Re: [PATCH v6] Add an explicit GIT_DIR to the list of excludes

Am 12.06.2014 01:28, schrieb Pasha Bolokhov:

+test_expect_success "setup" '
+	mkdir repo-inside/ &&
+	(
+		cd repo-inside/ &&
+		for f in a b c d
+		do
+			echo "DATA" >"$f" || exit 1
+		done &&
+		mkdir dir1 dir1/meta &&
+		mkdir dir1/ssubdir dir1/ssubdir/meta &&
+		for f in e f g h
+		do
+			echo "MORE DATA" >"dir1/$f" || exit 1
+		done &&
+		echo "EVEN more Data" >dir1/meta/aa &&
+		echo "Data and BAIT" >dir1/ssubdir/meta/aaa &&
+		mkdir dir2
+		echo "Not a Metadata File" >dir2/meta

&&-chain broken twice.

We already use 'mkdir -p' elsewhere; you can use it here as well to
contract several mkdir invocations.

+		git --git-dir=meta init
+	) &&
+	mkdir repo-outside/ repo-outside/external repo-outside/tree &&
+	(
+		cd repo-outside/tree &&
+		for f in n o p q
+		do
+			echo "Literal Data" >"$f" || exit 1
+		done &&
+		mkdir meta sub sub/meta &&
+		echo "Sample data" >meta/bb &&
+		echo "Stream of data" >sub/meta/bbb &&
+		git --git-dir=../external/meta init
+	)
+'
+
+
+#
+# The first set of tests (the repository is inside the work tree)
+#
+test_expect_success "'git status' ignores the repository directory" '
+	(
+		cd repo-inside &&
+		git --git-dir=meta --work-tree=. status --porcelain --untracked=all |
+			grep meta | sort >status.actual.2 &&

Please do not place a git invocation in a pipline such that it is not
the last command; its exist status would be ignored:

		git --git-dir=meta --work-tree=. status --porcelain --untracked=all

status.actual.2+ &&

		grep meta status.actual.2+ | sort >status.actual.2 &&

There are more cases like this later in the patch.

+		cat >status.expect.2 <<-\EOF &&
+		?? dir1/meta/aa
+		?? dir1/ssubdir/meta/aaa
+		?? dir2/meta
+		EOF
+		test_cmp status.actual.2 status.expect.2

It is customary to call the files 'expect' and 'actual'. Furthermore,
swap the order so that in case of a failure the diff shows how the
actual text was changed from the expected text:

		test_cmp status.expect.2 status.actual.2

+	)
+'
+
+test_expect_success "'git add -A' ignores the repository directory" '
+	(
+		cd repo-inside &&
+		git --git-dir=meta --work-tree=. add -A &&
+		git --git-dir=meta --work-tree=. status --porcelain | grep meta | sort >status.actual.3 &&
+		cat >status.expect.3 <<-\EOF &&
+		A  dir1/meta/aa
+		A  dir1/ssubdir/meta/aaa
+		A  dir2/meta
+		EOF
+		test_cmp status.actual.3 status.expect.3
+	)
+'
+
+test_expect_success "'git grep --exclude-standard' ignores the repository directory" '
+	(
+		cd repo-inside &&
+		test_might_fail git --git-dir=meta \
+			grep --no-index --exclude-standard BAIT >grep.actual.4 &&
+		cat >grep.expect.4 <<-\EOF &&
+		dir1/ssubdir/meta/aaa:Data and BAIT
+		EOF
+		test_cmp grep.actual.4 grep.expect.4
+	)
+'
+
+#
+# The second set of tests (the repository is outside of the work tree)
+#
+test_expect_success "'git status' acknowledges directories 'meta' \
+if repo is not within work tree" '
+	test_might_fail rm -rf meta/ &&

How might this fail? Only if permissions are wrong, and then we do want
this to fail.

Moreover, test_*_fail helpers are intended to be used only with git
commands; we don't expect system commands to fail in unexpected ways.

+	(
+		cd repo-outside/tree &&
+		git --git-dir=../external/meta init &&
+		git --git-dir=../external/meta --work-tree=. status --porcelain --untracked=all |
+			grep meta | sort >status.actual.5 &&
+		cat >status.expect.5 <<-\EOF &&
+		?? meta/bb
+		?? sub/meta/bbb
+		EOF
+		test_cmp status.actual.5 status.expect.5
+	)
+'
+
+test_expect_success "'git add -A' adds 'meta' if the repo is outside the work tree" '
+	(
+		cd repo-outside/tree &&
+		git --git-dir=../external/meta --work-tree=. add -A &&
+		git --git-dir=../external/meta --work-tree=. status --porcelain --untracked=all |
+			grep meta | sort >status.actual.6 &&
+		cat >status.expect.6 <<-\EOF &&
+		A  meta/bb
+		A  sub/meta/bbb
+		EOF
+		test_cmp status.actual.6 status.expect.6
+	)
+'
+
+test_done