[RFC][PATCH] gcc-runtime.inc: force libstdc++ to always use gcc4 compatible ABI

[RFC][PATCH] gcc-runtime.inc: force libstdc++ to always use gcc4 compatible ABI

[Andre McCurdy]

Experimental hack, not expected to be merged.

Without this, the http://www.cplusplus.com/reference/string/string/append/
example fails to build with -std=c++11 and it's not clear why (it's as if we
build a version of libstdc++ which supports only the old ABI but with headers
which try to use the new ABI when compiled with -std=c++11 ?). As a workaround,
force libstdc++ to provide, and default to using, the old ABI in all cases.

See also:

  https://gcc.gnu.org/onlinedocs/libstdc++/manual/configure.html
  https://mail.gnome.org/archives/gtkmm-list/2015-June/msg00026.html
  https://wiki.debian.org/GCC5

Signed-off-by: Andre McCurdy <armccurdy@gmail.com>
---
 meta/recipes-devtools/gcc/gcc-runtime.inc | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/meta/recipes-devtools/gcc/gcc-runtime.inc b/meta/recipes-devtools/gcc/gcc-runtime.inc
index f20d298..3a476cd 100644
--- a/meta/recipes-devtools/gcc/gcc-runtime.inc
+++ b/meta/recipes-devtools/gcc/gcc-runtime.inc
@@ -10,6 +10,19 @@ EXTRA_OECONF_PATHS = "\
 
 EXTRA_OECONF_append_linuxstdbase = " --enable-clocale=gnu"
 
+# gcc-runtime.inc: force libstdc++ to always use gcc4 compatible ABI
+# Experimental hack. Without this, the http://www.cplusplus.com/reference/string/string/append/
+# example fails to build with -std=c++11 and it's not clear why (it's as if we
+# build a version of libstdc++ which supports only the old ABI but with headers
+# which try to use the new ABI when compiled with -std=c++11 ?). As a workaround,
+# force libstdc++ to provide, and default to using, the old ABI in all cases.
+# See also:
+# https://gcc.gnu.org/onlinedocs/libstdc++/manual/configure.html
+# https://mail.gnome.org/archives/gtkmm-list/2015-June/msg00026.html
+# https://wiki.debian.org/GCC5
+#
+EXTRA_OECONF_append = " --disable-libstdcxx-dual-abi --with-default-libstdcxx-abi=gcc4-compatible"
+
 RUNTIMETARGET = "libssp libstdc++-v3 libgomp libatomic \
     ${@bb.utils.contains_any('FORTRAN', [',fortran',',f77'], 'libquadmath', '', d)} \
 "
-- 
1.9.1

Re: [RFC][PATCH] gcc-runtime.inc: force libstdc++ to always use gcc4 compatible ABI

[Khem Raj]

[-- Attachment #1: Type: text/plain, Size: 2835 bytes --]


> On Jan 26, 2016, at 12:10 AM, Andre McCurdy <armccurdy@gmail.com> wrote:
> 
> Experimental hack, not expected to be merged.
> 
> Without this, the http://www.cplusplus.com/reference/string/string/append/
> example fails to build with -std=c++11 and it's not clear why (it's as if we
> build a version of libstdc++ which supports only the old ABI but with headers
> which try to use the new ABI when compiled with -std=c++11 ?). As a workaround,
> force libstdc++ to provide, and default to using, the old ABI in all cases.
> 
> See also:
> 
>  https://gcc.gnu.org/onlinedocs/libstdc++/manual/configure.html
>  https://mail.gnome.org/archives/gtkmm-list/2015-June/msg00026.html
>  https://wiki.debian.org/GCC5
> 
> Signed-off-by: Andre McCurdy <armccurdy@gmail.com>
> ---
> meta/recipes-devtools/gcc/gcc-runtime.inc | 13 +++++++++++++
> 1 file changed, 13 insertions(+)
> 
> diff --git a/meta/recipes-devtools/gcc/gcc-runtime.inc b/meta/recipes-devtools/gcc/gcc-runtime.inc
> index f20d298..3a476cd 100644
> --- a/meta/recipes-devtools/gcc/gcc-runtime.inc
> +++ b/meta/recipes-devtools/gcc/gcc-runtime.inc
> @@ -10,6 +10,19 @@ EXTRA_OECONF_PATHS = "\
> 
> EXTRA_OECONF_append_linuxstdbase = " --enable-clocale=gnu"
> 
> +# gcc-runtime.inc: force libstdc++ to always use gcc4 compatible ABI
> +# Experimental hack. Without this, the http://www.cplusplus.com/reference/string/string/append/
> +# example fails to build with -std=c++11 and it's not clear why (it's as if we
> +# build a version of libstdc++ which supports only the old ABI but with headers
> +# which try to use the new ABI when compiled with -std=c++11 ?). As a workaround,
> +# force libstdc++ to provide, and default to using, the old ABI in all cases.
> +# See also:
> +# https://gcc.gnu.org/onlinedocs/libstdc++/manual/configure.html
> +# https://mail.gnome.org/archives/gtkmm-list/2015-June/msg00026.html
> +# https://wiki.debian.org/GCC5
> +#
> +EXTRA_OECONF_append = " --disable-libstdcxx-dual-abi --with-default-libstdcxx-abi=gcc4-compatible”

This change while makes it work compatible to older libstdc++ ABI, we should make the defaults to be
forward looking. I would rather keep --with-default-libstdcxx-abi=c++11 since gcc6 will switch to newer
c++ std as default. If we revert like you suggest, we will be up for more work in gcc6 time.

please define _GLIBCXX_USE_CXX11_ABI=0 for your app, where you see the problem.

> +

> RUNTIMETARGET = "libssp libstdc++-v3 libgomp libatomic \
>     ${@bb.utils.contains_any('FORTRAN', [',fortran',',f77'], 'libquadmath', '', d)} \
> "
> --
> 1.9.1
> 
> --
> _______________________________________________
> Openembedded-core mailing list
> Openembedded-core@lists.openembedded.org
> http://lists.openembedded.org/mailman/listinfo/openembedded-core


[-- Attachment #2: Message signed with OpenPGP using GPGMail --]
[-- Type: application/pgp-signature, Size: 211 bytes --]

Re: [RFC][PATCH] gcc-runtime.inc: force libstdc++ to always use gcc4 compatible ABI

[Andre McCurdy]

On Tue, Jan 26, 2016 at 8:38 AM, Khem Raj <raj.khem@gmail.com> wrote:
>
>> On Jan 26, 2016, at 12:10 AM, Andre McCurdy <armccurdy@gmail.com> wrote:
>>
>> Experimental hack, not expected to be merged.
>>
>> Without this, the http://www.cplusplus.com/reference/string/string/append/
>> example fails to build with -std=c++11 and it's not clear why (it's as if we
>> build a version of libstdc++ which supports only the old ABI but with headers
>> which try to use the new ABI when compiled with -std=c++11 ?). As a workaround,
>> force libstdc++ to provide, and default to using, the old ABI in all cases.
>>
>> See also:
>>
>>  https://gcc.gnu.org/onlinedocs/libstdc++/manual/configure.html
>>  https://mail.gnome.org/archives/gtkmm-list/2015-June/msg00026.html
>>  https://wiki.debian.org/GCC5
>>
>> Signed-off-by: Andre McCurdy <armccurdy@gmail.com>
>> ---
>> meta/recipes-devtools/gcc/gcc-runtime.inc | 13 +++++++++++++
>> 1 file changed, 13 insertions(+)
>>
>> diff --git a/meta/recipes-devtools/gcc/gcc-runtime.inc b/meta/recipes-devtools/gcc/gcc-runtime.inc
>> index f20d298..3a476cd 100644
>> --- a/meta/recipes-devtools/gcc/gcc-runtime.inc
>> +++ b/meta/recipes-devtools/gcc/gcc-runtime.inc
>> @@ -10,6 +10,19 @@ EXTRA_OECONF_PATHS = "\
>>
>> EXTRA_OECONF_append_linuxstdbase = " --enable-clocale=gnu"
>>
>> +# gcc-runtime.inc: force libstdc++ to always use gcc4 compatible ABI
>> +# Experimental hack. Without this, the http://www.cplusplus.com/reference/string/string/append/
>> +# example fails to build with -std=c++11 and it's not clear why (it's as if we
>> +# build a version of libstdc++ which supports only the old ABI but with headers
>> +# which try to use the new ABI when compiled with -std=c++11 ?). As a workaround,
>> +# force libstdc++ to provide, and default to using, the old ABI in all cases.
>> +# See also:
>> +# https://gcc.gnu.org/onlinedocs/libstdc++/manual/configure.html
>> +# https://mail.gnome.org/archives/gtkmm-list/2015-June/msg00026.html
>> +# https://wiki.debian.org/GCC5
>> +#
>> +EXTRA_OECONF_append = " --disable-libstdcxx-dual-abi --with-default-libstdcxx-abi=gcc4-compatible”
>
> This change while makes it work compatible to older libstdc++ ABI, we should make the defaults to be
> forward looking.

Yes, I agree. The problem is that currently our libstdc++ doesn't seem
to support the new ABI.

> I would rather keep --with-default-libstdcxx-abi=c++11 since gcc6 will switch to newer
> c++ std as default. If we revert like you suggest, we will be up for more work in gcc6 time.
>
> please define _GLIBCXX_USE_CXX11_ABI=0 for your app, where you see the problem.

My test case is the example app from:

  http://www.cplusplus.com/reference/string/string/append/

It fails when I try to compile to use the new ABI (ie by using
-stdc++11 or _GLIBCXX_USE_CXX11_ABI=1).

There is no legacy code involved.

>
>> RUNTIMETARGET = "libssp libstdc++-v3 libgomp libatomic \
>>     ${@bb.utils.contains_any('FORTRAN', [',fortran',',f77'], 'libquadmath', '', d)} \
>> "
>> --
>> 1.9.1
>>
>> --
>> _______________________________________________
>> Openembedded-core mailing list
>> Openembedded-core@lists.openembedded.org
>> http://lists.openembedded.org/mailman/listinfo/openembedded-core
>

Re: [RFC][PATCH] gcc-runtime.inc: force libstdc++ to always use gcc4 compatible ABI

[Khem Raj]

[-- Attachment #1: Type: text/plain, Size: 3710 bytes --]


> On Jan 26, 2016, at 12:18 PM, Andre McCurdy <armccurdy@gmail.com> wrote:
> 
> On Tue, Jan 26, 2016 at 8:38 AM, Khem Raj <raj.khem@gmail.com> wrote:
>> 
>>> On Jan 26, 2016, at 12:10 AM, Andre McCurdy <armccurdy@gmail.com> wrote:
>>> 
>>> Experimental hack, not expected to be merged.
>>> 
>>> Without this, the http://www.cplusplus.com/reference/string/string/append/
>>> example fails to build with -std=c++11 and it's not clear why (it's as if we
>>> build a version of libstdc++ which supports only the old ABI but with headers
>>> which try to use the new ABI when compiled with -std=c++11 ?). As a workaround,
>>> force libstdc++ to provide, and default to using, the old ABI in all cases.
>>> 
>>> See also:
>>> 
>>> https://gcc.gnu.org/onlinedocs/libstdc++/manual/configure.html
>>> https://mail.gnome.org/archives/gtkmm-list/2015-June/msg00026.html
>>> https://wiki.debian.org/GCC5
>>> 
>>> Signed-off-by: Andre McCurdy <armccurdy@gmail.com>
>>> ---
>>> meta/recipes-devtools/gcc/gcc-runtime.inc | 13 +++++++++++++
>>> 1 file changed, 13 insertions(+)
>>> 
>>> diff --git a/meta/recipes-devtools/gcc/gcc-runtime.inc b/meta/recipes-devtools/gcc/gcc-runtime.inc
>>> index f20d298..3a476cd 100644
>>> --- a/meta/recipes-devtools/gcc/gcc-runtime.inc
>>> +++ b/meta/recipes-devtools/gcc/gcc-runtime.inc
>>> @@ -10,6 +10,19 @@ EXTRA_OECONF_PATHS = "\
>>> 
>>> EXTRA_OECONF_append_linuxstdbase = " --enable-clocale=gnu"
>>> 
>>> +# gcc-runtime.inc: force libstdc++ to always use gcc4 compatible ABI
>>> +# Experimental hack. Without this, the http://www.cplusplus.com/reference/string/string/append/
>>> +# example fails to build with -std=c++11 and it's not clear why (it's as if we
>>> +# build a version of libstdc++ which supports only the old ABI but with headers
>>> +# which try to use the new ABI when compiled with -std=c++11 ?). As a workaround,
>>> +# force libstdc++ to provide, and default to using, the old ABI in all cases.
>>> +# See also:
>>> +# https://gcc.gnu.org/onlinedocs/libstdc++/manual/configure.html
>>> +# https://mail.gnome.org/archives/gtkmm-list/2015-June/msg00026.html
>>> +# https://wiki.debian.org/GCC5
>>> +#
>>> +EXTRA_OECONF_append = " --disable-libstdcxx-dual-abi --with-default-libstdcxx-abi=gcc4-compatible”
>> 
>> This change while makes it work compatible to older libstdc++ ABI, we should make the defaults to be
>> forward looking.
> 
> Yes, I agree. The problem is that currently our libstdc++ doesn't seem
> to support the new ABI.
> 
>> I would rather keep --with-default-libstdcxx-abi=c++11 since gcc6 will switch to newer
>> c++ std as default. If we revert like you suggest, we will be up for more work in gcc6 time.
>> 
>> please define _GLIBCXX_USE_CXX11_ABI=0 for your app, where you see the problem.
> 
> My test case is the example app from:
> 
>  http://www.cplusplus.com/reference/string/string/append/
> 
> It fails when I try to compile to use the new ABI (ie by using
> -stdc++11 or _GLIBCXX_USE_CXX11_ABI=1).

Lets not mix API ( std=c++11 ) and libstdc++ c++11 ABI, both are different.

have you tried the converse where you suspect the test is not c++11 ABI compliant for STL
in particular

str.append<int>(5,0x2E);

> 
> There is no legacy code involved.
> 
>> 
>>> RUNTIMETARGET = "libssp libstdc++-v3 libgomp libatomic \
>>>    ${@bb.utils.contains_any('FORTRAN', [',fortran',',f77'], 'libquadmath', '', d)} \
>>> "
>>> --
>>> 1.9.1
>>> 
>>> --
>>> _______________________________________________
>>> Openembedded-core mailing list
>>> Openembedded-core@lists.openembedded.org
>>> http://lists.openembedded.org/mailman/listinfo/openembedded-core
>> 


[-- Attachment #2: Message signed with OpenPGP using GPGMail --]
[-- Type: application/pgp-signature, Size: 211 bytes --]