[1/1] package/jsoncpp: broken with musl on x86

This patch replaces a previous submission made by Bernd Kuhls some while ago
[1].

The posix_memalign declaration is incompatible with musl for C++, because of
the exception specification [2]. Note, the referenced patch is part of the a
patch series to "Add musl support to GCC".

Instead of patching gcc in Buildroot we disable jsoncpp and any reverse
dependendy with the musl toolchain on x86.

[1] https://patchwork.ozlabs.org/patch/505425/
[2] https://gcc.gnu.org/ml/gcc-patches/2015-04/msg01084.html

Fixes:
http://autobuild.buildroot.net/results/588/5885d33b8d9f17878f565f3ae5371017dc89aede/
http://autobuild.buildroot.net/results/0d9/0d90c7c13ae1640b07709c4c686e0237ada1324d/
http://autobuild.buildroot.net/results/30b/30b98ac7362c27254218a6e521d29971070f8c3c/
http://autobuild.buildroot.net/results/292/29289c125166630aeaf884ccc006bcf12c8aa0f6/

and many more.

Signed-off-by: Jörg Krause <joerg.krause@embedded.rocks>
---
 package/cmake/Config.in            | 4 ++++
 package/jsoncpp/Config.in          | 5 +++++
 package/kodi-pvr-argustv/Config.in | 4 ++++
 package/kodi-pvr-filmon/Config.in  | 4 ++++
 package/kodi-pvr-pctv/Config.in    | 4 ++++
 package/kodi-pvr-stalker/Config.in | 4 ++++
 package/sysdig/Config.in           | 4 ++++
 7 files changed, 29 insertions(+)

Dear Jörg Krause,

On Tue, 20 Oct 2015 22:51:53 +0200, Jörg Krause wrote:
> This patch replaces a previous submission made by Bernd Kuhls some while ago
> [1].
> 
> The posix_memalign declaration is incompatible with musl for C++, because of
> the exception specification [2]. Note, the referenced patch is part of the a
> patch series to "Add musl support to GCC".
> 
> Instead of patching gcc in Buildroot we disable jsoncpp and any reverse
> dependendy with the musl toolchain on x86.
> 
> [1] https://patchwork.ozlabs.org/patch/505425/
> [2] https://gcc.gnu.org/ml/gcc-patches/2015-04/msg01084.html

It is indeed not very practical to fix gcc, since external toolchains
would continue to be affected. But isn't there a way of working around
the problem inside jsoncpp itself, so that we don't have to do this
complicated reverse-dependency propagation ?

Thanks,

Thomas

Hello,

On Wed, 21 Oct 2015 13:50:29 +0200, Thomas Petazzoni wrote:

> On Tue, 20 Oct 2015 22:51:53 +0200, Jörg Krause wrote:
> > This patch replaces a previous submission made by Bernd Kuhls some while ago
> > [1].
> > 
> > The posix_memalign declaration is incompatible with musl for C++, because of
> > the exception specification [2]. Note, the referenced patch is part of the a
> > patch series to "Add musl support to GCC".
> > 
> > Instead of patching gcc in Buildroot we disable jsoncpp and any reverse
> > dependendy with the musl toolchain on x86.
> > 
> > [1] https://patchwork.ozlabs.org/patch/505425/
> > [2] https://gcc.gnu.org/ml/gcc-patches/2015-04/msg01084.html
> 
> It is indeed not very practical to fix gcc, since external toolchains
> would continue to be affected. But isn't there a way of working around
> the problem inside jsoncpp itself, so that we don't have to do this
> complicated reverse-dependency propagation ?

There is indeed a solution: remove the -pedantic flags from

set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -Wall -Wconversion -Wshadow -Wextra -pedantic")

line in CMakeLists.txt. And it builds just fine. It seems a lot simpler
than propagating the reverse dependencies.

Could you submit a patch doing this ?

Best regards,

Thomas

Hi Thomas,

On Di, 2015-10-27 at 07:52 +0100, Thomas Petazzoni wrote:
> Hello,
> 
> On Wed, 21 Oct 2015 13:50:29 +0200, Thomas Petazzoni wrote:
> 
> > On Tue, 20 Oct 2015 22:51:53 +0200, Jörg Krause wrote:
> > > This patch replaces a previous submission made by Bernd Kuhls
> > > some while ago
> > > [1].
> > > 
> > > The posix_memalign declaration is incompatible with musl for C++,
> > > because of
> > > the exception specification [2]. Note, the referenced patch is
> > > part of the a
> > > patch series to "Add musl support to GCC".
> > > 
> > > Instead of patching gcc in Buildroot we disable jsoncpp and any
> > > reverse
> > > dependendy with the musl toolchain on x86.
> > > 
> > > [1] https://patchwork.ozlabs.org/patch/505425/
> > > [2] https://gcc.gnu.org/ml/gcc-patches/2015-04/msg01084.html
> > 
> > It is indeed not very practical to fix gcc, since external
> > toolchains
> > would continue to be affected. But isn't there a way of working
> > around
> > the problem inside jsoncpp itself, so that we don't have to do this
> > complicated reverse-dependency propagation ?
> 
> There is indeed a solution: remove the -pedantic flags from
> 
> set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -Wall -Wconversion
> -Wshadow -Wextra -pedantic")
> 
> line in CMakeLists.txt. And it builds just fine. It seems a lot
> simpler
> than propagating the reverse dependencies.
> 
> Could you submit a patch doing this ?

Indeed, this is much simpler. Many thanks for investigation ... I did
not had the time. I'll prepare a patch for this soon!

Best regards
Jörg Krause

Dear Jörg Krause,

On Tue, 27 Oct 2015 08:54:13 +0100, Jörg Krause wrote:

> > line in CMakeLists.txt. And it builds just fine. It seems a lot
> > simpler
> > than propagating the reverse dependencies.
> > 
> > Could you submit a patch doing this ?
> 
> Indeed, this is much simpler. Many thanks for investigation ... I did
> not had the time. I'll prepare a patch for this soon!

I actually didn't do much investigation. I just googled the issue, and
in the discussion around the musl related patches sent to gcc, this was
discussed, and it was pointed out that the error only occurs if
-pedantic is used.

Thomas