[PATCH] Teach format-patch to handle output directory relative to cwd

Without any explicit -o parameter, we correctly avoided putting the
resulting patch output to the toplevel.  We should do the same when
the user gave a relative pathname to be consistent with this case and also
with how other commands handle relative pathnames.

Noticed by Cesar Eduardo Barros.

Signed-off-by: Junio C Hamano <redacted>
---
Cesar Eduardo Barros [off-list ref] writes:

If you are in a subdirectory of your working copy (for instance,
linux-2.6/drivers/net) and use git-format-patch with -o to a sequence
of ../ (for instance, -o ../../../) to write to the working copy's
parent directory, it instead interprets the directory passed to -o as
relative to the root of the working copy, instead of the expected
current directory.

 builtin-log.c           |   13 ++++++++++-
 t/t4014-format-patch.sh |   52 ++++++++++++++++++++++++++++++++++++++++++++++-
 2 files changed, 62 insertions(+), 3 deletions(-)

diff --git c/builtin-log.c w/builtin-log.c
index 4a02ee9..3e404ae 100644
--- c/builtin-log.c
+++ w/builtin-log.c

@@ -731,6 +731,15 @@ static const char *clean_message_id(const char *msg_id)
 	return xmemdupz(a, z - a);
 }
 
+static const char *set_outdir(const char *prefix, const char *output_directory)
+{
+	if (!output_directory)
+		output_directory = ".";
+	if (prefix)
+		output_directory = xstrdup(prefix_filename(prefix, strlen(prefix), output_directory));
+	return output_directory;
+}
+
 int cmd_format_patch(int argc, const char **argv, const char *prefix)
 {
 	struct commit *commit;

@@ -908,8 +917,8 @@ int cmd_format_patch(int argc, const char **argv, const char *prefix)
 	if (!DIFF_OPT_TST(&rev.diffopt, TEXT) && !no_binary_diff)
 		DIFF_OPT_SET(&rev.diffopt, BINARY);
 
-	if (!output_directory && !use_stdout)
-		output_directory = prefix;
+	if (!use_stdout)
+		output_directory = set_outdir(prefix, output_directory);
 
 	if (output_directory) {
 		if (use_stdout)

diff --git c/t/t4014-format-patch.sh w/t/t4014-format-patch.sh
index 9d99dc2..5a9a63d 100755
--- c/t/t4014-format-patch.sh
+++ w/t/t4014-format-patch.sh

@@ -3,7 +3,7 @@
 # Copyright (c) 2006 Junio C Hamano
 #
 
-test_description='Format-patch skipping already incorporated patches'
+test_description='various format-patch tests'
 
 . ./test-lib.sh
 

@@ -255,4 +255,54 @@ test_expect_success 'format-patch respects -U' '
 
 '
 
+test_expect_success 'format-patch from a subdirectory (1)' '
+	filename=$(
+		rm -rf sub &&
+		mkdir -p sub/dir &&
+		cd sub/dir &&
+		git format-patch -1
+	) &&
+	case "$filename" in
+	sub/dir/*)
+		;; # ok
+	*)
+		echo "Oops? $filename"
+		false
+		;;
+	esac &&
+	test -f "$filename"
+'
+
+test_expect_success 'format-patch from a subdirectory (2)' '
+	filename=$(
+		rm -rf sub &&
+		mkdir -p sub/dir &&
+		cd sub/dir &&
+		git format-patch -1 -o ..
+	) &&
+	case "$filename" in
+	sub/dir/../0*)
+		;; # ok
+	*)
+		echo "Oops? $filename"
+		false
+		;;
+	esac &&
+	basename=$(expr "$filename" : ".*/\(.*\)") &&
+	test -f "sub/$basename"
+'
+
+test_expect_success 'format-patch from a subdirectory (3)' '
+	here="$TEST_DIRECTORY/$test" &&
+	rm -f 0* &&
+	filename=$(
+		rm -rf sub &&
+		mkdir -p sub/dir &&
+		cd sub/dir &&
+		git format-patch -1 -o "$here"
+	) &&
+	basename=$(expr "$filename" : ".*/\(.*\)") &&
+	test -f "$basename"
+'
+
 test_done

Junio C Hamano escreveu:

Without any explicit -o parameter, we correctly avoided putting the
resulting patch output to the toplevel.  We should do the same when
the user gave a relative pathname to be consistent with this case and also
with how other commands handle relative pathnames.

Works great, only the resulting output to the screen is a bit 
ugly/confusing:

drivers/net/../../../0001-sc92031-more-useful-banner-in-kernel-log.patch
drivers/net/../../../0002-sc92031-remove-meaningless-version-string.patch
drivers/net/../../../0003-sc92031-inline-SC92031_DESCRIPTION.patch
drivers/net/../../../0004-sc92031-use-device-id-directly-instead-of-made-up-n.patch
drivers/net/../../../0005-sc92031-add-a-link-to-the-datasheet.patch

I would expect:

../../../0001-sc92031-more-useful-banner-in-kernel-log.patch
../../../0002-sc92031-remove-meaningless-version-string.patch
../../../0003-sc92031-inline-SC92031_DESCRIPTION.patch
../../../0004-sc92031-use-device-id-directly-instead-of-made-up-n.patch
../../../0005-sc92031-add-a-link-to-the-datasheet.patch

(after all, I am still inside the drivers/net directory)

-- 
Cesar Eduardo Barros
cesarb@cesarb.net
cesar.barros@gmail.com