libwebsockets: Fix the build with -Os
PATCH 1/1 libwebsockets: Fix the build with -Os
lib/event-libs/libuv/libuv.c: In function 'elops_destroy_context1_uv':lib/event-libs/libuv/libuv.c:519:7: error: 'm' may be used uninitialized in this function [-Werror=maybe-uninitialized]
if (m)
^--- .../recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb | 2 ++ 1 file changed, 2 insertions(+)
meta-oe/recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb +2 -0
@@ -28,3 +28,5 @@ EXTRA_OECMAKE += " \ PACKAGES =+ "${PN}-testapps" FILES_${PN}-testapps += "${datadir}/libwebsockets-test-server/*" + +CFLAGS_append = " -Wno-error"
Signature · 1 line
2.17.1
Quoted 16 lines · Adrian Bunk
lib/event-libs/libuv/libuv.c: In function 'elops_destroy_context1_uv':lib/event-libs/libuv/libuv.c:519:7: error: 'm' may be used uninitialized in this function [-Werror=maybe-uninitialized]if (m) ^Signed-off-by Adrian Bunk <[email protected]>--- .../recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb | 2 ++ 1 file changed, 2 insertions(+)meta-oe/recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb +0
-0
meta-oe/recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb EXTRA_OECMAKE += " \ · line 32
Quoted 5 lines
meta-oe/recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb +2
-0@@ -28,3 +28,5 @@ EXTRA_OECMAKE += " \ PACKAGES =+ "${PN}-testapps" FILES_${PN}-testapps += "${datadir}/libwebsockets-test-server/*" + +CFLAGS_append = " -Wno-error"
is it possible to fix the underlying problem? since Os is not default it definitely could be a bug in upstream but
by disabling warnings for all kind of builds we are painting with broad brush
Quoted 8 lines
--
2.17.1--
_______________________________________________
Openembedded-devel mailing list
[email protected]
http://lists.openembedded.org/mailman/listinfo/openembedded-devel
Quoted 19 lines · Khem Raj
Quoted 18 lines · Adrian Bunk
lib/event-libs/libuv/libuv.c: In function 'elops_destroy_context1_uv':lib/event-libs/libuv/libuv.c:519:7: error: 'm' may be used uninitialized in this function [-Werror=maybe-uninitialized]if (m) ^Signed-off-by Adrian Bunk <[email protected]>--- .../recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb | 2 ++ 1 file changed, 2 insertions(+)meta-oe/recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb +0
-0@@ -28,3 +28,5 @@ EXTRA_OECMAKE += " \ PACKAGES =+ "${PN}-testapps"
Quoted 7 lines
meta-oe/recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb EXTRA_OECMAKE += " \ · line 32
Quoted 3 lines
meta-oe/recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb +2
-0FILES_${PN}-testapps += "${datadir}/libwebsockets-test-server/*" + +CFLAGS_append = " -Wno-error"is it possible to fix the underlying problem? since Os is not default it definitely could be a bug in upstream but
by disabling warnings for all kind of builds we are painting with broad brush
The underlying problem is that some gcc warnings are not reliable with -Os, there are bugs open in the gcc bugzilla for that.
cu
Adrian
Signature · 4 lines
"Is there not promise of rain?" Ling Tan asked suddenly out
of the darkness. There had been need of rain for many days.
"Only a promise," Lao Er said.
Pearl S. Buck - Dragon SeedQuoted 29 lines · Adrian Bunk
Quoted 19 lines · Khem Raj
Quoted 18 lines · Adrian Bunk
lib/event-libs/libuv/libuv.c: In function 'elops_destroy_context1_uv':lib/event-libs/libuv/libuv.c:519:7: error: 'm' may be used uninitialized in this function [-Werror=maybe-uninitialized]if (m) ^Signed-off-by Adrian Bunk <[email protected]>--- .../recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb | 2 ++ 1 file changed, 2 insertions(+)meta-oe/recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb +0
-0@@ -28,3 +28,5 @@ EXTRA_OECMAKE += " \ PACKAGES =+ "${PN}-testapps"Quoted 7 lines
meta-oe/recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb EXTRA_OECMAKE += " \ · line 32
Quoted 3 lines
meta-oe/recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb +2
-0FILES_${PN}-testapps += "${datadir}/libwebsockets-test-server/*" + +CFLAGS_append = " -Wno-error"is it possible to fix the underlying problem? since Os is not default it definitely could be a bug in upstream but
by disabling warnings for all kind of builds we are painting with broad brush
Quoted 3 lines
The underlying problem is that some gcc warnings are not reliable with -Os, there are bugs open in the gcc bugzilla for that.
I am aware of that for maybe-* warnings heuristics may go wrong, but then its better to just disable that
one warning from being treated as error if thats possible to add easily something like
-Wno-error=maybe-uninitialized could do it.
Quoted 10 lines
cu
Adrian--
"Is there not promise of rain?" Ling Tan asked suddenly out
of the darkness. There had been need of rain for many days.
"Only a promise," Lao Er said.
Pearl S. Buck - Dragon Seed
Quoted 30 lines · Khem Raj
Quoted 29 lines · Adrian Bunk
Quoted 26 lines · Khem Raj
Quoted 16 lines · Adrian Bunk
lib/event-libs/libuv/libuv.c: In function 'elops_destroy_context1_uv':lib/event-libs/libuv/libuv.c:519:7: error: 'm' may be used uninitialized in this function [-Werror=maybe-uninitialized]if (m) ^Signed-off-by Adrian Bunk <[email protected]>--- .../recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb | 2 ++ 1 file changed, 2 insertions(+)meta-oe/recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb +0
-0meta-oe/recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb EXTRA_OECMAKE += " \ · line 32
Quoted 5 lines
meta-oe/recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb +2
-0@@ -28,3 +28,5 @@ EXTRA_OECMAKE += " \ PACKAGES =+ "${PN}-testapps" FILES_${PN}-testapps += "${datadir}/libwebsockets-test-server/*" + +CFLAGS_append = " -Wno-error"is it possible to fix the underlying problem? since Os is not default it definitely could be a bug in upstream but
by disabling warnings for all kind of builds we are painting with broad brush
Quoted 8 lines
Quoted 3 lines
The underlying problem is that some gcc warnings are not reliable with -Os, there are bugs open in the gcc bugzilla for that.
I am aware of that for maybe-* warnings heuristics may go wrong, but then its better to just disable that
one warning from being treated as error if thats possible to add easily something like
-Wno-error=maybe-uninitialized could do it.
And then the package fails to build due to a different warning after the next gcc upgrade.
The few packages that manually set -Werror are causing so much trouble, and seeing warnings as errors in this code from 2018 that is currently 298 commits behind upstream master won't bring actual benefits even when the warnings are not gcc bugs.
If you are interested in warnings you shouldn't have -Werror in very few packages, but check for the most problematic warnings in all packages.
E.g. -Wimplicit-function-declaration warnings are often the cause for runime crashes, and a quick grep through my build logs shows that your gettid patches to snort/lttng-tools/lttng-ust are not correct.[1]
cu
Adrian
[1] pid_t is an int in glibc, which makes the lack of prototypes harmless in this specific case.
Signature · 4 lines
"Is there not promise of rain?" Ling Tan asked suddenly out
of the darkness. There had been need of rain for many days.
"Only a promise," Lao Er said.
Pearl S. Buck - Dragon SeedQuoted 35 lines · Adrian Bunk
Quoted 33 lines · Khem Raj
Quoted 32 lines · Adrian Bunk
Quoted 19 lines · Khem Raj
Quoted 18 lines · Adrian Bunk
lib/event-libs/libuv/libuv.c: In function 'elops_destroy_context1_uv':lib/event-libs/libuv/libuv.c:519:7: error: 'm' may be used uninitialized in this function [-Werror=maybe-uninitialized]if (m) ^Signed-off-by Adrian Bunk <[email protected]>--- .../recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb | 2 ++ 1 file changed, 2 insertions(+)meta-oe/recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb +0
-0@@ -28,3 +28,5 @@ EXTRA_OECMAKE += " \ PACKAGES =+ "${PN}-testapps"Quoted 7 lines
meta-oe/recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb EXTRA_OECMAKE += " \ · line 32
Quoted 3 lines
meta-oe/recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb +2
-0FILES_${PN}-testapps += "${datadir}/libwebsockets-test-server/*" + +CFLAGS_append = " -Wno-error"is it possible to fix the underlying problem? since Os is not default it definitely could be a bug in upstream but
by disabling warnings for all kind of builds we are painting with broad brushThe underlying problem is that some gcc warnings are not reliable with -Os, there are bugs open in the gcc bugzilla for that.
Quoted 9 lines
Quoted 5 lines
I am aware of that for maybe-* warnings heuristics may go wrong, but then its better to just disable that
one warning from being treated as error if thats possible to add easily something like
-Wno-error=maybe-uninitialized could do it.And then the package fails to build due to a different warning after the next gcc upgrade.
which is fine, since we can report it upstream to either gcc if its gcc's fault or to package itself
and better still with a patch. Ideally we should use package defaults or maybe be more strict
If everyone stop using these warnings then why is compiler adding
them in the first place
here I realize lies the difference of opinion.
Quoted 5 lines
The few packages that manually set -Werror are causing so much trouble, and seeing warnings as errors in this code from 2018 that is currently 298 commits behind upstream master won't bring actual benefits even when the warnings are not gcc bugs.
Quoted 7 lines
If you are interested in warnings you shouldn't have -Werror in very few packages, but check for the most problematic warnings in all packages.
E.g. -Wimplicit-function-declaration warnings are often the cause for runime crashes, and a quick grep through my build logs shows that your gettid patches to snort/lttng-tools/lttng-ust are not correct.[1]
maybe you should explain a bit more why they not correct these patches are to define local gettid function if system libc does not provide it
there is a new syscall wrapper in glibc 2.30, older glibcs wont have it.
Quoted 13 lines
cu
Adrian[1]
pid_tis an int in glibc, which makes the lack of prototypes harmless in this specific case.--
"Is there not promise of rain?" Ling Tan asked suddenly out
of the darkness. There had been need of rain for many days.
"Only a promise," Lao Er said.
Pearl S. Buck - Dragon Seed
Quoted 42 lines · Khem Raj
Quoted 41 lines · Adrian Bunk
Quoted 38 lines · Khem Raj
Quoted 29 lines · Adrian Bunk
Quoted 19 lines · Khem Raj
Quoted 18 lines · Adrian Bunk
lib/event-libs/libuv/libuv.c: In function 'elops_destroy_context1_uv':lib/event-libs/libuv/libuv.c:519:7: error: 'm' may be used uninitialized in this function [-Werror=maybe-uninitialized]if (m) ^Signed-off-by Adrian Bunk <[email protected]>--- .../recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb | 2 ++ 1 file changed, 2 insertions(+)meta-oe/recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb +0
-0@@ -28,3 +28,5 @@ EXTRA_OECMAKE += " \ PACKAGES =+ "${PN}-testapps"Quoted 7 lines
meta-oe/recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb EXTRA_OECMAKE += " \ · line 32
Quoted 3 lines
meta-oe/recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb +2
-0FILES_${PN}-testapps += "${datadir}/libwebsockets-test-server/*" + +CFLAGS_append = " -Wno-error"is it possible to fix the underlying problem? since Os is not default it definitely could be a bug in upstream but
by disabling warnings for all kind of builds we are painting with broad brushQuoted 3 lines
The underlying problem is that some gcc warnings are not reliable with -Os, there are bugs open in the gcc bugzilla for that.
I am aware of that for maybe-* warnings heuristics may go wrong, but then its better to just disable that
one warning from being treated as error if thats possible to add easily something like
-Wno-error=maybe-uninitialized could do it.
Quoted 6 lines
Quoted 3 lines
And then the package fails to build due to a different warning after the next gcc upgrade.
which is fine, since we can report it upstream to either gcc if its gcc's fault or to package itself
This is often a waste of time since the code in OE is often much older than upstream master.
Quoted 1 line
and better still with a patch.
But please do not add such patches to OE.
Patches from people who don't know the code well are often quite buggy, and fixing warnings then tends to add more bugs than it fixes.
Google "Debian OpenSSL disaster" for how the Debian maintainer "fixing" a Valgrind warning in the Debian OpenSSL package made private keys used for ssh authentication in Debian/Ubuntu predictable (AKA everyone on the internet could log into the affected machines).
Quoted 5 lines
Ideally we should use package defaults
or maybe be more strict
If everyone stop using these warnings then why is compiler adding
them in the first place
here I realize lies the difference of opinion.
When you are developing software these warnings are very useful.
A distribution is not developing software, it is distributing software developed by other people.
Quoted 8 lines
Quoted 8 lines
The few packages that manually set -Werror are causing so much trouble, and seeing warnings as errors in this code from 2018 that is currently 298 commits behind upstream master won't bring actual benefits even when the warnings are not gcc bugs.
If you are interested in warnings you shouldn't have -Werror in very few packages, but check for the most problematic warnings in all packages.
Quoted 8 lines
Quoted 3 lines
E.g. -Wimplicit-function-declaration warnings are often the cause for runime crashes, and a quick grep through my build logs shows that your gettid patches to snort/lttng-tools/lttng-ust are not correct.[1]
maybe you should explain a bit more why they not correct these patches are to define local gettid function if system libc does not provide it
there is a new syscall wrapper in glibc 2.30, older glibcs wont have it.
...../../../../lttng-tools-2.10.7/src/common/error.h:154:27: warning: implicit declaration of function 'getpid'; did you mean 'getpt'? [-Wimplicit-function-declaration] ../../../../lttng-tools-2.10.7/src/common/error.h:154:44: warning: implicit declaration of function 'gettid'; did you mean 'getcpu'? [-Wimplicit-function-declaration]
...
You ifdef'ed away the #include <unistd.h> that provides the prototypes for these functions in glibc 2.30.
cu
Adrian
Signature · 4 lines
"Is there not promise of rain?" Ling Tan asked suddenly out
of the darkness. There had been need of rain for many days.
"Only a promise," Lao Er said.
Pearl S. Buck - Dragon SeedQuoted 48 lines · Adrian Bunk
Quoted 46 lines · Khem Raj
Quoted 44 lines · Adrian Bunk
Quoted 30 lines · Khem Raj
Quoted 29 lines · Adrian Bunk
Quoted 26 lines · Khem Raj
Quoted 16 lines · Adrian Bunk
lib/event-libs/libuv/libuv.c: In function 'elops_destroy_context1_uv':lib/event-libs/libuv/libuv.c:519:7: error: 'm' may be used uninitialized in this function [-Werror=maybe-uninitialized]if (m) ^Signed-off-by Adrian Bunk <[email protected]>--- .../recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb | 2 ++ 1 file changed, 2 insertions(+)meta-oe/recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb +0
-0meta-oe/recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb EXTRA_OECMAKE += " \ · line 32
Quoted 5 lines
meta-oe/recipes-connectivity/libwebsockets/libwebsockets_3.1.0.bb +2
-0@@ -28,3 +28,5 @@ EXTRA_OECMAKE += " \ PACKAGES =+ "${PN}-testapps" FILES_${PN}-testapps += "${datadir}/libwebsockets-test-server/*" + +CFLAGS_append = " -Wno-error"is it possible to fix the underlying problem? since Os is not default it definitely could be a bug in upstream but
by disabling warnings for all kind of builds we are painting with broad brushQuoted 8 lines
Quoted 3 lines
The underlying problem is that some gcc warnings are not reliable with -Os, there are bugs open in the gcc bugzilla for that.
I am aware of that for maybe-* warnings heuristics may go wrong, but then its better to just disable that
one warning from being treated as error if thats possible to add easily something like
-Wno-error=maybe-uninitialized could do it.And then the package fails to build due to a different warning after the next gcc upgrade.
Quoted 6 lines
Quoted 2 lines
which is fine, since we can report it upstream to either gcc if its gcc's fault or to package itself
This is often a waste of time since the code in OE is often much older than upstream master.
which is partly because we dont have conducive workflow and tooling for devs to upstream the patches or
consider building master branch of upstream repos easily. Maybe some thought for future enhancements where
we can have a develop mode for packages which we can use for doing this kind of work.
Quoted 4 lines
Quoted 1 line
and better still with a patch.
But please do not add such patches to OE.
Quoted 8 lines
Patches from people who don't know the code well are often quite buggy, and fixing warnings then tends to add more bugs than it fixes.
Google "Debian OpenSSL disaster" for how the Debian maintainer "fixing" a Valgrind warning in the Debian OpenSSL package made private keys used for ssh authentication in Debian/Ubuntu predictable (AKA everyone on the internet could log into the affected machines).
right I remember that, but then I also know first-hand cases where the compiler was telling you all the way and it was
ignored which ended up in field bugs, so there is no right answer.
Quoted 11 lines
Quoted 5 lines
Ideally we should use package defaults
or maybe be more strict
If everyone stop using these warnings then why is compiler adding
them in the first place
here I realize lies the difference of opinion.When you are developing software these warnings are very useful.
A distribution is not developing software, it is distributing software developed by other people.
we also provide development environment along with distro building tools so its in interest for us to enable
not only packagers but also package developers
Quoted 35 lines
Quoted 8 lines
Quoted 8 lines
The few packages that manually set -Werror are causing so much trouble, and seeing warnings as errors in this code from 2018 that is currently 298 commits behind upstream master won't bring actual benefits even when the warnings are not gcc bugs.
If you are interested in warnings you shouldn't have -Werror in very few packages, but check for the most problematic warnings in all packages.
Quoted 8 lines
Quoted 3 lines
E.g. -Wimplicit-function-declaration warnings are often the cause for runime crashes, and a quick grep through my build logs shows that your gettid patches to snort/lttng-tools/lttng-ust are not correct.[1]
maybe you should explain a bit more why they not correct these patches are to define local gettid function if system libc does not provide it
there is a new syscall wrapper in glibc 2.30, older glibcs wont have it....
../../../../lttng-tools-2.10.7/src/common/error.h:154:27: warning: implicit declaration of function 'getpid'; did you mean 'getpt'? [-Wimplicit-function-declaration]../../../../lttng-tools-2.10.7/src/common/error.h:154:44: warning: implicit declaration of function 'gettid'; did you mean 'getcpu'? [-Wimplicit-function-declaration]
...You ifdef'ed away the #include <
unistd.h> that provides the prototypes for these functions in glibc 2.30.cu
Adrian--
"Is there not promise of rain?" Ling Tan asked suddenly out
of the darkness. There had been need of rain for many days.
"Only a promise," Lao Er said.
Pearl S. Buck - Dragon Seed
Quoted 15 lines · Khem Raj
Quoted 1 line · Adrian Bunk
On Thu, Aug 29, 2019 at 01:51:05PM -0700, Khem Raj wrote:
...
Quoted 11 lines
Quoted 1 line
and better still with a patch.
But please do not add such patches to OE.
Patches from people who don't know the code well are often quite buggy, and fixing warnings then tends to add more bugs than it fixes.
Google "Debian OpenSSL disaster" for how the Debian maintainer "fixing" a Valgrind warning in the Debian OpenSSL package made private keys used for ssh authentication in Debian/Ubuntu predictable (AKA everyone on the internet could log into the affected machines).
Quoted 4 lines
right I remember that, but then I also know first-hand cases where the compiler was telling you all the way and it was
ignored which ended up in field bugs, so there is no right answer.
...
That's a lesson for upstream, not so much for a distribution.
The worst case is when people are just doing whatever is the fastest code "fix" to silence a warning/error.
When the compiler is telling that the C library does not support FNM_EXTMATCH, then ignoring the error with
#define FNM_EXTMATCH 0
can turn it into a field bug.
Ignoring the compile error when the C library does not support qsort_r by using qsort instead can create exactly the runtime race conditions qsort_r is designed to avoid.
Finding correct solutions can be hard and time-consuming, especially when the person doing the change does not know the code in question well.
But few correct fixes are better than many quick fixes that might introduce more bugs than they fix.
And there is also a blame game involved:
If upstream software contains bugs, the blame goes to upstream. If distribution patches introduce bugs, the blame goes to the distribution.
Heartbleed was even worse than the above mentioned bug, but noone could blame Debian for it.
cu
Adrian
Signature · 4 lines
"Is there not promise of rain?" Ling Tan asked suddenly out
of the darkness. There had been need of rain for many days.
"Only a promise," Lao Er said.
Pearl S. Buck - Dragon Seed