|//*|/|g
> +			s|^\(\./\)*||
> +			s|/\./|/|g
> +			:start
> +			s|\([^/]*\)/\.\./||
> +			tstart
> +			s|/*$||
> +		')
> +
>  	# assure repo is absolute or relative to parent
>  	case "$repo" in
>  	./*|../*)
>  		# dereference source url relative to parent's url
> -		realrepo=$(resolve_relative_url "$repo") || exit
> +		realrepo=$(resolve_relative_url "$repo" "$sm_path") || exit
>  		;;
>  	*:*|/*)
>  		# absolute url
> @@ -250,18 +267,6 @@ cmd_add()
>  	;;
>  	esac
>  
> -	# normalize path:
> -	# multiple //; leading ./; /./; /../; trailing /
> -	sm_path=$(printf '%s/\n' "$sm_path" |
> -		sed -e '
> -			s|//*|/|g
> -			s|^\(\./\)*||
> -			s|/\./|/|g
> -			:start
> -			s|\([^/]*\)/\.\./||
> -			tstart
> -			s|/*$||
> -		')
>  	git ls-files --error-unmatch "$sm_path" > /dev/null 2>&1 &&
>  	die "$(eval_gettext "'\$sm_path' already exists in the index")"
>  
> @@ -401,13 +406,14 @@ cmd_init()
>  		if test -z "$(git config "submodule.$name.url")"
>  		then
>  			url=$(git config -f .gitmodules submodule."$name".url)
> +			sm_path=$(git config -f .gitmodules submodule."$name".path)

Isn't sm_path already set correctly here? I think this line should
be dropped.

>  			test -z "$url" &&
>  			die "$(eval_gettext "No url found for submodule path '\$sm_path' in .gitmodules")"
>  
>  			# Possibly a url relative to parent
>  			case "$url" in
>  			./*|../*)
> -				url=$(resolve_relative_url "$url") || exit
> +				url=$(resolve_relative_url "$url" "$sm_path") || exit
>  				;;
>  			esac
>  			git config submodule."$name".url "$url" ||
> @@ -960,11 +966,12 @@ cmd_sync()
>  	do
>  		name=$(module_name "$sm_path")
>  		url=$(git config -f .gitmodules --get submodule."$name".url)
> +		sm_path=$(git config -f .gitmodules --get submodule."$name".path)

Same here.

>  
>  		# Possibly a url relative to parent
>  		case "$url" in
>  		./*|../*)
> -			url=$(resolve_relative_url "$url") || exit
> +			url=$(resolve_relative_url "$url" "$sm_path") || exit
>  			;;
>  		esac
>  

Other than that the patch looks fine.

Re: [RFC] submodule: fix handling of supermodules with relative origin URLs.

[Jon Seymour]

On Sat, May 19, 2012 at 5:58 AM, Jens Lehmann <Jens.Lehmann@web.de> wrote:
> This version of the patch does break some existing tests, but your
> follow up suggests you already found that out yourself ;-)
>

Actually, that was just by reading  the patch and noticing odd-ness. I
realise now that I snuck in that bogus edit after I had run the tests
which were otherwise clean - d'oh.

>
> Me thinks up_path should be set in the case below, which is the only
> place where it is used.

Ok.

>
> Isn't sm_path already set correctly here? I think this line should
> be dropped.
>

Yes, good point. Thanks.

>
> Same here.
>

Thanks.

>
> Other than that the patch looks fine.

Thanks for the review - I'll add some tests for the new behaviour.

jon.