Skip to contentSkip to conversation

builtin/gc.c: fix -Wdeclaration-after-statement

PATCH In review Ævar Arnfjörð Bjarmason 4 participants 10 messages Nov 15, 2022

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 builtin/gc.c: fix -Wdeclaration-after-statement

In 1f80129d61b (maintenance: add option to register in a specific config, 2022-11-09) code was added which triggers a "-Wdeclaration-after-statement" warning, which is on by default with DEVELOPER=1.

---

This landed in the latest "next" push-out, causing e.g this CI failure:
https://github.com/git/git/actions/runs/3467538041/jobs/5792504315

 builtin/gc.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
builtin/gc.c +1 -1
index 56b107e7f0b..22a990db0be 100644
@@ -1543,6 +1543,7 @@ static int maintenance_unregister(int argc, const char **argv, const char *prefi
 	int found = 0;
 	struct string_list_item *item;
 	const struct string_list *list;
+	struct config_set cs;
 
 	argc = parse_options(argc, argv, prefix, options,
 			     builtin_maintenance_unregister_usage, 0);
@@ -1550,7 +1551,6 @@ static int maintenance_unregister(int argc, const char **argv, const char *prefi
 		usage_with_options(builtin_maintenance_unregister_usage,
 				   options);
 
-	struct config_set cs;
 	if (config_file) {
 		git_configset_init(&cs);
 		git_configset_add_file(&cs, config_file);
Signature · 1 line
2.38.0.1473.g172bcc0511c
#1 lore raw

Hi Ævar,

Quoted 11 lines · Ævar Arnfjörð Bjarmason

In 1f80129d61b (maintenance: add option to register in a specific config, 2022-11-09) code was added which triggers a "-Wdeclaration-after-statement" warning, which is on by default with DEVELOPER=1.

---

This landed in the latest "next" push-out, causing e.g this CI failure:
https://github.com/git/git/actions/runs/3467538041/jobs/5792504315

I looked at the same thing all morning, and found that
https://lore.kernel.org/git/[email protected]/
_probably_ was designed to fix the same issue (and I think you agree that that patch should be used instead of yours because it is more complete in fixing left-over issues).

However, try as I might, I did not find out yet why it does not apply cleanly over here (I got side-tracked into range-diff'ing patches downloaded from the archive, which is quite the challenge and not even fun because range-diff requires commits, not emails, so I side-tracked myself into teaching range-diff to accept mbox arguments).

Maybe you can adjust that patch so it applies cleanly?

Ciao,
Dscho

Quoted 29 lines
 builtin/gc.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
builtin/gc.c +1 -2
index 56b107e7f0b..22a990db0be 100644
@@ -1543,6 +1543,7 @@ static int maintenance_unregister(int argc, const char **argv, const char *prefi
 	int found = 0;
 	struct string_list_item *item;
 	const struct string_list *list;
+	struct config_set cs;
 
 	argc = parse_options(argc, argv, prefix, options,
 			     builtin_maintenance_unregister_usage, 0);
@@ -1550,7 +1551,6 @@ static int maintenance_unregister(int argc, const char **argv, const char *prefi
 		usage_with_options(builtin_maintenance_unregister_usage,
 				   options);
 
-	struct config_set cs;
 	if (config_file) {
 		git_configset_init(&cs);
 		git_configset_add_file(&cs, config_file);
--

2.38.0.1473.g172bcc0511c

#2 lore raw

Hi Ævar,

Quoted 20 lines · Johannes Schindelin
Quoted 11 lines · Ævar Arnfjörð Bjarmason

In 1f80129d61b (maintenance: add option to register in a specific config, 2022-11-09) code was added which triggers a "-Wdeclaration-after-statement" warning, which is on by default with DEVELOPER=1.

---

This landed in the latest "next" push-out, causing e.g this CI failure:
https://github.com/git/git/actions/runs/3467538041/jobs/5792504315

I looked at the same thing all morning, and found that
https://lore.kernel.org/git/[email protected]/
_probably_ was designed to fix the same issue (and I think you agree that that patch should be used instead of yours because it is more complete in fixing left-over issues).

Quoted 7 lines

However, try as I might, I did not find out yet why it does not apply cleanly over here (I got side-tracked into range-diff'ing patches downloaded from the archive, which is quite the challenge and not even fun because range-diff requires commits, not emails, so I side-tracked myself into teaching range-diff to accept mbox arguments).

Maybe you can adjust that patch so it applies cleanly?

Seems that the v2 of "maintenance: improve error reporting for unregister" [*1*] was _partially_ folded into v3 of "maintenance: add option to register in a specific config": At least the config change that talks about --config-file inste of --config _is_ part of v3, and one reason that the former patch does not apply cleanly on top of the latter patch. Which is a bit funny because the former patch was sent out two days _after_ the latter patch.

And it looks as if there is one more thing that needs to be forward-ported from that no-longer-applying patch: the initialization of the configset, otherwise we're potentially trying to clear an uninitialized data structure.

This is the patch I am currently testing (on top of Git for Windows' shears/seen branch:
https://github.com/git-for-windows/git/commit/cd7b86d19f):

-- snip --

builtin/gc.c +2 -2
index 635f12499d68..a3f63880dfba 100644
@@ -1569,6 +1569,7 @@ static int maintenance_unregister(int argc, const char **argv, const char *prefi
 	int found = 0;
 	struct string_list_item *item;
 	const struct string_list *list;
+	struct config_set cs;
 
 	argc = parse_options(argc, argv, prefix, options,
 			     builtin_maintenance_unregister_usage, 0);
@@ -1576,9 +1577,8 @@ static int maintenance_unregister(int argc, const char **argv, const char *prefi
 		usage_with_options(builtin_maintenance_unregister_usage,
 				   options);
 
-	struct config_set cs;
+	git_configset_init(&cs);
 	if (config_file) {
-		git_configset_init(&cs);
 		git_configset_add_file(&cs, config_file);
 		list = git_configset_get_value_multi(&cs, key);
 	} else {

-- snap --

Ævar, can you please have a thorough look and see whether there is anything else we were missing before advancing this to next?

Thanks,
Dscho

Footnote *1*: https://lore.kernel.org/git/[email protected]/
Footnote *2*: https://lore.kernel.org/git/[email protected]/

#3 lore raw
Quoted 75 lines · Johannes Schindelin

Hi Ævar,

Quoted 20 lines · Johannes Schindelin
Quoted 11 lines · Ævar Arnfjörð Bjarmason

In 1f80129d61b (maintenance: add option to register in a specific config, 2022-11-09) code was added which triggers a "-Wdeclaration-after-statement" warning, which is on by default with DEVELOPER=1.

---

This landed in the latest "next" push-out, causing e.g this CI failure:
https://github.com/git/git/actions/runs/3467538041/jobs/5792504315

I looked at the same thing all morning, and found that
https://lore.kernel.org/git/[email protected]/
_probably_ was designed to fix the same issue (and I think you agree that that patch should be used instead of yours because it is more complete in fixing left-over issues).

Quoted 7 lines

However, try as I might, I did not find out yet why it does not apply cleanly over here (I got side-tracked into range-diff'ing patches downloaded from the archive, which is quite the challenge and not even fun because range-diff requires commits, not emails, so I side-tracked myself into teaching range-diff to accept mbox arguments).

Maybe you can adjust that patch so it applies cleanly?

Seems that the v2 of "maintenance: improve error reporting for unregister" [*1*] was _partially_ folded into v3 of "maintenance: add option to register in a specific config": At least the config change that talks about --config-file inste of --config _is_ part of v3, and one reason that the former patch does not apply cleanly on top of the latter patch. Which is a bit funny because the former patch was sent out two days _after_ the latter patch.

And it looks as if there is one more thing that needs to be forward-ported from that no-longer-applying patch: the initialization of the configset, otherwise we're potentially trying to clear an uninitialized data structure.

This is the patch I am currently testing (on top of Git for Windows' shears/seen branch:
https://github.com/git-for-windows/git/commit/cd7b86d19f):

-- snip --

builtin/gc.c +2 -3
index 635f12499d68..a3f63880dfba 100644
@@ -1569,6 +1569,7 @@ static int maintenance_unregister(int argc, const char **argv, const char *prefi
 	int found = 0;
 	struct string_list_item *item;
 	const struct string_list *list;
+	struct config_set cs;
 
 	argc = parse_options(argc, argv, prefix, options,
 			     builtin_maintenance_unregister_usage, 0);
@@ -1576,9 +1577,8 @@ static int maintenance_unregister(int argc, const char **argv, const char *prefi
 		usage_with_options(builtin_maintenance_unregister_usage,
 				   options);
 
-	struct config_set cs;
+	git_configset_init(&cs);
 	if (config_file) {
-		git_configset_init(&cs);
 		git_configset_add_file(&cs, config_file);
 		list = git_configset_get_value_multi(&cs, key);
 	} else {
-- snap --
Quoted 2 lines

Ævar, can you please have a thorough look and see whether there is anything else we were missing before advancing this to next?

Thanks, I submitted a v2 just now that addresses the unit'd issue as well, and as we're fixing that we can just reduce the scope of the variable, which makes it lifetime more obvious:

https://lore.kernel.org/git/[email protected]/

#4 lore raw

PATCH v2 1/1 maintenance --unregister: fix uninit'd data use & -Wdeclaration-after-statement

In reply to Ævar Arnfjörð Bjarmason

Since (maintenance: add option to register in a specific config, 2022-11-09) we've been unable to build with "DEVELOPER=1" without "DEVOPTS=no-error", as the added code triggers a "-Wdeclaration-after-statement" warning.

And worse than that, the data handed to git_configset_clear() is uninitialized, as can be spotted with e.g.:

	./t7900-maintenance.sh -vixd --run=23 --valgrind
	[...]
	+ git maintenance unregister --force
	Conditional jump or move depends on uninitialised value(s)
	   at 0x6B5F1E: git_configset_clear (config.c:2367)
	   by 0x4BA64E: maintenance_unregister (gc.c:1619)
	   by 0x4BD278: cmd_maintenance (gc.c:2650)
	   by 0x409905: run_builtin (git.c:466)
	   by 0x40A21C: handle_builtin (git.c:721)
	   by 0x40A58E: run_argv (git.c:788)
	   by 0x40AF68: cmd_main (git.c:926)
	   by 0x5D39FE: main (common-main.c:57)
	 Uninitialised value was created by a stack allocation
	   at 0x4BA22C: maintenance_unregister (gc.c:1557)

Let's fix both of these issues, and also move the scope of the variable to the "if" statement it's used in, to make it obvious where it's used.

---
Range-diff against v1:
1: 54d405f15f1 ! 1: f37e99c9d59 builtin/gc.c: fix -Wdeclaration-after-statement
@@ Metadata
Author: Ævar Arnfjörð Bjarmason <[email protected]>

## Commit message ##
- builtin/gc.c: fix -Wdeclaration-after-statement
+ maintenance --unregister: fix uninit'd data use & -Wdeclaration-after-statement

- In 1f80129d61b (maintenance: add option to register in a specific
- config, 2022-11-09) code was added which triggers a
- "-Wdeclaration-after-statement" warning, which is on by default with
- DEVELOPER=1.
+ Since (maintenance: add option to register in a specific config,
+ 2022-11-09) we've been unable to build with "DEVELOPER=1" without
+ "DEVOPTS=no-error", as the added code triggers a
+ "-Wdeclaration-after-statement" warning.

+ And worse than that, the data handed to git_configset_clear() is
+ uninitialized, as can be spotted with e.g.:
+
+ ./t7900-maintenance.sh -vixd --run=23 --valgrind
+ [...]
+ + git maintenance unregister --force
+ Conditional jump or move depends on uninitialised value(s)
+ at 0x6B5F1E: git_configset_clear (config.c:2367)
+ by 0x4BA64E: maintenance_unregister (gc.c:1619)
+ by 0x4BD278: cmd_maintenance (gc.c:2650)
+ by 0x409905: run_builtin (git.c:466)
+ by 0x40A21C: handle_builtin (git.c:721)
+ by 0x40A58E: run_argv (git.c:788)
+ by 0x40AF68: cmd_main (git.c:926)
+ by 0x5D39FE: main (common-main.c:57)
+ Uninitialised value was created by a stack allocation
+ at 0x4BA22C: maintenance_unregister (gc.c:1557)
+
+ Let's fix both of these issues, and also move the scope of the
+ variable to the "if" statement it's used in, to make it obvious where
+ it's used.
+
+ Helped-by: Johannes Schindelin <[email protected]>

## builtin/gc.c ##
-@@ builtin/gc.c: static int maintenance_unregister(int argc, const char **argv, const char *prefi
- int found = 0;
- struct string_list_item *item;
- const struct string_list *list;
-+ struct config_set cs;
-
- argc = parse_options(argc, argv, prefix, options,
- builtin_maintenance_unregister_usage, 0);
@@ builtin/gc.c: static int maintenance_unregister(int argc, const char **argv, const char *prefi
usage_with_options(builtin_maintenance_unregister_usage,
options);

- struct config_set cs;
if (config_file) {
++ struct config_set cs;
++
git_configset_init(&cs);
git_configset_add_file(&cs, config_file);
+ list = git_configset_get_value_multi(&cs, key);
++ git_configset_clear(&cs);
+ } else {
+ list = git_config_get_value_multi(key);
+ }
+@@ builtin/gc.c: static int maintenance_unregister(int argc, const char **argv, const char *prefi
+ die(_("repository '%s' is not registered"), maintpath);
+ }
+
+- git_configset_clear(&cs);
+ free(maintpath);
+ return 0;
+ }

 builtin/gc.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
builtin/gc.c +3 -2
index 56b107e7f0b..d87cf84041f 100644
@@ -1550,11 +1550,13 @@ static int maintenance_unregister(int argc, const char **argv, const char *prefi
 		usage_with_options(builtin_maintenance_unregister_usage,
 				   options);
 
-	struct config_set cs;
 	if (config_file) {
+		struct config_set cs;
+
 		git_configset_init(&cs);
 		git_configset_add_file(&cs, config_file);
 		list = git_configset_get_value_multi(&cs, key);
+		git_configset_clear(&cs);
 	} else {
 		list = git_config_get_value_multi(key);
 	}
@@ -1590,7 +1592,6 @@ static int maintenance_unregister(int argc, const char **argv, const char *prefi
 		die(_("repository '%s' is not registered"), maintpath);
 	}
 
-	git_configset_clear(&cs);
 	free(maintpath);
 	return 0;
 }
Signature · 1 line
2.38.0.1473.g172bcc0511c
#5 lore raw

Re: [PATCH v2] maintenance --unregister: fix uninit'd data use & -Wdeclaration-after-statement

Quoted 7 lines · Ævar Arnfjörð Bjarmason

Since (maintenance: add option to register in a specific config, 2022-11-09) we've been unable to build with "DEVELOPER=1" without "DEVOPTS=no-error", as the added code triggers a "-Wdeclaration-after-statement" warning.

And worse than that, the data handed to git_configset_clear() is uninitialized, as can be spotted with e.g.:

Sorry, I'm a little confused. I had sent v1, v2, and v3 of the "maintenance: add option [...]" patch, but as I understand it v2 had already been applied so I was asked to resubmit the changes from v3 rebased on next.

I had done that in [1], but these issues were caught in review so I submitted a v2 of that correction in [2] which declares the configsset earlier and unconditionally initializes it is cleared. Are these further issues discovered after [2] was applied, or was there some issue rebasing the patches?

[1] https://lore.kernel.org/git/[email protected]/
[2] https://lore.kernel.org/git/[email protected]/

#6 lore raw

Re: [PATCH v2] maintenance --unregister: fix uninit'd data use & -Wdeclaration-after-statement

Quoted 8 lines · On Tue, Nov 15, 2022 at 04:32:03PM +0000, [email protected] wrote:

I had done that in [1], but these issues were caught in review so I submitted a v2 of that correction in [2] which declares the configsset earlier and unconditionally initializes it is cleared. Are these further issues discovered after [2] was applied, or was there some issue rebasing the patches?

[1] https://lore.kernel.org/git/[email protected]/
[2] https://lore.kernel.org/git/[email protected]/

It was human error on my part. Please let me know if the result in 'next' now is OK to you.

Thanks,
Taylor

#7 lore raw

Re: [PATCH v2] maintenance --unregister: fix uninit'd data use & -Wdeclaration-after-statement

Quoted 2 lines · Taylor Blau

It was human error on my part. Please let me know if the result in 'next' now is OK to you.

Had a look and I'm just wondering if we're sure it's alright to clear the configset before using the list we found in it. Doesn't that list point into the configset?

#8 lore raw

Re: [PATCH v2] maintenance --unregister: fix uninit'd data use & -Wdeclaration-after-statement

Quoted 8 lines · On Tue, Nov 15, 2022 at 06:00:45PM +0000, [email protected] wrote:
Quoted 2 lines · Taylor Blau

It was human error on my part. Please let me know if the result in 'next' now is OK to you.

Had a look and I'm just wondering if we're sure it's alright to clear the configset before using the list we found in it. Doesn't that list point into the configset?

It does. I should not have trusted the incoming patch so blindly. See:

    https://lore.kernel.org/git/2cbead254b77cb02d219bca8f628dc4362c045b0.1668538355.git.me@ttaylorr.com/

for a fixup on top.

Thanks,
Taylor

#9 lore raw

Re: [PATCH v2] maintenance --unregister: fix uninit'd data use & -Wdeclaration-after-statement

In reply to Ævar Arnfjörð Bjarmason
Quoted 23 lines · Ævar Arnfjörð Bjarmason

Since (maintenance: add option to register in a specific config, 2022-11-09) we've been unable to build with "DEVELOPER=1" without "DEVOPTS=no-error", as the added code triggers a "-Wdeclaration-after-statement" warning.

And worse than that, the data handed to git_configset_clear() is uninitialized, as can be spotted with e.g.:

	./t7900-maintenance.sh -vixd --run=23 --valgrind
	[...]
	+ git maintenance unregister --force
	Conditional jump or move depends on uninitialised value(s)
	   at 0x6B5F1E: git_configset_clear (config.c:2367)
	   by 0x4BA64E: maintenance_unregister (gc.c:1619)
	   by 0x4BD278: cmd_maintenance (gc.c:2650)
	   by 0x409905: run_builtin (git.c:466)
	   by 0x40A21C: handle_builtin (git.c:721)
	   by 0x40A58E: run_argv (git.c:788)
	   by 0x40AF68: cmd_main (git.c:926)
	   by 0x5D39FE: main (common-main.c:57)
	 Uninitialised value was created by a stack allocation
	   at 0x4BA22C: maintenance_unregister (gc.c:1557)
Quoted 6 lines

Let's fix both of these issues, and also move the scope of the variable to the "if" statement it's used in, to make it obvious where it's used.

Thanks, both. I could have sworn I picked up the right version of this topic when queuing, but apparently not.

I pushed this out to 'next' and rebuilt jch and seen based on the new version. The result now should be OK. Thanks again for noticing and working together.

Thanks,
Taylor