Skip to contentSkip to conversation

stash: disable literal treatment when passing top pathspec

PATCH In review Kyle Meyer 4 participants 7 messages Apr 8, 2022 – Apr 11, 2022

lore
Share & tools
lore

Kyle Meyer

author
#0 lore raw

PATCH 1/1 stash: disable literal treatment when passing top pathspec

do_push_stash() passes ":/" as the pathspec to two subprocess calls. When pathspecs are interpreted literally for the main process, these subprocess calls do not behave as intended:

* the 'git clean' call, triggered by --include-untracked, does not remove untracked files from the working tree

* the 'git checkout' call, triggered by --keep-index, fails with a message about ":/" not matching any known files, and the main command exits with a non-zero status

Fix both of these spots by passing --no-literal-pathspecs to the subprocess commands.

---
 builtin/stash.c                    | 5 ++++-
 t/t3903-stash.sh                   | 5 +++++
 t/t3905-stash-include-untracked.sh | 5 +++++
 3 files changed, 14 insertions(+), 1 deletion(-)
builtin/stash.c +4 -1
index 0c7b6a9588..afc8400c5d 100644
@@ -1529,7 +1529,8 @@ static int do_push_stash(const struct pathspec *ps, const char *stash_msg, int q
 					     GIT_WORK_TREE_ENVIRONMENT,
 					     the_repository->worktree);
 			}
-			strvec_pushl(&cp.args, "clean", "--force",
+			strvec_pushl(&cp.args, "--no-literal-pathspecs",
+				     "clean", "--force",
 				     "--quiet", "-d", ":/", NULL);
 			if (include_untracked == INCLUDE_ALL_FILES)
 				strvec_push(&cp.args, "-x");
@@ -1592,6 +1593,8 @@ static int do_push_stash(const struct pathspec *ps, const char *stash_msg, int q
 			struct child_process cp = CHILD_PROCESS_INIT;
 
 			cp.git_cmd = 1;
+			if (!ps->nr)
+				strvec_push(&cp.args, "--no-literal-pathspecs");
 			strvec_pushl(&cp.args, "checkout", "--no-overlay",
 				     oid_to_hex(&info.i_tree), "--", NULL);
 			if (!ps->nr)
t/t3903-stash.sh +5 -0
index 4abbc8fcca..f85c3a06cb 100755
@@ -1383,6 +1383,11 @@ test_expect_success 'stash --keep-index with file deleted in index does not resu
 	test_path_is_missing to-remove
 '
 
+test_expect_success 'stash --keep-index succeeds with --literal-pathspecs' '
+	echo modified >file &&
+	git --literal-pathspecs stash --keep-index
+'
+
 test_expect_success 'stash apply should succeed with unmodified file' '
 	echo base >file &&
 	git add file &&
t/t3905-stash-include-untracked.sh +5 -0
index 5390eec4e3..2f216274b2 100755
@@ -427,5 +427,10 @@ test_expect_success 'stash -u ignores sub-repository' '
 	git init sub-repo &&
 	git stash -u
 '
+test_expect_success 'stash -u works with --literal-pathspecs' '
+	>untracked &&
+	git --literal-pathspecs stash -u &&
+	test_path_is_missing untracked
+'
 
 test_done

base-commit: bf23fe5c37d62f37267d31d4afa1a1444f70cdac

Signature · 1 line
2.34.0
#1 lore raw

t/t3905-stash-include-untracked.sh test_expect_success 'stash -u ignores sub-repository' ' · line 434

Quoted 5 lines · Kyle Meyer
t/t3905-stash-include-untracked.sh +5 -0
+test_expect_success 'stash -u works with --literal-pathspecs' '
+	>untracked &&
+	git --literal-pathspecs stash -u &&
+	test_path_is_missing untracked
+'

Why not "touch untracked" instead?

Signature · 1 line
An old man doll... just what I always wanted! - Clara
#2 lore raw
Quoted 8 lines · Bagas Sanjaya

t/t3905-stash-include-untracked.sh test_expect_success 'stash -u ignores sub-repository' ' · line 434

Quoted 5 lines · Kyle Meyer
t/t3905-stash-include-untracked.sh +5 -0
+test_expect_success 'stash -u works with --literal-pathspecs' '
+	>untracked &&
+	git --literal-pathspecs stash -u &&
+	test_path_is_missing untracked
+'

Why not "touch untracked" instead?

The ">" form is correct here. We use "touch" when updating the timestamp to something in particular is important, but here we're just creating an empty file.

#3 lore raw
Quoted 11 lines · Ævar Arnfjörð Bjarmason
Quoted 8 lines · Bagas Sanjaya

t/t3905-stash-include-untracked.sh test_expect_success 'stash -u ignores sub-repository' ' · line 434

Quoted 5 lines · Kyle Meyer
t/t3905-stash-include-untracked.sh +5 -0
+test_expect_success 'stash -u works with --literal-pathspecs' '
+	>untracked &&
+	git --literal-pathspecs stash -u &&
+	test_path_is_missing untracked
+'

Why not "touch untracked" instead?

Quoted 3 lines

The ">" form is correct here. We use "touch" when updating the timestamp to something in particular is important, but here we're just creating an empty file.

Maybe it is better to have it in CodingGuidelines or t/README?

Thanks.

#4 lore raw
In reply to Kyle Meyer
Quoted 7 lines · Kyle Meyer

do_push_stash() passes ":/" as the pathspec to two subprocess calls. When pathspecs are interpreted literally for the main process, these subprocess calls do not behave as intended:

* the 'git clean' call, triggered by --include-untracked, does not remove untracked files from the working tree

Quoted 6 lines

* the 'git checkout' call, triggered by --keep-index, fails with a message about ":/" not matching any known files, and the main command exits with a non-zero status

Fix both of these spots by passing --no-literal-pathspecs to the subprocess commands.

Yuck (to the original problem, not to the proposed solution).

I wonder if stopping to use ":/" (or using "." instead, if we need to give _some_ pathspec) is a better approach. Don't we move to the top of the working tree by the time cmd_stash() is called and whatever subprocess we spawn via run_command() interface will start at the top anyway, no?

Quoted 63 lines
---
 builtin/stash.c                    | 5 ++++-
 t/t3903-stash.sh                   | 5 +++++
 t/t3905-stash-include-untracked.sh | 5 +++++
 3 files changed, 14 insertions(+), 1 deletion(-)
builtin/stash.c +4 -1
index 0c7b6a9588..afc8400c5d 100644
@@ -1529,7 +1529,8 @@ static int do_push_stash(const struct pathspec *ps, const char *stash_msg, int q
 					     GIT_WORK_TREE_ENVIRONMENT,
 					     the_repository->worktree);
 			}
-			strvec_pushl(&cp.args, "clean", "--force",
+			strvec_pushl(&cp.args, "--no-literal-pathspecs",
+				     "clean", "--force",
 				     "--quiet", "-d", ":/", NULL);
 			if (include_untracked == INCLUDE_ALL_FILES)
 				strvec_push(&cp.args, "-x");
@@ -1592,6 +1593,8 @@ static int do_push_stash(const struct pathspec *ps, const char *stash_msg, int q
 			struct child_process cp = CHILD_PROCESS_INIT;
 
 			cp.git_cmd = 1;
+			if (!ps->nr)
+				strvec_push(&cp.args, "--no-literal-pathspecs");
 			strvec_pushl(&cp.args, "checkout", "--no-overlay",
 				     oid_to_hex(&info.i_tree), "--", NULL);
 			if (!ps->nr)
t/t3903-stash.sh +5 -0
index 4abbc8fcca..f85c3a06cb 100755
@@ -1383,6 +1383,11 @@ test_expect_success 'stash --keep-index with file deleted in index does not resu
 	test_path_is_missing to-remove
 '
 
+test_expect_success 'stash --keep-index succeeds with --literal-pathspecs' '
+	echo modified >file &&
+	git --literal-pathspecs stash --keep-index
+'
+
 test_expect_success 'stash apply should succeed with unmodified file' '
 	echo base >file &&
 	git add file &&
t/t3905-stash-include-untracked.sh +5 -0
index 5390eec4e3..2f216274b2 100755
@@ -427,5 +427,10 @@ test_expect_success 'stash -u ignores sub-repository' '
 	git init sub-repo &&
 	git stash -u
 '
+test_expect_success 'stash -u works with --literal-pathspecs' '
+	>untracked &&
+	git --literal-pathspecs stash -u &&
+	test_path_is_missing untracked
+'
 
 test_done

base-commit: bf23fe5c37d62f37267d31d4afa1a1444f70cdac

author
#5 lore raw
Quoted 1 line · Junio C Hamano

Kyle Meyer <[email protected]> writes:

[...]

Quoted 10 lines
Quoted 9 lines

* the 'git clean' call, triggered by --include-untracked, does not remove untracked files from the working tree

* the 'git checkout' call, triggered by --keep-index, fails with a message about ":/" not matching any known files, and the main command exits with a non-zero status

Fix both of these spots by passing --no-literal-pathspecs to the subprocess commands.

Quoted 7 lines

Yuck (to the original problem, not to the proposed solution).

I wonder if stopping to use ":/" (or using "." instead, if we need to give _some_ pathspec) is a better approach. Don't we move to the top of the working tree by the time cmd_stash() is called and whatever subprocess we spawn via run_command() interface will start at the top anyway, no?

For the --keep-index/checkout case, yes, it looks like the command starts from the top-level. Passing "." as the pathspec to checkout works fine, as far as I can tell.

However, for --include-untracked/clean case, the subprocess directory is set to startup_info->original_cwd since 0fce211ccc (stash: do not attempt to remove startup_info->original_cwd, 2021-12-09).

#6 lore raw
Quoted 3 lines · Kyle Meyer

However, for --include-untracked/clean case, the subprocess directory is set to startup_info->original_cwd since 0fce211ccc (stash: do not attempt to remove startup_info->original_cwd, 2021-12-09).

Interesting. I find the logic there a bit convoluted. IIUC, it goes like this:

- we do not want to lose the directory our process was originally in, which is recorded in startup_info->original_cmd.

- we have gone up to the root of the working tree, and running "clean" from there is what we want---even if we started "git stash" from a subdirectory, we want to make the entire working tree clean, not just inside our subdirectory.

- but we came up with a hack that allows us to skip removing the directory the Git process started at. To take advantage of the mechanism, we'd need to start from that original_cmd.

- but then "clean" run from that subdirectory normally cleans only that subdirectory, which is not what we want to do. To work it around, we'd need to pass :/ pathspec to say that we are cleaning from the top.

It makes me suspect that "we protect current directory" is a too specialized way that didn't really consider the possibility that we sometimes spawn a subcommand. Even "we protect this directory" may not be sufficient and we may need a "we protect these directories", I suspect. When the user originally starts "git foo" in one directory, which may have to run "git bar" in another directory, and "git bar" would want to protect the directory it starts in and also where "git foo" started from, no? It almost makes me suspect that we'd want some "git" wide option that allows us to pass a list of paths not to rmdir, whose default value is ["."], or something.

Elijah, thoughts?

But as a short-term fix, I think "--no-literal-pathspecs" is fine for this code path.

Thanks.