[Buildroot] [PATCH/next/RFC 1/2] package/mariadb: handle missing ucontext_t

[Buildroot] [PATCH/next/RFC 1/2] package/mariadb: handle missing ucontext_t

[Bernd Kuhls]

On some archs uclibc does not provide ucontext_t, for details see
https://git.buildroot.net/buildroot/commit/?id=f1cbfeea95e6287c7a666aafc182ffa318eff262

Mariadb uses ucontext_t optionally, its use can be disabled by the
corresponding cmake option. Instead of copying the Config.in code from
libsigsegv we shamelessly use its ARCH_SUPPORTS option while thinking
that a more general approach might be needed for ucontext_t support so
please consider this patch a RFC.

This patch fixes a build error

In file included from /home/bernd/buildroot/output/build/mariadb-10.2.17/libmariadb/plugins/pvio/pvio_socket.c:32:0:
/home/bernd/buildroot/output/build/mariadb-10.2.17/libmariadb/include/ma_context.h:62:3: error: unknown type name ?ucontext_t?
   ucontext_t base_context;
   ^~~~~~~~~~

using this defconfig:

BR2_arcle=y
BR2_archs38=y
BR2_TOOLCHAIN_EXTERNAL=y
BR2_TOOLCHAIN_EXTERNAL_CUSTOM=y
BR2_TOOLCHAIN_EXTERNAL_DOWNLOAD=y
BR2_TOOLCHAIN_EXTERNAL_URL="http://autobuild.buildroot.org/toolchains/tarballs/br-arcle-hs38-full-2018.02-926-gb393c24.tar.bz2"
BR2_TOOLCHAIN_EXTERNAL_GCC_7=y
BR2_TOOLCHAIN_EXTERNAL_HEADERS_4_1=y
BR2_TOOLCHAIN_EXTERNAL_LOCALE=y
BR2_TOOLCHAIN_EXTERNAL_CXX=y
BR2_PACKAGE_MYSQL=y
BR2_PACKAGE_MARIADB=y
BR2_PACKAGE_MARIADB_SERVER=y

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
---
 package/mariadb/mariadb.mk | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/package/mariadb/mariadb.mk b/package/mariadb/mariadb.mk
index 075582288a..7074e3510e 100644
--- a/package/mariadb/mariadb.mk
+++ b/package/mariadb/mariadb.mk
@@ -61,6 +61,11 @@ else
 MARIADB_CONF_OPTS += -DWITHOUT_SERVER=ON
 endif
 
+# hack: do we need a ucontext-specific option?
+ifeq ($(BR2_PACKAGE_LIBSIGSEGV_ARCH_SUPPORTS),)
+MARIADB_CONF_OPTS += -DHAVE_UCONTEXT_H=0
+endif
+
 MARIADB_CONF_OPTS += \
 	-DINSTALL_DOCDIR=share/doc/mariadb-$(MARIADB_VERSION) \
 	-DINSTALL_DOCREADMEDIR=share/doc/mariadb-$(MARIADB_VERSION) \
-- 
2.18.0

[Buildroot] [PATCH/next 2/2] package/mariadb: bump version to 10.3.9

[Bernd Kuhls]

From: Peter Seiderer <ps.report@gmx.net>

Release notes: https://mariadb.com/kb/en/mariadb-1039-release-notes/
Changelog: https://mariadb.com/kb/en/mariadb-1039-changelog/

- change WITH_SSL for host build from bundled to system (and add
  host-openssl dependency) to avoid the following configure failure:

  CMake Error at /usr/share/cmake/Modules/FindPackageHandleStandardArgs.cmake:137 (message):
    Could NOT find GnuTLS (missing: GNUTLS_LIBRARY GNUTLS_INCLUDE_DIR)
    (Required is at least version "3.3.24")
  Call Stack (most recent call first):
    /usr/share/cmake/Modules/FindPackageHandleStandardArgs.cmake:378 (_FPHSA_FAILURE_MESSAGE)
    /usr/share/cmake/Modules/FindGnuTLS.cmake:54 (FIND_PACKAGE_HANDLE_STANDARD_ARGS)
    libmariadb/CMakeLists.txt:298 (FIND_PACKAGE)

This bump also fixes broken atomic support for
BR2_TOOLCHAIN_HAS_LIBATOMIC=y with mariadb-10.2:

[ 59%] Linking CXX executable mariabackup
../../storage/innobase/libinnobase.a(handler0alter.cc.o): In function `my_atomic_add64':
/home/bernd/buildroot/output/build/mariadb-10.2.17/include/my_atomic.h:274: undefined reference to `__sync_fetch_and_add_8'

with this defconfig

BR2_arcle=y
BR2_archs38=y
BR2_TOOLCHAIN_EXTERNAL=y
BR2_TOOLCHAIN_EXTERNAL_CUSTOM=y
BR2_TOOLCHAIN_EXTERNAL_DOWNLOAD=y
BR2_TOOLCHAIN_EXTERNAL_URL="http://autobuild.buildroot.org/toolchains/tarballs/br-arcle-hs38-full-2018.02-926-gb393c24.tar.bz2"
BR2_TOOLCHAIN_EXTERNAL_GCC_7=y
BR2_TOOLCHAIN_EXTERNAL_HEADERS_4_1=y
BR2_TOOLCHAIN_EXTERNAL_LOCALE=y
BR2_TOOLCHAIN_EXTERNAL_CXX=y
BR2_PACKAGE_MYSQL=y
BR2_PACKAGE_MARIADB=y
BR2_PACKAGE_MARIADB_SERVER=y

Adding "-latomic" to CXXFLAGS and/or adding -DHAVE_GCC_ATOMIC_BUILTINS=1
to CONF_OPTS, like we do for mariadb-10.3, does not fix the build error
with mariadb-10.2. There the build would stop even earlier without these
options:

/home/bernd/buildroot/output/build/mariadb-10.2.17/include/my_atomic.h:138:2:
 error: #error atomic ops for this platform are not implemented
 #error atomic ops for this platform are not implemented

Mariadb-10.3 contains improved atomic support, this build error does
not occur here.

To prevent adding "depends on BR2_TOOLCHAIN_HAS_SYNC_8" to mariadb-10.2
and then removing it now we bump and fix the atomic problem in one
commit instead.

Updated license hash of README.md after upstream commits:
https://github.com/MariaDB/server/commits/10.3/README.md

my-small.cnf.sh is not provided anymore by upstream:
https://github.com/MariaDB/server/commit/7fee164faf8fce7be4ebe322d2178efd3d075eae

[Peter Seiderer: bumped to version 10.3.7]
Signed-off-by: Peter Seiderer <ps.report@gmx.net>
[Bernd: bumped to version 10.3.9, fixed atomic support,
 fixed my-small.cnf.sh, updated license hash]
Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
---
 package/mariadb/mariadb.hash | 12 ++++++------
 package/mariadb/mariadb.mk   | 14 ++++++++++----
 2 files changed, 16 insertions(+), 10 deletions(-)

diff --git a/package/mariadb/mariadb.hash b/package/mariadb/mariadb.hash
index d44b4f35ab..41e60037be 100644
--- a/package/mariadb/mariadb.hash
+++ b/package/mariadb/mariadb.hash
@@ -1,9 +1,9 @@
-# From https://downloads.mariadb.org/mariadb/10.2.17/
-md5 97dac7c5c288dbbbdd97768972daeb2e  mariadb-10.2.17.tar.gz
-sha1 38f2ba16297e14baa2f641d2273bbb92f56d0b94  mariadb-10.2.17.tar.gz
-sha256 e7b3078f8de874a4d451242a8a3eed49bf6f916dcd52fc3efa55886f5f35be27  mariadb-10.2.17.tar.gz
-sha512 9714f786abde43b418d8b5accbbd485de213b90db18c879625ce30777526d28fdafd3bdb43922ab6872b300cfa06895b5b3fad5eb0c0e804ad66a1fd1793a141  mariadb-10.2.17.tar.gz
+# From https://downloads.mariadb.org/mariadb/10.3.9
+md5 39c871e65134aabbc12c856691912435  mariadb-10.3.9.tar.gz
+sha1 51e9870847daa76f1e82ade97d9cf60a31df7656  mariadb-10.3.9.tar.gz
+sha256 561c6969bbd24dbb22d1d196a6b037665389b91e6dab881f39c5616389f156f4  mariadb-10.3.9.tar.gz
+sha512 7c7a678fb0c85da64b6df3df17b8cf7f2ccf862e27dcf8535f7403f8693c76b3b1a912f4b6ca876a591ada60a99577a947bfe7a63c602f754332e1e4f689e6e7  mariadb-10.3.9.tar.gz
 
 # Hash for license files
-sha256 dec3c92fdc58070facec08db681aa6b9cb05fd21e463358161c1df92e1c0653f  README.md
+sha256 5baa5057c525cacc9f7814215582ac150e5bb0b0007aa8f6ebc50a5c1b7a496d  README.md
 sha256 ab15fd526bd8dd18a9e77ebc139656bf4d33e97fc7238cd11bf60e2b9b8666c6  COPYING
diff --git a/package/mariadb/mariadb.mk b/package/mariadb/mariadb.mk
index 7074e3510e..f3c8ed4081 100644
--- a/package/mariadb/mariadb.mk
+++ b/package/mariadb/mariadb.mk
@@ -4,7 +4,7 @@
 #
 ################################################################################
 
-MARIADB_VERSION = 10.2.17
+MARIADB_VERSION = 10.3.9
 MARIADB_SITE = https://downloads.mariadb.org/interstitial/mariadb-$(MARIADB_VERSION)/source
 MARIADB_LICENSE = GPL-2.0 (server), GPL-2.0 with FLOSS exception (GPL client library), LGPL-2.0 (LGPL client library)
 # Tarball no longer contains LGPL license text
@@ -66,7 +66,14 @@ ifeq ($(BR2_PACKAGE_LIBSIGSEGV_ARCH_SUPPORTS),)
 MARIADB_CONF_OPTS += -DHAVE_UCONTEXT_H=0
 endif
 
+MARIADB_CXXFLAGS = $(TARGET_CXXFLAGS)
+
+ifeq ($(BR2_TOOLCHAIN_HAS_LIBATOMIC),y)
+MARIADB_CXXFLAGS += -latomic
+endif
+
 MARIADB_CONF_OPTS += \
+	-DCMAKE_CXX_FLAGS="$(MARIADB_CXXFLAGS)" \
 	-DINSTALL_DOCDIR=share/doc/mariadb-$(MARIADB_VERSION) \
 	-DINSTALL_DOCREADMEDIR=share/doc/mariadb-$(MARIADB_VERSION) \
 	-DINSTALL_MANDIR=share/man \
@@ -80,7 +87,8 @@ MARIADB_CONF_OPTS += \
 	-DMYSQL_DATADIR=/var/lib/mysql \
 	-DMYSQL_UNIX_ADDR=$(MYSQL_SOCKET)
 
-HOST_MARIADB_CONF_OPTS += -DWITH_SSL=OFF
+HOST_MARIADB_DEPENDENCIES = host-openssl
+HOST_MARIADB_CONF_OPTS += -DWITH_SSL=system
 
 # Some helpers must be compiled for host in order to crosscompile mariadb for
 # the target. They are then included by import_executables.cmake which is
@@ -121,8 +129,6 @@ endif
 # We also don't need the test suite on the target
 define MARIADB_POST_INSTALL
 	mkdir -p $(TARGET_DIR)/var/lib/mysql
-	$(INSTALL) -D -m 644 $(TARGET_DIR)/usr/share/mysql/my-small.cnf \
-		$(TARGET_DIR)/etc/mysql/my.cnf
 	$(RM) $(TARGET_DIR)/usr/bin/mysql_config
 	$(RM) -r $(TARGET_DIR)/usr/share/mysql/test
 endef
-- 
2.18.0

[Buildroot] [PATCH/next 2/2] package/mariadb: bump version to 10.3.9

[Thomas Petazzoni]

Hello,

On Sat, 25 Aug 2018 22:44:03 +0200, Bernd Kuhls wrote:
> From: Peter Seiderer <ps.report@gmx.net>
> 
> Release notes: https://mariadb.com/kb/en/mariadb-1039-release-notes/
> Changelog: https://mariadb.com/kb/en/mariadb-1039-changelog/
> 
> - change WITH_SSL for host build from bundled to system (and add
>   host-openssl dependency) to avoid the following configure failure:
> 
>   CMake Error at /usr/share/cmake/Modules/FindPackageHandleStandardArgs.cmake:137 (message):
>     Could NOT find GnuTLS (missing: GNUTLS_LIBRARY GNUTLS_INCLUDE_DIR)
>     (Required is at least version "3.3.24")
>   Call Stack (most recent call first):
>     /usr/share/cmake/Modules/FindPackageHandleStandardArgs.cmake:378 (_FPHSA_FAILURE_MESSAGE)
>     /usr/share/cmake/Modules/FindGnuTLS.cmake:54 (FIND_PACKAGE_HANDLE_STANDARD_ARGS)
>     libmariadb/CMakeLists.txt:298 (FIND_PACKAGE)

We changed from -DWITH_SSL=bundled to -DWITH_SSL=OFF very recently, in
commit 9276747defdd4f91ec21ba0930afdf6d947730b6. Do we really need SSL
support in the host mariadb build ?

The move to -DWITH_SSL=OFF in 9276747defdd4f91ec21ba0930afdf6d947730b6
was also done to fix exactly the same build issue, according to the
commit log.

> This bump also fixes broken atomic support for
> BR2_TOOLCHAIN_HAS_LIBATOMIC=y with mariadb-10.2:
> 
> [ 59%] Linking CXX executable mariabackup
> ../../storage/innobase/libinnobase.a(handler0alter.cc.o): In function `my_atomic_add64':
> /home/bernd/buildroot/output/build/mariadb-10.2.17/include/my_atomic.h:274: undefined reference to `__sync_fetch_and_add_8'

Is it the bump that fixes the atomic issue, or the fact that this patch
adds the logic to link against libatomic when needed ?

Thanks,

Thomas
-- 
Thomas Petazzoni, CTO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com

[Buildroot] [PATCH/next 2/2] package/mariadb: bump version to 10.3.9

[Bernd Kuhls]

Hi Thomas,

Am Sat, 25 Aug 2018 23:05:14 +0200 schrieb Thomas Petazzoni:

> We changed from -DWITH_SSL=bundled to -DWITH_SSL=OFF very recently, in
> commit 9276747defdd4f91ec21ba0930afdf6d947730b6. Do we really need SSL
> support in the host mariadb build ?

will test, thanks for the hint.

>> This bump also fixes broken atomic support for
>> BR2_TOOLCHAIN_HAS_LIBATOMIC=y with mariadb-10.2:
>> 
>> [ 59%] Linking CXX executable mariabackup
>> ../../storage/innobase/libinnobase.a(handler0alter.cc.o): In function
>> `my_atomic_add64':
>> /home/bernd/buildroot/output/build/mariadb-10.2.17/include/my_atomic.h:
274:
>> undefined reference to `__sync_fetch_and_add_8'
> 
> Is it the bump that fixes the atomic issue, or the fact that this patch
> adds the logic to link against libatomic when needed ?

Both ;)

In unpatched next branch the build on arc fails with
"error atomic ops for this platform are not implemented", no cxxflags nor 
CONF_OPTS will solve that.

Only when bumping to 10.3 you get past this build error just to see the 
"undefined reference to `__sync_fetch_and_add_8'" error later on.
To solve this the changes in this patch are needed.

Regards, Bernd

[Buildroot] [PATCH/next 2/2] package/mariadb: bump version to 10.3.9

[Thomas Petazzoni]

Hello,

On Sat, 25 Aug 2018 23:26:48 +0200, Bernd Kuhls wrote:

> In unpatched next branch the build on arc fails with
> "error atomic ops for this platform are not implemented", no cxxflags nor 
> CONF_OPTS will solve that.
> 
> Only when bumping to 10.3 you get past this build error just to see the 
> "undefined reference to `__sync_fetch_and_add_8'" error later on.
> To solve this the changes in this patch are needed.

So it still needs some atomic intrinsics, so the package should:

	depends on BR2_TOOLCHAIN_HAS_ATOMIC || BR2_TOOLCHAIN_HAS_SYNC_4

Indeed, atomic intrinsics are not always available, even when you link
with -latomic: gcc versions before gcc 4.8 did not have libatomic.
Build a SPARC gcc 4.7 toolchain, and try to build this package, you'll
be surprised :-)

Best regards,

Thomas
-- 
Thomas Petazzoni, CTO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com

[Buildroot] [PATCH/next 2/2] package/mariadb: bump version to 10.3.9

[Bernd Kuhls]

Am Sat, 25 Aug 2018 23:39:58 +0200 schrieb Thomas Petazzoni:

> Indeed, atomic intrinsics are not always available, even when you link
> with -latomic: gcc versions before gcc 4.8 did not have libatomic. Build
> a SPARC gcc 4.7 toolchain, and try to build this package, you'll be
> surprised :-)

Hi Thomas,

ok, will also do that. Where can I get such a toolchain? Could you please 
point me to a defconfig?

Regards, Bernd

[Buildroot] [PATCH/next 2/2] package/mariadb: bump version to 10.3.9

[Thomas Petazzoni]

Hello,

On Sat, 25 Aug 2018 23:54:10 +0200, Bernd Kuhls wrote:

> ok, will also do that. Where can I get such a toolchain? Could you please 
> point me to a defconfig?

I don't have anything like that readily available, unfortunately. You'd
have to pick a Buildroot version that is old enough to still support
gcc 4.7, or see if Crosstool-NG can help here.

Unfortunately, due to the autobuild.b.o server crash in October 2017, I
lost all pre-built toolchains I had built before that :-/

Best regards,

Thomas
-- 
Thomas Petazzoni, CTO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com

[Buildroot] [PATCH/next 2/2] package/mariadb: bump version to 10.3.9

[Bernd Kuhls]

Am Sat, 25 Aug 2018 23:39:58 +0200 schrieb Thomas Petazzoni:

> Hello,
> 
> On Sat, 25 Aug 2018 23:26:48 +0200, Bernd Kuhls wrote:
> 
>> Only when bumping to 10.3 you get past this build error just to see the
>> "undefined reference to `__sync_fetch_and_add_8'" error later on.
>> To solve this the changes in this patch are needed.
> 
> So it still needs some atomic intrinsics, so the package should:
> 
> 	depends on BR2_TOOLCHAIN_HAS_ATOMIC || BR2_TOOLCHAIN_HAS_SYNC_4

Hi Thomas,

why "BR2_TOOLCHAIN_HAS_SYNC_4" and not "BR2_TOOLCHAIN_HAS_SYNC_8"?

Regards, Bernd

[Buildroot] [PATCH/next 2/2] package/mariadb: bump version to 10.3.9

[Thomas Petazzoni]

Hello,

On Sun, 26 Aug 2018 13:48:59 +0200, Bernd Kuhls wrote:

> > So it still needs some atomic intrinsics, so the package should:
> > 
> > 	depends on BR2_TOOLCHAIN_HAS_ATOMIC || BR2_TOOLCHAIN_HAS_SYNC_4  
> 
> Hi Thomas,
> 
> why "BR2_TOOLCHAIN_HAS_SYNC_4" and not "BR2_TOOLCHAIN_HAS_SYNC_8"?

I misread the error message you pasted. Indeed it should be
BR2_TOOLCHAIN_HAS_SYNC_8. Sorry about that :-/

Best regards,

Thomas
-- 
Thomas Petazzoni, CTO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com