[OE-core] [PATCH] libunwind: Fix build race conflict with gcc and musl

Khem Raj raj.khem at gmail.com
Thu Aug 18 16:49:05 UTC 2016


> On Aug 18, 2016, at 9:43 AM, Khem Raj <raj.khem at gmail.com> wrote:
> 
> 
>> On Aug 18, 2016, at 6:46 AM, Richard Purdie <richard.purdie at linuxfoundation.org> wrote:
>> 
>> Building libunwind, then gcc-runtime causes build failures. This is hard
>> to fix since gcc-runtime wants the internal gcc unwind.h header but libunwind
>> wants to provide this. There are differences in include behaviour between gcc
>> and glibc which are by design.
> 
> based upon C library being musl or glibc
> 
>> 
>> This patch hacks around the issue by looking for a define used during gcc-runtime's
>> build and skipping to the internal header in that case. The patch is only enabled
>> on musl and is the best workaround I could come up with to unblock failing builds
>> on our auto builder.
> 
> I also saw gcc build failures not only gcc-runtime. I am not able to find the logs
> once I have a reproducer, I will check if it was in libstdc++ as well or elsewhere
> nevertheless this is ok to go in.

It was compiling gcc itself for arm/musl target see

http://errors.yoctoproject.org/Errors/Details/73163/

thats why I disabled it on world builds on arm. Real fix would be
for gcc to get its act together.

> 
>> 
>> [YOCTO #10129]
>> 
>> Signed-off-by: Richard Purdie <richard.purdie at linuxfoundation.org>
>> 
>> diff --git a/meta/recipes-support/libunwind/libunwind/musl-header-conflict.patch b/meta/recipes-support/libunwind/libunwind/musl-header-conflict.patch
>> new file mode 100644
>> index 0000000..79f63fd
>> --- /dev/null
>> +++ b/meta/recipes-support/libunwind/libunwind/musl-header-conflict.patch
>> @@ -0,0 +1,37 @@
>> +If you:
>> +
>> +TCLIBC=musl bitbake unwind
>> +TCLIBC=musl bitbake gcc-runtime -c cleansstate
>> +TCLIBC=musl bitbake gcc-runtime
>> +
>> +you will see libstdc++ fail to build due to finding libunwind's header file.
>> +
>> +Khem: "When we build any of gcc components they expect to use internal version
>> +and that works with glibc based gcc since the search headers first look into gcc
>> +headers, however with musl the gcc headers are searched after the standard
>> +headers ( which is by design the right thing )."
>> +
>> +This patch hacks around the issue by looking for a define used during gcc-runtime's
>> +build and skipping to the internal header in that case.
>> +
>> +[YOCTO #10129]
>> +
>> +RP 2016/8/18
>> +
>> +Upstream-Status: Inappropriate [really need to fix gcc]
>> +
>> +Index: git/include/unwind.h
>> +===================================================================
>> +--- git.orig/include/unwind.h
>> ++++ git/include/unwind.h
>> +@@ -23,6 +23,10 @@ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER L
>> + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
>> + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.  */
>> +
>> ++#ifdef _GLIBCXX_SHARED
>> ++#include_next <unwind.h>
>> ++#endif
> 
> This is a solution we can live with since _GLIBCXX_SHARED is gcc internal define, it wont
> be visible to other packages.
> 
>> ++
>> + #ifndef _UNWIND_H
>> + #define _UNWIND_H
>> +
>> diff --git a/meta/recipes-support/libunwind/libunwind_git.bb b/meta/recipes-support/libunwind/libunwind_git.bb
>> index 6d1d9fb..cd4cb89 100644
>> --- a/meta/recipes-support/libunwind/libunwind_git.bb
>> +++ b/meta/recipes-support/libunwind/libunwind_git.bb
>> @@ -13,6 +13,7 @@ SRC_URI = "git://git.sv.gnu.org/libunwind.git \
>>           file://0001-ppc32-Consider-ucontext-mismatches-between-glibc-and.patch \
>>           "
>> 
>> +SRC_URI_append_libc-musl = " file://musl-header-conflict.patch”
> 
> It should be fine to use it in general too.
> 
>> EXTRA_OECONF_append_libc-musl = " --disable-documentation --disable-tests "
>> 
>> # http://errors.yoctoproject.org/Errors/Details/20487/
>> 
>> 
> 

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 204 bytes
Desc: Message signed with OpenPGP using GPGMail
URL: <http://lists.openembedded.org/pipermail/openembedded-core/attachments/20160818/d2b1ea70/attachment-0002.sig>


More information about the Openembedded-core mailing list