rng-tools: ignore incompatible-pointer-types errors for now
PATCH v2 1/11 rng-tools: ignore incompatible-pointer-types errors for now
From: Martin Jansa <[email protected]>
* causes libargp test to fail:
http://errors.yoctoproject.org/Errors/Details/766951/
| configure: error: in '/OE/build/luneos-styhead/tmp-glibc/work/qemux86_64-webos-linux/rng-tools/6.16/build':
| configure: error: libargp disabled and libc does not have argp
config.log shows:
configure:8424: x86_64-webos-linux-gcc -m64 -march=nehalem -mtune=generic -mfpmath=sse -msse4.2 --sysroot=/OE/build/luneos-styhead/tmp-glibc/work/qemux86_64-webos-linux/rng-tools/6.16/recipe-sysroot -o conftest -O2 -pipe -g -feliminate-unused-debug-types -fcanon-prefix-map -fmacro-prefix-map=/OE/build/luneos-styhead/tmp-glibc/work/qemux86_64-webos-linux/rng-tools/6.16/git=/usr/src/debug/rng-tools/6.16 -fdebug-prefix-map=/OE/build/luneos-styhead/tmp-glibc/work/qemux86_64-webos-linux/rng-tools/6.16/git=/usr/src/debug/rng-tools/6.16 -fmacro-prefix-map=/OE/build/luneos-styhead/tmp-glibc/work/qemux86_64-webos-linux/rng-tools/6.16/build=/usr/src/debug/rng-tools/6.16 -fdebug-prefix-map=/OE/build/luneos-styhead/tmp-glibc/work/qemux86_64-webos-linux/rng-tools/6.16/build=/usr/src/debug/rng-tools/6.16 -fdebug-prefix-map=/OE/build/luneos-styhead/tmp-glibc/work/qemux86_64-webos-linux/rng-tools/6.16/recipe-sysroot= -fmacro-prefix-map=/OE/build/luneos-styhead/tmp-glibc/work/qemux86_64-webos-linux/rng-tools/6.16/recipe-sysroot= -fdebug-prefix-map=/OE/build/luneos-styhead/tmp-glibc/work/qemux86_64-webos-linux/rng-tools/6.16/recipe-sysroot-native= -DJENT_CONF_ENABLE_INTERNAL_TIMER -Wl,-O1 -Wl,--hash-style=gnu -Wl,--as-needed -fcanon-prefix-map -fmacro-prefix-map=/OE/build/luneos-styhead/tmp-glibc/work/qemux86_64-webos-linux/rng-tools/6.16/git=/usr/src/debug/rng-tools/6.16 -fdebug-prefix-map=/OE/build/luneos-styhead/tmp-glibc/work/qemux86_64-webos-linux/rng-tools/6.16/git=/usr/src/debug/rng-tools/6.16 -fmacro-prefix-map=/OE/build/luneos-styhead/tmp-glibc/work/qemux86_64-webos-linux/rng-tools/6.16/build=/usr/src/debug/rng-tools/6.16 -fdebug-prefix-map=/OE/build/luneos-styhead/tmp-glibc/work/qemux86_64-webos-linux/rng-tools/6.16/build=/usr/src/debug/rng-tools/6.16 -fdebug-prefix-map=/OE/build/luneos-styhead/tmp-glibc/work/qemux86_64-webos-linux/rng-tools/6.16/recipe-sysroot= -fmacro-prefix-map=/OE/build/luneos-styhead/tmp-glibc/work/qemux86_64-webos-linux/rng-tools/6.16/recipe-sysroot= -fdebug-prefix-map=/OE/build/luneos-styhead/tmp-glibc/work/qemux86_64-webos-linux/rng-tools/6.16/recipe-sysroot-native= conftest.c -ljitterentropy >&5conftest.c: In function 'main':
conftest.c:51:52: error: passing argument 3 of 'argp_parse' from incompatible pointer type [-Wincompatible-pointer-types]
51 | int argc=1; char *argv={"test"}; argp_parse(0,argc,argv,0,0,0); return 0;
| ^~~~
| |
| char *
In file included from conftest.c:47:
/OE/build/luneos-styhead/tmp-glibc/work/qemux86_64-webos-linux/rng-tools/6.16/recipe-sysroot/usr/include/argp.h:371:58: note: expected 'char ** restrict' but argument is of type 'char *'
371 | int __argc, char **__restrict __argv,
| ~~~~~~~~~~~~~~~~~~^~~~~~--- meta/recipes-support/rng-tools/rng-tools_6.16.bb | 4 ++++ 1 file changed, 4 insertions(+)
meta/recipes-support/rng-tools/rng-tools_6.16.bb +4 -0
@@ -67,3 +67,7 @@ do_install:append() { ${D}${systemd_system_unitdir}/rng-tools.service fi } + +# libargp detection fails +# http://errors.yoctoproject.org/Errors/Details/766951/ +CFLAGS += "-Wno-error=incompatible-pointer-types"
Signature · 1 line
2.45.0
PATCH v2 2/11 expect: ignore various issues now fatal with gcc-14
From: Martin Jansa <[email protected]>
--- meta/recipes-devtools/expect/expect_5.45.4.bb | 4 ++++ 1 file changed, 4 insertions(+)
meta/recipes-devtools/expect/expect_5.45.4.bb +4 -0
@@ -81,3 +81,7 @@ FILES:${PN} += "${libdir}/libexpect${PV}.so \ " BBCLASSEXTEND = "native nativesdk" + +# http://errors.yoctoproject.org/Errors/Details/766950/ +# expect5.45.4/exp_chan.c:62:5: error: initialization of 'struct Tcl_ChannelTypeVersion_ *' from incompatible pointer type 'int (*)(void *, int)' [-Wincompatible-pointer-types] +CFLAGS += "-Wno-error=incompatible-pointer-types"
Signature · 1 line
2.45.0
PATCH v2 3/11 libunwind: ignore various issues now fatal with gcc-14
In reply to Khem RajFrom: Martin Jansa <[email protected]>
http://gecko.lge.com:8000/Errors/Details/821679
--- meta/recipes-support/libunwind/libunwind_1.6.2.bb | 8 ++++++++ 1 file changed, 8 insertions(+)
meta/recipes-support/libunwind/libunwind_1.6.2.bb +8 -0
@@ -40,3 +40,11 @@ do_install:append () { } BBCLASSEXTEND = "native" + +# http://errors.yoctoproject.org/Errors/Build/183144/ +# libunwind-1.6.2/include/tdep-aarch64/libunwind_i.h:123:47: error: passing argument 1 of '_ULaarch64_uc_addr' from incompatible pointer type [-Wincompatible-pointer-types] +# libunwind-1.6.2/src/aarch64/Ginit.c:348:28: error: initialization of 'unw_tdep_context_t *' from incompatible pointer type 'ucontext_t *' [-Wincompatible-pointer-types] +# libunwind-1.6.2/src/aarch64/Ginit.c:377:28: error: initialization of 'unw_tdep_context_t *' from incompatible pointer type 'ucontext_t *' [-Wincompatible-pointer-types] +# libunwind-1.6.2/src/aarch64/Ginit_local.c:51:9: error: assignment to 'ucontext_t *' from incompatible pointer type 'unw_context_t *' {aka 'unw_tdep_context_t *'} [-Wincompatible-pointer-types] +# libunwind-1.6.2/src/aarch64/Gresume.c:37:28: error: initialization of 'unw_tdep_context_t *' from incompatible pointer type 'ucontext_t *' [-Wincompatible-pointer-types] +CFLAGS += "-Wno-error=incompatible-pointer-types"
Signature · 1 line
2.45.0
PATCH v2 4/11 zip lrzsz connman-gnome libfm: ignore various issues fatal with gcc-14
In reply to Khem RajFrom: Martin Jansa <[email protected]>
work arounds for:
oe-core/meta/recipes-bsp/lrzsz/lrzsz_0.12.20.bb:do_compile oe-core/meta/recipes-connectivity/connman/connman-gnome_0.7.bb:do_compile oe-core/meta/recipes-support/libfm/libfm_1.3.2.bb:do_compile oe-core/meta/recipes-extended/zip/zip_3.0.bb:do_compile
http://errors.yoctoproject.org/Errors/Build/183127/
More fixes on ML (especially for -native with gcc-14 on host)
zip: https://lists.openembedded.org/g/openembedded-core/message/198783
cdrtools: https://lists.openembedded.org/g/openembedded-core/message/198899
syslinux: https://lists.openembedded.org/g/openembedded-core/message/198901
--- meta/recipes-bsp/lrzsz/lrzsz_0.12.20.bb | 6 ++++++ meta/recipes-connectivity/connman/connman-gnome_0.7.bb | 4 ++++ meta/recipes-extended/zip/zip_3.0.bb | 5 +++++ meta/recipes-support/libfm/libfm_1.3.2.bb | 4 ++++ 4 files changed, 19 insertions(+)
meta/recipes-bsp/lrzsz/lrzsz_0.12.20.bb +6 -0
@@ -46,3 +46,9 @@ ALTERNATIVE_TARGET[rb] = "${bindir}/lrz" ALTERNATIVE_TARGET[sz] = "${bindir}/lsz" ALTERNATIVE_TARGET[sx] = "${bindir}/lsz" ALTERNATIVE_TARGET[sb] = "${bindir}/lsz" + +# http://errors.yoctoproject.org/Errors/Details/766929/ +# lrzsz-0.12.20/src/tcp.c:75:56: error: passing argument 3 of 'getsockname' from incompatible pointer type [-Wincompatible-pointer-types] +# lrzsz-0.12.20/src/tcp.c:83:52: error: passing argument 3 of 'getsockname' from incompatible pointer type [-Wincompatible-pointer-types] +# lrzsz-0.12.20/src/tcp.c:103:51: error: passing argument 3 of 'accept' from incompatible pointer type [-Wincompatible-pointer-types] +CFLAGS += "-Wno-error=incompatible-pointer-types"
meta/recipes-connectivity/connman/connman-gnome_0.7.bb +4 -0
@@ -28,3 +28,7 @@ RDEPENDS:${PN} = "connman" do_install:append() { install -m 0644 ${UNPACKDIR}/images/* ${D}/usr/share/icons/hicolor/22x22/apps/ } + +# http://errors.yoctoproject.org/Errors/Details/766926/ +# connman-client.c:200:15: error: assignment to 'GtkTreeModel *' {aka 'struct _GtkTreeModel *'} from incompatible pointer type 'GtkTreeStore *' {aka 'struct _GtkTreeStore *'} [-Wincompatible-pointer-types] +CFLAGS += "-Wno-error=incompatible-pointer-types"
meta/recipes-extended/zip/zip_3.0.bb +5 -0
@@ -53,3 +53,8 @@ BBCLASSEXTEND = "native" # exclude version 2.3.2 which triggers a false positive UPSTREAM_CHECK_REGEX = "^zip(?P<pver>(?!232).+)\.tgz" + +# http://errors.yoctoproject.org/Errors/Details/766953/ +# unix/unix.c:84:51: error: passing argument 4 of 'fread' from incompatible pointer type [-Wincompatible-pointer-types] +CFLAGS += "-Wno-error=incompatible-pointer-types" +# proper fix on ML: https://lists.openembedded.org/g/openembedded-core/message/198783
meta/recipes-support/libfm/libfm_1.3.2.bb +4 -0
@@ -53,3 +53,7 @@ do_install:append () { rm -f ${D}${libdir}/libfm-extra.a rm -f ${D}${libdir}/libfm-extra.la } + +# http://errors.yoctoproject.org/Errors/Details/766924/ +# libfm-1.3.2/src/actions/action.c:2050:25: error: assignment to 'gchar **' {aka 'char **'} from incompatible pointer type 'const gchar * const*' {aka 'const char * const*'} [-Wincompatible-pointer-types] +CFLAGS += "-Wno-error=incompatible-pointer-types"
Signature · 1 line
2.45.0
Re: [PATCH v2 04/11] zip lrzsz connman-gnome libfm: ignore various issues fatal with gcc-14
Quoted 19 lines · Khem Raj
From: Martin Jansa <[email protected]>
work arounds for:
oe-core/meta/recipes-bsp/lrzsz/lrzsz_0.12.20.bb:do_compile oe-core/meta/recipes-connectivity/connman/connman-gnome_0.7.bb:do_compile oe-core/meta/recipes-support/libfm/libfm_1.3.2.bb:do_compile oe-core/meta/recipes-extended/zip/zip_3.0.bb:do_compilehttp://errors.yoctoproject.org/Errors/Build/183127/
More fixes on ML (especially for -native with gcc-14 on host)
zip: https://lists.openembedded.org/g/openembedded-core/message/198783
cdrtools: https://lists.openembedded.org/g/openembedded-core/message/198899
syslinux: https://lists.openembedded.org/g/openembedded-core/message/198901Signed-off-by Martin Jansa <[email protected]>Signed-off-by Khem Raj <[email protected]>---
...
meta/recipes-extended/zip/zip_3.0.bb BBCLASSEXTEND = "native" · line 60
Quoted 13 lines
meta/recipes-extended/zip/zip_3.0.bb +5
-0@@ -53,3 +53,8 @@ BBCLASSEXTEND = "native" # exclude version 2.3.2 which triggers a false positive UPSTREAM_CHECK_REGEX = "^zip(?P<pver>(?!232).+)\.tgz" + +# http://errors.yoctoproject.org/Errors/Details/766953/ +# unix/unix.c:84:51: error: passing argument 4 of 'fread' from incompatible pointer type [-Wincompatible-pointer-types] +CFLAGS += "-Wno-error=incompatible-pointer-types" +# proper fix on ML: https://lists.openembedded.org/g/openembedded-core/message/198783
I think we should take the proper fix from ML instead of this work around (I just haven't tested it yet).
Re: [PATCH v2 04/11] zip lrzsz connman-gnome libfm: ignore various issues fatal with gcc-14
Quoted 28 lines · Martin Jansa
Quoted 19 lines · Khem Raj
From: Martin Jansa <[email protected]>
work arounds for:
oe-core/meta/recipes-bsp/lrzsz/lrzsz_0.12.20.bb:do_compile oe-core/meta/recipes-connectivity/connman/connman-gnome_0.7.bb:do_compile oe-core/meta/recipes-support/libfm/libfm_1.3.2.bb:do_compile oe-core/meta/recipes-extended/zip/zip_3.0.bb:do_compilehttp://errors.yoctoproject.org/Errors/Build/183127/
More fixes on ML (especially for -native with gcc-14 on host)
zip: https://lists.openembedded.org/g/openembedded-core/message/198783
cdrtools: https://lists.openembedded.org/g/openembedded-core/message/198899
syslinux: https://lists.openembedded.org/g/openembedded-core/message/198901Signed-off-by Martin Jansa <[email protected]>Signed-off-by Khem Raj <[email protected]>---
...
Quoted 6 lines
meta/recipes-extended/zip/zip_3.0.bb +0
-0@@ -53,3 +53,8 @@ BBCLASSEXTEND = "native"
Quoted 10 lines
meta/recipes-extended/zip/zip_3.0.bb BBCLASSEXTEND = "native" · line 60
Quoted 7 lines
meta/recipes-extended/zip/zip_3.0.bb +5
-0# exclude version 2.3.2 which triggers a false positive UPSTREAM_CHECK_REGEX = "^zip(?P<pver>(?!232).+)\.tgz" + +# http://errors.yoctoproject.org/Errors/Details/766953/ +# unix/unix.c:84:51: error: passing argument 4 of 'fread' from incompatible pointer type [-Wincompatible-pointer-types] +CFLAGS += "-Wno-error=incompatible-pointer-types" +# proper fix on ML: https://lists.openembedded.org/g/openembedded-core/message/198783I think we should take the proper fix from ML instead of this work around (I just haven't tested it yet).
I have cherry-picked the fix from ML and dropped zip from workaround in my branches
PATCH v2 5/11 p11-kit: ignore various issues fatal with gcc-14 (for 32bit MACHINEs)
In reply to Khem RajFrom: Martin Jansa <[email protected]>
* these are reproducible only with 32bit MACHINEs and probably related to time64.inc
* http://errors.yoctoproject.org/Errors/Build/183154/
--- meta/recipes-support/p11-kit/p11-kit_0.25.3.bb | 10 ++++++++++ 1 file changed, 10 insertions(+)
meta/recipes-support/p11-kit/p11-kit_0.25.3.bb +10 -0
@@ -32,3 +32,13 @@ FILES:${PN} += " \ INSANE_SKIP:${PN} = "dev-so" BBCLASSEXTEND = "native nativesdk" + +# # This one is reproducible only on 32bit MACHINEs +# http://errors.yoctoproject.org/Errors/Details/766969/ +# git/p11-kit/import-object.c:223:62: error: passing argument 3 of 'p11_asn1_read' from incompatible pointer type [-Wincompatible-pointer-types] +# git/p11-kit/import-object.c:229:70: error: passing argument 3 of 'p11_asn1_read' from incompatible pointer type [-Wincompatible-pointer-types] +# git/p11-kit/import-object.c:264:78: error: passing argument 3 of 'p11_asn1_read' from incompatible pointer type [-Wincompatible-pointer-types] +# git/p11-kit/import-object.c:223:62: error: passing argument 3 of 'p11_asn1_read' from incompatible pointer type [-Wincompatible-pointer-types] +# git/p11-kit/import-object.c:229:70: error: passing argument 3 of 'p11_asn1_read' from incompatible pointer type [-Wincompatible-pointer-types] +# git/p11-kit/import-object.c:264:78: error: passing argument 3 of 'p11_asn1_read' from incompatible pointer type [-Wincompatible-pointer-types] +CFLAGS += "-Wno-error=incompatible-pointer-types"
Signature · 1 line
2.45.0
PATCH v2 6/11 pcmanfm: Disable incompatible-pointer-types warning as error
In reply to Khem Raj--- meta/recipes-sato/pcmanfm/pcmanfm_1.3.2.bb | 2 ++ 1 file changed, 2 insertions(+)
meta/recipes-sato/pcmanfm/pcmanfm_1.3.2.bb +2 -0
@@ -27,6 +27,8 @@ inherit autotools pkgconfig features_check mime-xdg REQUIRED_DISTRO_FEATURES = "x11" EXTRA_OECONF = "--with-gtk=3" +# GCC 14 finds extra incompatible pointer type warnings which are treated as errors +CFLAGS += "-Wno-error=incompatible-pointer-types" do_install:append () { install -d ${D}/${datadir}
Signature · 1 line
2.45.0
PATCH v2 7/11 kea: Remove -fvisibility-inlines-hidden from C++ flags
In reply to Khem RajThis fixes build with gcc-14, where default visibility is extended to inline functions and getAll() function now falls into this category and functions are marked hidden resulting in linking errors
Fixes
/OE/build/oe-core/tmp-glibc/work/core2-64-oe-linux/kea/2.5.8/recipe-sysroot-native/usr/bin/x86_64-oe-linux/../../libexec/x86_64-oe-linux/gcc/x86_64-oe-linux/14.0.1/ld: ./.libs/libdhcp4.a(dhcp4_srv.o): in function `isc::dhcp::Dhcpv4Srv::appendRequestedVendorOptions(isc::dhcp::Dhcpv4Exchange&)':
/usr/src/debug/kea/2.5.8/src/bin/dhcp4/dhcp4_srv.cc:2356:(.text+0xaac2): undefined reference to `isc::dhcp::CfgOption::getAll(unsigned int) const'
/OE/build/oe-core/tmp-glibc/work/core2-64-oe-linux/kea/2.5.8/recipe-sysroot-native/usr/bin/x86_64-oe-linux/../../libexec/x86_64-oe-linux/gcc/x86_64-oe-linux/14.0.1/ld: ./.libs/libdhcp4.a(dhcp4_srv.o): in function `isc::dhcp::OptionDescriptor isc::dhcp::CfgOption::get<unsigned int>(unsigned int const&, unsigned short) const':
/usr/src/debug/kea/2.5.8/src/lib/dhcpsrv/cfg_option.h:609:(.text+0xb288): undefined reference to `isc::dhcp::CfgOption::getAll(unsigned int) const'
/OE/build/oe-core/tmp-glibc/work/core2-64-oe-linux/kea/2.5.8/recipe-sysroot-native/usr/bin/x86_64-oe-linux/../../libexec/x86_64-oe-linux/gcc/x86_64-oe-linux/14.0.1/ld: ./.libs/libdhcp4.a(dhcp4_srv.o): in function `isc::dhcp::Dhcpv4Srv::appendRequestedOptions(isc::dhcp::Dhcpv4Exchange&)':
/usr/src/debug/kea/2.5.8/src/bin/dhcp4/dhcp4_srv.cc:2128:(.text+0xc556): undefined reference to `isc::dhcp::CfgOption::getAll(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&) const' /OE/build/oe-core/tmp-glibc/work/core2-64-oe-linux/kea/2.5.8/recipe-sysroot-native/usr/bin/x86_64-oe-linux/../../libexec/x86_64-oe-linux/gcc/x86_64-oe-linux/14.0.1/ld: ./.libs/libdhcp4.a(dhcp4_srv.o): in function `std::vector<isc::dhcp::OptionDescriptor, std::allocator<isc::dhcp::OptionDescriptor> > isc::dhcp::CfgOption::getList<char [6]>(char const (&) [6], unsigned short) const':
/usr/src/debug/kea/2.5.8/src/lib/dhcpsrv/cfg_option.h:641:(.text._ZNK3isc4dhcp9CfgOption7getListIA6_cEESt6vectorINS0_16OptionDescriptorESaIS5_EERKT_t[_ZNK3isc4dhcp9CfgOption7getListIA6_cEESt6vectorINS0_16OptionDescriptorESaIS5_EERKT_t]+0x86): undefined reference to `isc::dhcp::CfgOption::getAll(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&) const' /OE/build/oe-core/tmp-glibc/work/core2-64-oe-linux/kea/2.5.8/recipe-sysroot-native/usr/bin/x86_64-oe-linux/../../libexec/x86_64-oe-linux/gcc/x86_64-oe-linux/14.0.1/ld: ./.libs/libdhcp4.a(dhcp4_srv.o): in function `isc::dhcp::OptionDescriptor isc::dhcp::CfgOption::get<char [6]>(char const (&) [6], unsigned short) const':
/usr/src/debug/kea/2.5.8/src/lib/dhcpsrv/cfg_option.h:609:(.text._ZNK3isc4dhcp9CfgOption3getIA6_cEENS0_16OptionDescriptorERKT_t[_ZNK3isc4dhcp9CfgOption3getIA6_cEENS0_16OptionDescriptorERKT_t]+0x77): undefined reference to `isc::dhcp::CfgOption::getAll(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&) const'
collect2: error: ld returned 1 exit status
make[5]: *** [Makefile:651: kea-dhcp4] Error 1
--- meta/recipes-connectivity/kea/kea_2.4.1.bb | 1 + 1 file changed, 1 insertion(+)
meta/recipes-connectivity/kea/kea_2.4.1.bb +1 -0
@@ -38,6 +38,7 @@ DEBUG_OPTIMIZATION:append:mipsel = " -O" BUILD_OPTIMIZATION:remove:mipsel = " -Og" BUILD_OPTIMIZATION:append:mipsel = " -O" +CXXFLAGS:remove = "-fvisibility-inlines-hidden" EXTRA_OECONF = "--with-boost-libs=-lboost_system \ --with-log4cplus=${STAGING_DIR_TARGET}${prefix} \ --with-openssl=${STAGING_DIR_TARGET}${prefix}"
Signature · 1 line
2.45.0
Re: [OE-core] [PATCH v2 07/11] kea: Remove -fvisibility-inlines-hidden from C++ flags
Quoted 1 line · Khem Raj via lists.openembedded.org
+CXXFLAGS:remove = "-fvisibility-inlines-hidden"
Same comment as v1: this seems like the wrong way around. Why do we have it as a global glibc/musl setting in the first place? Shouldn't we just remove it from there? I did some digging and this line was imported from oe-classic in 2010 with eglibc definition, so most likely whatever reasons there used to be are long obsolete.
Alex
Re: [OE-core] [PATCH v2 07/11] kea: Remove -fvisibility-inlines-hidden from C++ flags
Quoted 11 lines · Alexander Kanavin
Quoted 1 line · Khem Raj via lists.openembedded.org
+CXXFLAGS:remove = "-fvisibility-inlines-hidden"
Same comment as v1: this seems like the wrong way around. Why do we have it as a global glibc/musl setting in the first place? Shouldn't we just remove it from there? I did some digging and this line was imported from oe-classic in 2010 with eglibc definition, so most likely whatever reasons there used to be are long obsolete.
Firstly I am inclined towards removing it if we can, since I think it should be packages to decide to use it, then they can maintain it better from testing POV, However, there is something
to consider w.r.t. to this option. This does help in optimizing loading DSOs by reducing the number of PLTs and improves loading time for DSOs which might be quite meaningful for embedded devices that we target. However, we only enable hidden visibility only for inlines so it won't be full benefits but we do have better default compatibility.
Since we have had this option for a long time and it still seems relevant, I would
think it would make sense to do some benchmarking on a sizeable C++ program perhaps chromium or webkit, and see the difference in DSOs sizes with and without
this option and also benchmark the loading time of the resulting chromium package.
maybe we will find that the gains are not as significant, in that case we might drop
it from defaults.
Quoted 2 lines
Alex
Re: [OE-core] [PATCH v2 07/11] kea: Remove -fvisibility-inlines-hidden from C++ flags
Quoted 17 lines · Khem Raj
Firstly I am inclined towards removing it if we can, since I think it should be packages to decide to use it, then they can maintain it better from testing POV, However, there is something
to consider w.r.t. to this option. This does help in optimizing loading DSOs by reducing the number of PLTs and improves loading time for DSOs which might be quite meaningful for embedded devices that we target. However, we only enable hidden visibility only for inlines so it won't be full benefits but we do have better default compatibility.Since we have had this option for a long time and it still seems relevant, I would
think it would make sense to do some benchmarking on a sizeable C++ program perhaps chromium or webkit, and see the difference in DSOs sizes with and without
this option and also benchmark the loading time of the resulting chromium package.
Quoted 3 lines
maybe we will find that the gains are not as significant, in that case we might drop
it from defaults.
I digged deeper and this is where I think it originated:
https://git.openembedded.org/openembedded/commit/?id=ddd6d1fd26416a3766d754eeda4237a680a65520
Ah the simpler times of oe-classic, when one could bundle four unrelated changes into one commit and explain none of them.
Benchmarking I have no time for right now, just wanted to point out that adding the flag was never justified in commit log. Can you experiment with webkit?
Alex
Re: [OE-core] [PATCH v2 07/11] kea: Remove -fvisibility-inlines-hidden from C++ flags
Quoted 26 lines · Alexander Kanavin
Quoted 17 lines · Khem Raj
Firstly I am inclined towards removing it if we can, since I think it should be packages to decide to use it, then they can maintain it better from testing POV, However, there is something
to consider w.r.t. to this option. This does help in optimizing loading DSOs by reducing the number of PLTs and improves loading time for DSOs which might be quite meaningful for embedded devices that we target. However, we only enable hidden visibility only for inlines so it won't be full benefits but we do have better default compatibility.Since we have had this option for a long time and it still seems relevant, I would
think it would make sense to do some benchmarking on a sizeable C++ program perhaps chromium or webkit, and see the difference in DSOs sizes with and without
this option and also benchmark the loading time of the resulting chromium package.Quoted 3 lines
maybe we will find that the gains are not as significant, in that case we might drop
it from defaults.I digged deeper and this is where I think it originated:
https://git.openembedded.org/openembedded/commit/?id=ddd6d1fd26416a3766d754eeda4237a680a65520
Quoted 7 lines
Ah the simpler times of oe-classic, when one could bundle four unrelated changes into one commit and explain none of them.
Benchmarking I have no time for right now, just wanted to point out that adding the flag was never justified in commit log. Can you experiment with webkit?
Yes, I have added it to my backlog.
Quoted 1 line
Alex
Re: [OE-core] [PATCH v2 07/11] kea: Remove -fvisibility-inlines-hidden from C++ flags
In reply to Alexander KanavinQuoted 24 lines · Alexander Kanavin via lists.openembedded.org
Quoted 17 lines · Khem Raj
Firstly I am inclined towards removing it if we can, since I think it should be packages to decide to use it, then they can maintain it better from testing POV, However, there is something
to consider w.r.t. to this option. This does help in optimizing loading DSOs by reducing the number of PLTs and improves loading time for DSOs which might be quite meaningful for embedded devices that we target. However, we only enable hidden visibility only for inlines so it won't be full benefits but we do have better default compatibility.Since we have had this option for a long time and it still seems relevant, I would
think it would make sense to do some benchmarking on a sizeable C++ program perhaps chromium or webkit, and see the difference in DSOs sizes with and without
this option and also benchmark the loading time of the resulting chromium package.Quoted 3 lines
maybe we will find that the gains are not as significant, in that case we might drop
it from defaults.I digged deeper and this is where I think it originated:
https://git.openembedded.org/openembedded/commit/?id=ddd6d1fd26416a3766d754eeda4237a680a65520
Quoted 6 lines
Ah the simpler times of oe-classic, when one could bundle four unrelated changes into one commit and explain none of them.
Benchmarking I have no time for right now, just wanted to point out that adding the flag was never justified in commit log. Can you experiment with webkit?
I tried to build chromium with and without this flag but ran into build problems - sigh.
I pulled out the next most-bloated single executable that I could think of: nodejs
The bottom line is that the flag reduces the nodejs executable size by ~6%. Actual sizes:
Quoted 1 line
Quoted 1 line
Quoted 1 line
37333922 - 35245092
2088830
so 2 MB out of 37 MB !
Chromium has been updated over the weekend so I'll try that next.
What else would people like to see to assess the utility of this compiler option?
../Randy
Original build:
❯ size ../chromium-poky-m-aab1335523/tmp/work/core2-64-poky-linux/nodejs/20.12.2/packages-split/nodejs/usr/bin/node
text data bss dec hex filename
35245092 628548 154792 36028432 225c010 ../chromium-poky-m-aab1335523/tmp/work/core2-64-poky-linux/nodejs/20.12.2/packages-split/nodejs/usr/bin/node
With this patch:
❯ git diff
meta-oe/recipes-devtools/nodejs/nodejs_20.12.2.bb +0 -0
@@ -42,6 +42,8 @@ S = "${WORKDIR}/node-v${PV}"
CVE_PRODUCT += "node.js"
+CXXFLAGS:remove = "-fvisibility-inlines-hidden"
+
# v8 errors out if you have set CCACHE CCACHE = ""
❯ size tmp/work/core2-64-poky-linux/nodejs/20.12.2/packages-split/nodejs/usr/bin/node
text data bss dec hex filename
37333922 628548 154792 38117262 2459f8e tmp/work/core2-64-poky-linux/nodejs/20.12.2/packages-split/nodejs/usr/bin/node
❯ python3
Python 3.10.12 (main, Nov 20 2023, 15:14:05) [GCC 11.4.0] on linux Type "help", "copyright", "credits" or "license" for more information.
Quoted 1 line
Quoted 1 line
Quoted 1 line
35245092.0/37333922.0*100.0
94.40500786389386
Quoted 11 lines
Alex
-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#199104):https://lists.openembedded.org/g/openembedded-core/message/199104 Mute This Topic:https://lists.openembedded.org/mt/105955497/3616765
Group Owner:[email protected] Unsubscribe:https://lists.openembedded.org/g/openembedded-core/unsub [[email protected]]
-=-=-=-=-=-=-=-=-=-=-=-
Signature · 2 lines
# Randy MacLeod # Wind River Linux
Re: [OE-core] [PATCH v2 07/11] kea: Remove -fvisibility-inlines-hidden from C++ flags
Quoted 44 lines · Randy MacLeod
On 2024-05-07 3:56 p.m., Alexander Kanavin via lists.openembedded.org wrote:
On Tue, 7 May 2024 at 18:13, Khem Raj <[email protected]> wrote:
Firstly I am inclined towards removing it if we can, since I think it should be packages to decide to use it, then they can maintain it better from testing POV, However, there is something
to consider w.r.t. to this option. This does help in optimizing loading DSOs by reducing the number of PLTs and improves loading time for DSOs which might be quite meaningful for embedded devices that we target. However, we only enable hidden visibility only for inlines so it won't be full benefits but we do have better default compatibility.Since we have had this option for a long time and it still seems relevant, I would
think it would make sense to do some benchmarking on a sizeable C++ program perhaps chromium or webkit, and see the difference in DSOs sizes with and without
this option and also benchmark the loading time of the resulting chromium package.maybe we will find that the gains are not as significant, in that case we might drop
it from defaults.I digged deeper and this is where I think it originated:
https://git.openembedded.org/openembedded/commit/?id=ddd6d1fd26416a3766d754eeda4237a680a65520Ah the simpler times of oe-classic, when one could bundle four unrelated changes into one commit and explain none of them.
Benchmarking I have no time for right now, just wanted to point out that adding the flag was never justified in commit log. Can you experiment with webkit?
I tried to build chromium with and without this flag but ran into build problems - sigh.
I pulled out the next most-bloated single executable that I could think of: nodejs
The bottom line is that the flag reduces the nodejs executable size by ~6%. Actual sizes:
Quoted 9 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
37333922 - 35245092
2088830
so 2 MB out of 37 MB !
Chromium has been updated over the weekend so I'll try that next.
What else would people like to see to assess the utility of this compiler option?
Thanks for taking this up.
I think chromium and nodejs are large enough packages to give us a good idea of its benefits.
We should add a comment to config files where this option is added for future reference.
Quoted 55 lines
../Randy
Original build:
❯ size ../chromium-poky-m-aab1335523/tmp/work/core2-64-poky-linux/nodejs/20.12.2/packages-split/nodejs/usr/bin/node
text data bss dec hex filename
35245092 628548 154792 36028432 225c010 ../chromium-poky-m-aab1335523/tmp/work/core2-64-poky-linux/nodejs/20.12.2/packages-split/nodejs/usr/bin/nodeWith this patch:
❯ git diff
meta-oe/recipes-devtools/nodejs/nodejs_20.12.2.bb +2
-0@@ -42,6 +42,8 @@ S = "${WORKDIR}/node-v${PV}" CVE_PRODUCT += "node.js" +CXXFLAGS:remove = "-fvisibility-inlines-hidden" + # v8 errors out if you have set CCACHE CCACHE = ""❯ size tmp/work/core2-64-poky-linux/nodejs/20.12.2/packages-split/nodejs/usr/bin/node
text data bss dec hex filename
37333922 628548 154792 38117262 2459f8e tmp/work/core2-64-poky-linux/nodejs/20.12.2/packages-split/nodejs/usr/bin/node❯ python3
Python 3.10.12 (main, Nov 20 2023, 15:14:05) [GCC 11.4.0] on linux Type "help", "copyright", "credits" or "license" for more information.Quoted 1 line
Quoted 1 line
Quoted 1 line
35245092.0/37333922.0*100.0
94.40500786389386
Alex
-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#199104): https://lists.openembedded.org/g/openembedded-core/message/199104
Mute This Topic: https://lists.openembedded.org/mt/105955497/3616765
Group Owner: [email protected]
Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [[email protected]]
-=-=-=-=-=-=-=-=-=-=-=---
# Randy MacLeod
# Wind River Linux
Re: [OE-core] [PATCH v2 07/11] kea: Remove -fvisibility-inlines-hidden from C++ flags
Can you also try webkitgtk please? It's directly in core, no extra layer or special setup needed.
Alex
Quoted 117 lines · Khem Raj
Quoted 44 lines · Randy MacLeod
On 2024-05-07 3:56 p.m., Alexander Kanavin via lists.openembedded.org wrote:
On Tue, 7 May 2024 at 18:13, Khem Raj <[email protected]> wrote:
Firstly I am inclined towards removing it if we can, since I think it should be packages to decide to use it, then they can maintain it better from testing POV, However, there is something
to consider w.r.t. to this option. This does help in optimizing loading DSOs by reducing the number of PLTs and improves loading time for DSOs which might be quite meaningful for embedded devices that we target. However, we only enable hidden visibility only for inlines so it won't be full benefits but we do have better default compatibility.Since we have had this option for a long time and it still seems relevant, I would
think it would make sense to do some benchmarking on a sizeable C++ program perhaps chromium or webkit, and see the difference in DSOs sizes with and without
this option and also benchmark the loading time of the resulting chromium package.maybe we will find that the gains are not as significant, in that case we might drop
it from defaults.I digged deeper and this is where I think it originated:
https://git.openembedded.org/openembedded/commit/?id=ddd6d1fd26416a3766d754eeda4237a680a65520Ah the simpler times of oe-classic, when one could bundle four unrelated changes into one commit and explain none of them.
Benchmarking I have no time for right now, just wanted to point out that adding the flag was never justified in commit log. Can you experiment with webkit?
I tried to build chromium with and without this flag but ran into build problems - sigh.
I pulled out the next most-bloated single executable that I could think of: nodejs
The bottom line is that the flag reduces the nodejs executable size by ~6%. Actual sizes:
Quoted 9 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
37333922 - 35245092
2088830
so 2 MB out of 37 MB !
Chromium has been updated over the weekend so I'll try that next.
What else would people like to see to assess the utility of this compiler option?
Thanks for taking this up.
I think chromium and nodejs are large enough packages to give us a good idea of its benefits.
We should add a comment to config files where this option is added for future reference.Quoted 55 lines
../Randy
Original build:
❯ size ../chromium-poky-m-aab1335523/tmp/work/core2-64-poky-linux/nodejs/20.12.2/packages-split/nodejs/usr/bin/node
text data bss dec hex filename
35245092 628548 154792 36028432 225c010 ../chromium-poky-m-aab1335523/tmp/work/core2-64-poky-linux/nodejs/20.12.2/packages-split/nodejs/usr/bin/nodeWith this patch:
❯ git diff
meta-oe/recipes-devtools/nodejs/nodejs_20.12.2.bb +2
-0@@ -42,6 +42,8 @@ S = "${WORKDIR}/node-v${PV}" CVE_PRODUCT += "node.js" +CXXFLAGS:remove = "-fvisibility-inlines-hidden" + # v8 errors out if you have set CCACHE CCACHE = ""❯ size tmp/work/core2-64-poky-linux/nodejs/20.12.2/packages-split/nodejs/usr/bin/node
text data bss dec hex filename
37333922 628548 154792 38117262 2459f8e tmp/work/core2-64-poky-linux/nodejs/20.12.2/packages-split/nodejs/usr/bin/node❯ python3
Python 3.10.12 (main, Nov 20 2023, 15:14:05) [GCC 11.4.0] on linux Type "help", "copyright", "credits" or "license" for more information.Quoted 1 line
Quoted 1 line
Quoted 1 line
35245092.0/37333922.0*100.0
94.40500786389386
Alex
-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#199104): https://lists.openembedded.org/g/openembedded-core/message/199104
Mute This Topic: https://lists.openembedded.org/mt/105955497/3616765
Group Owner: [email protected]
Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [[email protected]]
-=-=-=-=-=-=-=-=-=-=-=---
# Randy MacLeod
# Wind River Linux
PATCH v2 8/11 consolekit: Disable incompatible-pointer-types warning as error
In reply to Khem RajFixes build with gcc-14
--- meta/recipes-support/consolekit/consolekit_0.4.6.bb | 3 +++ 1 file changed, 3 insertions(+)
meta/recipes-support/consolekit/consolekit_0.4.6.bb +3 -0
@@ -31,6 +31,9 @@ PACKAGECONFIG[pam] = "--enable-pam-module --with-pam-module-dir=${base_libdir}/s PACKAGECONFIG[polkit] = "--with-polkit,--without-polkit,polkit" PACKAGECONFIG[systemd] = "--with-systemdsystemunitdir=${systemd_system_unitdir}/,--with-systemdsystemunitdir=" +# Fails to build with GCC14 with incompatible pointer error warning being treated as error +CFLAGS += "-Wno-error=incompatible-pointer-types" + FILES:${PN} += "${exec_prefix}/lib/ConsoleKit \ ${libdir}/ConsoleKit ${systemd_unitdir} ${base_libdir} \ ${datadir}/dbus-1 ${datadir}/PolicyKit ${datadir}/polkit*"
Signature · 1 line
2.45.0
PATCH v2 9/11 gtk4: Disable int-conversion warning as error
In reply to Khem RajThis helps builds on 32bit machines when using GCC-14 there is a bug filed for fixing this in code [1]
[1] https://gitlab.gnome.org/GNOME/gtk/-/issues/6033
--- meta/recipes-gnome/gtk+/gtk4_4.14.2.bb | 4 ++++ 1 file changed, 4 insertions(+)
meta/recipes-gnome/gtk+/gtk4_4.14.2.bb +4 -0
@@ -76,6 +76,10 @@ PACKAGECONFIG[gstreamer] = "-Dmedia-gstreamer=enabled,-Dmedia-gstreamer=disabled PACKAGECONFIG[tracker] = "-Dtracker=enabled,-Dtracker=disabled,tracker,tracker-miners" PACKAGECONFIG[vulkan] = "-Dvulkan=enabled,-Dvulkan=disabled, vulkan-loader vulkan-headers shaderc-native" +# Disable int-conversion warning as error until [1] is fixed +# [1] https://gitlab.gnome.org/GNOME/gtk/-/issues/6033 +CFLAGS:append = " -Wno-error=int-conversion" + LIBV = "4.0.0" FILES:${PN}:append = " \
Signature · 1 line
2.45.0