Skip to contentSkip to conversation

kbuild, LLVMLinux: Don't suppress format warnings

PATCH In review [email protected] 4 participants 11 messages Feb 1, 2019 – Feb 7, 2019 Kbuild

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 kbuild, LLVMLinux: Don't suppress format warnings

From: Jon Flatley <[email protected]>

gcc produces format warnings that clang suppresses. To keep behavior consistent between gcc and clang, don't suppress format warnings in clang.

---
 scripts/Makefile.extrawarn | 1 -
 1 file changed, 1 deletion(-)
scripts/Makefile.extrawarn +0 -1
index 768306add591..94daf4e1fc73 100644
@@ -68,7 +68,6 @@ else
 ifdef CONFIG_CC_IS_CLANG
 KBUILD_CFLAGS += $(call cc-disable-warning, initializer-overrides)
 KBUILD_CFLAGS += $(call cc-disable-warning, unused-value)
-KBUILD_CFLAGS += $(call cc-disable-warning, format)
 KBUILD_CFLAGS += $(call cc-disable-warning, sign-compare)
 KBUILD_CFLAGS += $(call cc-disable-warning, format-zero-length)
 KBUILD_CFLAGS += $(call cc-disable-warning, uninitialized)
Signature · 1 line
2.20.1.611.gfbb209baf1-goog
#1 lore raw
Quoted 9 lines · On Sat, Feb 2, 2019 at 6:10 AM <[email protected]> wrote:

From: Jon Flatley <[email protected]>

gcc produces format warnings that clang suppresses. To keep behavior consistent between gcc and clang, don't suppress format warnings in clang.

---

Applied to linux-kbuild.
Thanks.

Quoted 18 lines
 scripts/Makefile.extrawarn | 1 -
 1 file changed, 1 deletion(-)
scripts/Makefile.extrawarn +0 -2
index 768306add591..94daf4e1fc73 100644
@@ -68,7 +68,6 @@ else
 ifdef CONFIG_CC_IS_CLANG
 KBUILD_CFLAGS += $(call cc-disable-warning, initializer-overrides)
 KBUILD_CFLAGS += $(call cc-disable-warning, unused-value)
-KBUILD_CFLAGS += $(call cc-disable-warning, format)
 KBUILD_CFLAGS += $(call cc-disable-warning, sign-compare)
 KBUILD_CFLAGS += $(call cc-disable-warning, format-zero-length)
 KBUILD_CFLAGS += $(call cc-disable-warning, uninitialized)
--

2.20.1.611.gfbb209baf1-goog

Signature · 2 lines
Best Regards
Masahiro Yamada
#2 lore raw
Quoted 8 lines · Masahiro Yamada
Quoted 7 lines · On Sat, Feb 2, 2019 at 6:10 AM <[email protected]> wrote:

From: Jon Flatley <[email protected]>

gcc produces format warnings that clang suppresses. To keep behavior consistent between gcc and clang, don't suppress format warnings in clang.

Quoted 7 lines
Quoted 2 lines

---

Applied to linux-kbuild.
Thanks.

Hi Jon and Masahiro,

Just as a heads up, this introduces a ton of warnings (duh). Isn't the typical plan behind turning on warnings that were disabled to build with 'W=', fix them all, then turn them on so as not to pollute the build?

Log file: https://gist.github.com/443db156e56cd3c0f6b21d9d77728d80

Note a big chunk of them come from one scnprintf call in include/linux/usb/wusb.h but still, there are many other warnings that make quite a bit of noise. Some seem relatively easy to fix, which I suppose I will try to tackle soon.

Thanks,
Nathan

Quoted 25 lines
Quoted 20 lines

 scripts/Makefile.extrawarn | 1 -
 1 file changed, 1 deletion(-)
scripts/Makefile.extrawarn +0 -2
index 768306add591..94daf4e1fc73 100644
@@ -68,7 +68,6 @@ else
 ifdef CONFIG_CC_IS_CLANG
 KBUILD_CFLAGS += $(call cc-disable-warning, initializer-overrides)
 KBUILD_CFLAGS += $(call cc-disable-warning, unused-value)
-KBUILD_CFLAGS += $(call cc-disable-warning, format)
 KBUILD_CFLAGS += $(call cc-disable-warning, sign-compare)
 KBUILD_CFLAGS += $(call cc-disable-warning, format-zero-length)
 KBUILD_CFLAGS += $(call cc-disable-warning, uninitialized)
--

2.20.1.611.gfbb209baf1-goog

Signature · 2 lines
Best Regards
Masahiro Yamada
#3 lore raw
Quoted 26 lines · Nathan Chancellor
Quoted 8 lines · Masahiro Yamada
Quoted 7 lines · On Sat, Feb 2, 2019 at 6:10 AM <[email protected]> wrote:

From: Jon Flatley <[email protected]>

gcc produces format warnings that clang suppresses. To keep behavior consistent between gcc and clang, don't suppress format warnings in clang.

Quoted 7 lines
Quoted 2 lines

---

Applied to linux-kbuild.
Thanks.

Hi Jon and Masahiro,

Just as a heads up, this introduces a ton of warnings (duh). Isn't the typical plan behind turning on warnings that were disabled to build with 'W=', fix them all, then turn them on so as not to pollute the build?

Log file: https://gist.github.com/443db156e56cd3c0f6b21d9d77728d80

Quoted 8 lines

Note a big chunk of them come from one scnprintf call in include/linux/usb/wusb.h but still, there are many other warnings that make quite a bit of noise. Some seem relatively easy to fix, which I suppose I will try to tackle soon.

Thanks,
Nathan

Hi Nathan,

This was definitely not my intention.
I noticed the added warnings this morning and was considering calling for a revert on this patch.

The intent was to match the behavior of gcc, as it has -Wformat enabled. It was rather naive of me to assume the behavior of -Wformat would be the same in both gcc and clang.
Indeed, it seems gcc is more permissive about what format substitutions it allows.

For example passing int to the "%hu" format specifier is fine in gcc under -Wformat but produces a warning in clang.
Maybe this was the motivation for adding -Wno-format to clang in the first place.
This difference is puzzling to me, and I wonder if it's by design.

Considering the whole point of this patch was to sync up this behavior between gcc and clang I am OK with reverting this.

Thanks,
Jon

Quoted 25 lines
Quoted 25 lines
Quoted 18 lines

 scripts/Makefile.extrawarn | 1 -
 1 file changed, 1 deletion(-)
scripts/Makefile.extrawarn +0 -2
index 768306add591..94daf4e1fc73 100644
@@ -68,7 +68,6 @@ else
 ifdef CONFIG_CC_IS_CLANG
 KBUILD_CFLAGS += $(call cc-disable-warning, initializer-overrides)
 KBUILD_CFLAGS += $(call cc-disable-warning, unused-value)
-KBUILD_CFLAGS += $(call cc-disable-warning, format)
 KBUILD_CFLAGS += $(call cc-disable-warning, sign-compare)
 KBUILD_CFLAGS += $(call cc-disable-warning, format-zero-length)
 KBUILD_CFLAGS += $(call cc-disable-warning, uninitialized)
--
Quoted 2 lines

2.20.1.611.gfbb209baf1-goog

--
Best Regards
Masahiro Yamada

#4 lore raw
Quoted 21 lines · Jon Flatley
Quoted 18 lines · Nathan Chancellor
Quoted 15 lines · Masahiro Yamada
Quoted 9 lines · On Sat, Feb 2, 2019 at 6:10 AM <[email protected]> wrote:

From: Jon Flatley <[email protected]>

gcc produces format warnings that clang suppresses. To keep behavior consistent between gcc and clang, don't suppress format warnings in clang.

---

Applied to linux-kbuild.
Thanks.

Quoted 7 lines
Quoted 7 lines

Hi Jon and Masahiro,

Just as a heads up, this introduces a ton of warnings (duh). Isn't the typical plan behind turning on warnings that were disabled to build with 'W=', fix them all, then turn them on so as not to pollute the build?

Log file: https://gist.github.com/443db156e56cd3c0f6b21d9d77728d80

Oh boy, that's a lot. Too many to fix quickly IMO.

Quoted 22 lines
Quoted 9 lines

Note a big chunk of them come from one scnprintf call in include/linux/usb/wusb.h but still, there are many other warnings that make quite a bit of noise. Some seem relatively easy to fix, which I suppose I will try to tackle soon.

Thanks,
Nathan

Hi Nathan,

This was definitely not my intention.
I noticed the added warnings this morning and was considering calling for a revert on this patch.

The intent was to match the behavior of gcc, as it has -Wformat enabled. It was rather naive of me to assume the behavior of -Wformat would be the same in both gcc and clang.
Indeed, it seems gcc is more permissive about what format substitutions it allows.

Quoted 4 lines

For example passing int to the "%hu" format specifier is fine in gcc under -Wformat but produces a warning in clang.
Maybe this was the motivation for adding -Wno-format to clang in the first place.

Sorry, I'm late to this thread. What is it reverting; who authored the original patch? Was it [email protected]?

Quoted 1 line

This difference is puzzling to me, and I wonder if it's by design.

Probably; internally let's sync up with the Clang devs to understand this difference more.

Quoted 3 lines

Considering the whole point of this patch was to sync up this behavior between gcc and clang I am OK with reverting this.

Is this patch in -next, or has it already hit mainline? I think it's better to revert, then start upstreaming fixes, then re-land it once we're warning free.

Signature · 2 lines
Thanks,
~Nick Desaulniers
#5 lore raw
Quoted 24 lines · Nick Desaulniers
Quoted 23 lines · Jon Flatley
Quoted 20 lines · Nathan Chancellor
Quoted 8 lines · Masahiro Yamada
Quoted 7 lines · On Sat, Feb 2, 2019 at 6:10 AM <[email protected]> wrote:

From: Jon Flatley <[email protected]>

gcc produces format warnings that clang suppresses. To keep behavior consistent between gcc and clang, don't suppress format warnings in clang.

Quoted 7 lines
Quoted 2 lines

---

Applied to linux-kbuild.
Thanks.

Hi Jon and Masahiro,

Quoted 30 lines
Quoted 5 lines
Quoted 5 lines

Just as a heads up, this introduces a ton of warnings (duh). Isn't the typical plan behind turning on warnings that were disabled to build with 'W=', fix them all, then turn them on so as not to pollute the build?

Log file: https://gist.github.com/443db156e56cd3c0f6b21d9d77728d80

Oh boy, that's a lot. Too many to fix quickly IMO.

Quoted 22 lines
Quoted 9 lines

Note a big chunk of them come from one scnprintf call in include/linux/usb/wusb.h but still, there are many other warnings that make quite a bit of noise. Some seem relatively easy to fix, which I suppose I will try to tackle soon.

Thanks,
Nathan

Hi Nathan,

This was definitely not my intention.
I noticed the added warnings this morning and was considering calling for a revert on this patch.

The intent was to match the behavior of gcc, as it has -Wformat enabled. It was rather naive of me to assume the behavior of -Wformat would be the same in both gcc and clang.
Indeed, it seems gcc is more permissive about what format substitutions it allows.

My guess is that it has something to do with how the compilers internally handle certain specifier promotions (GCC probably just silently ignores the 'h' part of the specifier whereas Clang warns) but I didn't do any actual research into the matter. Probably should before looking into all of the warnings :)

Quoted 8 lines
Quoted 4 lines

For example passing int to the "%hu" format specifier is fine in gcc under -Wformat but produces a warning in clang.
Maybe this was the motivation for adding -Wno-format to clang in the first place.

Sorry, I'm late to this thread. What is it reverting; who authored the original patch? Was it [email protected]?

This patch is turning on '-Wformat' for Clang, which was disabled in commit 3d3d6b847420 ("kbuild: LLVMLinux: Adapt warnings for compilation with clang").

Quoted 5 lines
Quoted 1 line

This difference is puzzling to me, and I wonder if it's by design.

Probably; internally let's sync up with the Clang devs to understand this difference more.

Yes, I do think it would be a good idea to turn this on eventually but it'd be wise to understand why Clang emits a warning but GCC doesn't.

Quoted 8 lines
Quoted 3 lines

Considering the whole point of this patch was to sync up this behavior between gcc and clang I am OK with reverting this.

Is this patch in -next, or has it already hit mainline? I think it's better to revert, then start upstreaming fixes, then re-land it once we're warning free.

It's in linux-kbuild/kbuild, it hasn't hit -next yet.

Nathan

Quoted 3 lines
Signature · 2 lines
Thanks,
~Nick Desaulniers
#6 lore raw
Quoted 89 lines · Nathan Chancellor
Quoted 24 lines · Nick Desaulniers
Quoted 23 lines · Jon Flatley
Quoted 20 lines · Nathan Chancellor
Quoted 8 lines · Masahiro Yamada
Quoted 7 lines · On Sat, Feb 2, 2019 at 6:10 AM <[email protected]> wrote:

From: Jon Flatley <[email protected]>

gcc produces format warnings that clang suppresses. To keep behavior consistent between gcc and clang, don't suppress format warnings in clang.

Quoted 7 lines
Quoted 2 lines

---

Applied to linux-kbuild.
Thanks.

Hi Jon and Masahiro,

Quoted 30 lines
Quoted 5 lines
Quoted 5 lines

Just as a heads up, this introduces a ton of warnings (duh). Isn't the typical plan behind turning on warnings that were disabled to build with 'W=', fix them all, then turn them on so as not to pollute the build?

Log file: https://gist.github.com/443db156e56cd3c0f6b21d9d77728d80

Oh boy, that's a lot. Too many to fix quickly IMO.

Quoted 22 lines
Quoted 9 lines

Note a big chunk of them come from one scnprintf call in include/linux/usb/wusb.h but still, there are many other warnings that make quite a bit of noise. Some seem relatively easy to fix, which I suppose I will try to tackle soon.

Thanks,
Nathan

Hi Nathan,

This was definitely not my intention.
I noticed the added warnings this morning and was considering calling for a revert on this patch.

The intent was to match the behavior of gcc, as it has -Wformat enabled. It was rather naive of me to assume the behavior of -Wformat would be the same in both gcc and clang.
Indeed, it seems gcc is more permissive about what format substitutions it allows.

My guess is that it has something to do with how the compilers internally handle certain specifier promotions (GCC probably just silently ignores the 'h' part of the specifier whereas Clang warns) but I didn't do any actual research into the matter. Probably should before looking into all of the warnings :)

Quoted 8 lines
Quoted 4 lines

For example passing int to the "%hu" format specifier is fine in gcc under -Wformat but produces a warning in clang.
Maybe this was the motivation for adding -Wno-format to clang in the first place.

Sorry, I'm late to this thread. What is it reverting; who authored the original patch? Was it [email protected]?

This patch is turning on '-Wformat' for Clang, which was disabled in commit 3d3d6b847420 ("kbuild: LLVMLinux: Adapt warnings for compilation with clang").

Quoted 5 lines
Quoted 1 line

This difference is puzzling to me, and I wonder if it's by design.

Probably; internally let's sync up with the Clang devs to understand this difference more.

Yes, I do think it would be a good idea to turn this on eventually but it'd be wise to understand why Clang emits a warning but GCC doesn't.

Quoted 4 lines
Quoted 3 lines

Considering the whole point of this patch was to sync up this behavior between gcc and clang I am OK with reverting this.

Quoted 6 lines
Quoted 4 lines

Is this patch in -next, or has it already hit mainline? I think it's better to revert, then start upstreaming fixes, then re-land it once we're warning free.

It's in linux-kbuild/kbuild, it hasn't hit -next yet.

Right, it is just staying in my branch.
I will drop it.

I should have tested this carefully. Sorry.

Signature · 2 lines
Best Regards
Masahiro Yamada
#7 lore raw
Quoted 97 lines · Masahiro Yamada
Quoted 94 lines · Nathan Chancellor
Quoted 24 lines · Nick Desaulniers
Quoted 23 lines · Jon Flatley
Quoted 20 lines · Nathan Chancellor
Quoted 8 lines · Masahiro Yamada
Quoted 7 lines · On Sat, Feb 2, 2019 at 6:10 AM <[email protected]> wrote:

From: Jon Flatley <[email protected]>

gcc produces format warnings that clang suppresses. To keep behavior consistent between gcc and clang, don't suppress format warnings in clang.

Quoted 7 lines
Quoted 2 lines

---

Applied to linux-kbuild.
Thanks.

Hi Jon and Masahiro,

Quoted 30 lines
Quoted 5 lines
Quoted 5 lines

Just as a heads up, this introduces a ton of warnings (duh). Isn't the typical plan behind turning on warnings that were disabled to build with 'W=', fix them all, then turn them on so as not to pollute the build?

Log file: https://gist.github.com/443db156e56cd3c0f6b21d9d77728d80

Oh boy, that's a lot. Too many to fix quickly IMO.

Quoted 22 lines
Quoted 9 lines

Note a big chunk of them come from one scnprintf call in include/linux/usb/wusb.h but still, there are many other warnings that make quite a bit of noise. Some seem relatively easy to fix, which I suppose I will try to tackle soon.

Thanks,
Nathan

Hi Nathan,

This was definitely not my intention.
I noticed the added warnings this morning and was considering calling for a revert on this patch.

The intent was to match the behavior of gcc, as it has -Wformat enabled. It was rather naive of me to assume the behavior of -Wformat would be the same in both gcc and clang.
Indeed, it seems gcc is more permissive about what format substitutions it allows.

My guess is that it has something to do with how the compilers internally handle certain specifier promotions (GCC probably just silently ignores the 'h' part of the specifier whereas Clang warns) but I didn't do any actual research into the matter. Probably should before looking into all of the warnings :)

Quoted 8 lines
Quoted 4 lines

For example passing int to the "%hu" format specifier is fine in gcc under -Wformat but produces a warning in clang.
Maybe this was the motivation for adding -Wno-format to clang in the first place.

Sorry, I'm late to this thread. What is it reverting; who authored the original patch? Was it [email protected]?

This patch is turning on '-Wformat' for Clang, which was disabled in commit 3d3d6b847420 ("kbuild: LLVMLinux: Adapt warnings for compilation with clang").

Quoted 5 lines
Quoted 1 line

This difference is puzzling to me, and I wonder if it's by design.

Probably; internally let's sync up with the Clang devs to understand this difference more.

Yes, I do think it would be a good idea to turn this on eventually but it'd be wise to understand why Clang emits a warning but GCC doesn't.

Quoted 8 lines
Quoted 3 lines

Considering the whole point of this patch was to sync up this behavior between gcc and clang I am OK with reverting this.

Is this patch in -next, or has it already hit mainline? I think it's better to revert, then start upstreaming fixes, then re-land it once we're warning free.

Quoted 8 lines
Quoted 1 line

It's in linux-kbuild/kbuild, it hasn't hit -next yet.

Right, it is just staying in my branch.
I will drop it.

I should have tested this carefully. Sorry.

BTW, it would be nice if kbuild test robot
could test building with clang ...

Signature · 2 lines
Best Regards
Masahiro Yamada
#8 lore raw
Quoted 2 lines · Masahiro Yamada

BTW, it would be nice if kbuild test robot
could test building with clang ...

Is "kbuild test robot" different than 0day, kernelCI, or ClangBuiltLinux's CI?

Signature · 2 lines
Thanks,
~Nick Desaulniers
#9 lore raw
Quoted 9 lines · Nick Desaulniers
Quoted 2 lines · Masahiro Yamada

BTW, it would be nice if kbuild test robot
could test building with clang ...

Is "kbuild test robot" different than 0day, kernelCI, or ClangBuiltLinux's CI?

Signature · 2 lines
Thanks,
~Nick Desaulniers

kbuild test robot is 0day (that's the name the emails come under when reporting issues).

Nathan

#10 lore raw
Quoted 7 lines · Nathan Chancellor
Quoted 5 lines · Nick Desaulniers
Quoted 2 lines · Masahiro Yamada

BTW, it would be nice if kbuild test robot
could test building with clang ...

Quoted 7 lines
Quoted 4 lines

Is "kbuild test robot" different than 0day, kernelCI, or ClangBuiltLinux's CI? --
Thanks,
~Nick Desaulniers

kbuild test robot is 0day (that's the name the emails come under when reporting issues).

Ah, ok, I've email the 0day team about a year ago. Once we land the asm goto patches and get everything working again on x86_64, then I'll contact them again.

Signature · 2 lines
Thanks,
~Nick Desaulniers