[Buildroot] [PATCH v3 3/8] postgresql: enable bfin compile

[Buildroot] [PATCH v3 3/8] postgresql: enable bfin compile

[Peter Seiderer]

Hello Thomas,

On Sun, Mar 15, 2015 at 07:47:53PM +0100, Thomas Petazzoni wrote:
> Dear Peter Seiderer,
> 
> On Tue, 10 Mar 2015 20:59:39 +0100, Peter Seiderer wrote:
> 
> > +# unsupported platform, overwrite manually
> > +ifeq ($(BR2_bfin),y)
> > +POSTGRESQL_CONF_OPTS += --with-template=linux
> > +endif
> 
> Do we really need to make this conditional? I believe
> --with-template=linux is fine in all cases since Buildroot only targets
> Linux.
> 

Did take a look at postgresql-9.4.1/configure.in, the decision is made on
a variable called $host_os, but used for (what I would call) $target_os:

- for the bfin test case:
checking build system type... x86_64-unknown-linux-gnu
checking host system type... bfin-buildroot-uclinux-uclibc

- for a buildroot toolchain:
checking build system type... x86_64-unknown-linux-gnu
checking host system type... arm-buildroot-linux-uclibcgnueabi

> So what about having --with-template=linux unconditionally in
> POSTGRESQL_CONF_OPTS instead?
> 

Yes, I think this is save...

> If that's OK for you, I can adjust this when applying.
>

Yes, thats fine with me...

Regards,
Peter
 
> Thanks,
> 
> Thomas
> -- 
> Thomas Petazzoni, CTO, Free Electrons
> Embedded Linux, Kernel and Android engineering
> http://free-electrons.com

[Buildroot] [PATCH v3 3/8] postgresql: enable bfin compile

[Thomas Petazzoni]

Dear Peter Seiderer,

On Mon, 16 Mar 2015 20:40:03 +0100, Peter Seiderer wrote:

> Did take a look at postgresql-9.4.1/configure.in, the decision is made on
> a variable called $host_os, but used for (what I would call) $target_os:
> 
> - for the bfin test case:
> checking build system type... x86_64-unknown-linux-gnu
> checking host system type... bfin-buildroot-uclinux-uclibc
> 
> - for a buildroot toolchain:
> checking build system type... x86_64-unknown-linux-gnu
> checking host system type... arm-buildroot-linux-uclibcgnueabi

Yes, the "linux" OS is recognized, but not "uclinux", I saw that. But
it means that passing --with-template=linux unconditionally is fine.

> > If that's OK for you, I can adjust this when applying.
> 
> Yes, thats fine with me...

Ok, thanks.

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com

[Buildroot] [PATCH v3 3/8] postgresql: enable bfin compile

[Peter Seiderer]

Hello Thomas,

On Tue, Mar 17, 2015 at 09:29:39PM +0100, Thomas Petazzoni wrote:
> Dear Peter Seiderer,
> 
> On Tue, 17 Mar 2015 21:26:02 +0100, Peter Seiderer wrote:
> 
> > The only minor purpose leftover is as gcc ICE exampel for 'bfin-uclinux-gcc (ADI-2014R1-RC2) 4.3.5' (see [1])...
> 
> I don't understand, why do you say it's the "only minor purpose
> leftover" ? 
> 
> If PostgreSQL uses fork(), it *cannot* work on noMMU architecture, so
> there is no way to build PostgreSQL on Blackfin.
>

...should have added 'until Blackfin compile is disabled because of noMMU' as a footnote ;-)

But the remaining question: Revert the patch? Should I send a revert-patch?
Or can you revert/commit directly? Or leave as is?

Regards,
Peter

> Thomas
> -- 
> Thomas Petazzoni, CTO, Free Electrons
> Embedded Linux, Kernel and Android engineering
> http://free-electrons.com

[Buildroot] [PATCH v3 3/8] postgresql: enable bfin compile

[Thomas Petazzoni]

Dear Peter Seiderer,

On Tue, 17 Mar 2015 21:45:48 +0100, Peter Seiderer wrote:

> ...should have added 'until Blackfin compile is disabled because of noMMU' as a footnote ;-)
> 
> But the remaining question: Revert the patch? Should I send a revert-patch?
> Or can you revert/commit directly? Or leave as is?

Well, please send a patch that adds the BR2_USE_MMU dependency, and as
part of it, removes the attempts at supporting Blackfin.

Thanks a lot!

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com

[Buildroot] [PATCH v3 3/8] postgresql: enable bfin compile

[Peter Seiderer]

Hello Thomas,

On Tue, Mar 17, 2015 at 08:59:23PM +0100, Thomas Petazzoni wrote:
> Dear Peter Seiderer,
> 
> On Tue, 17 Mar 2015 20:52:03 +0100, Peter Seiderer wrote:
> 
> > Will provide a patch for postgresql to check for fork/BR2_USE_MMU, is there any special string
> > for the comment besides 'needs threads'?
> 
> No there is no comment about a BR2_USE_MMU dependency, because it's not
> something the user can "fix" : if your architecture doesn't have a MMU,
> it's like this, there's nothing you can do about it.
> 
> > No special point, had the build environment still from [1] around...
> > 
> > In retrospect (in case bfin is disabled because of fork) the patch subject should
> > have been something different...
> 
> Well, the patch by itself is then completely useless, because Blackfin
> is a noMMU architecture, so postgresql will never work on it.
>

You are right (did not know that the 'uclinux' in 'bfin-buildroot-uclinux-uclibc' stands
for microcontroller Linux/noMMU)...

The only minor purpose leftover is as gcc ICE exampel for 'bfin-uclinux-gcc (ADI-2014R1-RC2) 4.3.5' (see [1])...

Revert the patch or keep it?

Regards,
Peter

[1] http://lists.busybox.net/pipermail/buildroot/2015-March/121094.html

> Thomas
> -- 
> Thomas Petazzoni, CTO, Free Electrons
> Embedded Linux, Kernel and Android engineering
> http://free-electrons.com

[Buildroot] [PATCH v3 3/8] postgresql: enable bfin compile

[Thomas Petazzoni]

Dear Peter Seiderer,

On Tue, 17 Mar 2015 21:26:02 +0100, Peter Seiderer wrote:

> The only minor purpose leftover is as gcc ICE exampel for 'bfin-uclinux-gcc (ADI-2014R1-RC2) 4.3.5' (see [1])...

I don't understand, why do you say it's the "only minor purpose
leftover" ? 

If PostgreSQL uses fork(), it *cannot* work on noMMU architecture, so
there is no way to build PostgreSQL on Blackfin.

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com

[Buildroot] [PATCH v3 3/8] postgresql: enable bfin compile

[Peter Seiderer]

Hello Thomas,

On Mon, Mar 16, 2015 at 11:07:25PM +0100, Thomas Petazzoni wrote:
> Dear Peter Seiderer,
> 
> On Tue, 10 Mar 2015 20:59:39 +0100, Peter Seiderer wrote:
> > - overwrite platform support with linux template
> > - disable spinlock support
> > 
> > Signed-off-by: Peter Seiderer <ps.report@gmx.net>
> 
> postgresql uses fork(), which is not available on Blackfin:
> http://autobuild.buildroot.org/results/c9c/c9c93156fe64e89f35627aa592c47d8c5181ed7d/build-end.log.
> 

Will provide a patch for postgresql to check for fork/BR2_USE_MMU, is there any special string
for the comment besides 'needs threads'?

> So what was the point of enabling postgresql on Blackfin ?

No special point, had the build environment still from [1] around...

In retrospect (in case bfin is disabled because of fork) the patch subject should
have been something different...

Regards,
Peter

[1] http://git.buildroot.net/buildroot/commit/?id=fc2e91d5bd32047055a611ed0d62f050bc90377c

> 
> Thomas
> -- 
> Thomas Petazzoni, CTO, Free Electrons
> Embedded Linux, Kernel and Android engineering
> http://free-electrons.com

[Buildroot] [PATCH v3 3/8] postgresql: enable bfin compile

[Thomas Petazzoni]

Dear Peter Seiderer,

On Tue, 17 Mar 2015 20:52:03 +0100, Peter Seiderer wrote:

> Will provide a patch for postgresql to check for fork/BR2_USE_MMU, is there any special string
> for the comment besides 'needs threads'?

No there is no comment about a BR2_USE_MMU dependency, because it's not
something the user can "fix" : if your architecture doesn't have a MMU,
it's like this, there's nothing you can do about it.

> No special point, had the build environment still from [1] around...
> 
> In retrospect (in case bfin is disabled because of fork) the patch subject should
> have been something different...

Well, the patch by itself is then completely useless, because Blackfin
is a noMMU architecture, so postgresql will never work on it.

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com

[Buildroot] [PATCH v3 1/8] postgresql: enable uclibc build

[Peter Seiderer]

Build with uclibc and locale support failes with the following error
message:

regc_pg_locale.c: In function ?pg_wc_isdigit?:
regc_pg_locale.c:312:6: error: dereferencing pointer to incomplete type
      isdigit_l((unsigned char) c, pg_regex_locale));

This can be fixed by overwriting the HAVE_LOCALE_T detection (thanks
to Alex Potapenko <opotapenko@gmail.com> for the hint [1]).

[1] http://lists.busybox.net/pipermail/buildroot/2015-March/121088.html

Signed-off-by: Peter Seiderer <ps.report@gmx.net>
---
Changes v1 --> v3:
  - no changes

Tested with the following buildroot defonfig:

BR2_arm=y
BR2_arm1176jzf_s=y
BR2_JLEVEL=1
BR2_KERNEL_HEADERS_VERSION=y
BR2_DEFAULT_KERNEL_VERSION="3.18"
BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_3_18=y
BR2_TOOLCHAIN_BUILDROOT_INET_IPV6=y
BR2_TOOLCHAIN_BUILDROOT_LOCALE=y
BR2_TARGET_GENERIC_GETTY_PORT="tty1"
BR2_LINUX_KERNEL=y
BR2_LINUX_KERNEL_CUSTOM_GIT=y
BR2_LINUX_KERNEL_CUSTOM_REPO_URL="git://github.com/raspberrypi/linux.git"
BR2_LINUX_KERNEL_CUSTOM_REPO_VERSION="246530ff8a4f302b8666c6d9cf509407d8c1257a"
BR2_LINUX_KERNEL_USE_CUSTOM_CONFIG=y
BR2_LINUX_KERNEL_CUSTOM_CONFIG_FILE="board/raspberrypi/linux-3.18-rpi-dt.defconfig"
BR2_LINUX_KERNEL_ZIMAGE=y
BR2_PACKAGE_RPI_FIRMWARE=y
BR2_PACKAGE_RPI_FIRMWARE_INSTALL_DTBS=y
BR2_PACKAGE_POSTGRESQL=y
---
 package/postgresql/Config.in     | 7 -------
 package/postgresql/postgresql.mk | 8 ++++++++
 2 files changed, 8 insertions(+), 7 deletions(-)

diff --git a/package/postgresql/Config.in b/package/postgresql/Config.in
index 3ef5e6c..519b2a1 100644
--- a/package/postgresql/Config.in
+++ b/package/postgresql/Config.in
@@ -1,9 +1,5 @@
 config BR2_PACKAGE_POSTGRESQL
 	bool "postgresql"
-	# PostgreSQL does not build against uClibc with locales
-	# enabled, due to an uClibc bug, see
-	# http://lists.uclibc.org/pipermail/uclibc/2014-April/048326.html.
-	depends on BR2_TOOLCHAIN_USES_GLIBC
 	help
 	  PostgreSQL is a powerful, open source object-relational
 	  database system.
@@ -16,6 +12,3 @@ config BR2_PACKAGE_POSTGRESQL
 	  archives with pg_dump and pg_restore.
 
 	  http://www.postgresql.org
-
-comment "postgresql needs a toolchain w/ glibc"
-	depends on !BR2_TOOLCHAIN_USES_GLIBC
diff --git a/package/postgresql/postgresql.mk b/package/postgresql/postgresql.mk
index 7dfef06..03154af 100644
--- a/package/postgresql/postgresql.mk
+++ b/package/postgresql/postgresql.mk
@@ -13,6 +13,14 @@ POSTGRESQL_LICENSE_FILES = COPYRIGHT
 POSTGRESQL_INSTALL_STAGING = YES
 POSTGRESQL_CONFIG_SCRIPTS = pg_config
 
+ifeq ($(BR2_TOOLCHAIN_USES_UCLIBC),y)
+# PostgreSQL does not build against uClibc with locales
+# enabled, due to an uClibc bug, see
+# http://lists.uclibc.org/pipermail/uclibc/2014-April/048326.html
+# so overwrite automatic detection and disable locale support
+POSTGRESQL_CONF_ENV += pgac_cv_type_locale_t=no
+endif
+
 ifneq ($(BR2_TOOLCHAIN_HAS_THREADS),y)
 	POSTGRESQL_CONF_OPTS += --disable-thread-safety
 endif
-- 
2.1.4

[Buildroot] [PATCH v3 3/8] postgresql: enable bfin compile

[Peter Seiderer]

- overwrite platform support with linux template
- disable spinlock support

Signed-off-by: Peter Seiderer <ps.report@gmx.net>
---
Changes v1 --> v3:
  - no changes
---
 package/postgresql/postgresql.mk | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/package/postgresql/postgresql.mk b/package/postgresql/postgresql.mk
index 7ebf795..76b19d5 100644
--- a/package/postgresql/postgresql.mk
+++ b/package/postgresql/postgresql.mk
@@ -13,6 +13,11 @@ POSTGRESQL_LICENSE_FILES = COPYRIGHT
 POSTGRESQL_INSTALL_STAGING = YES
 POSTGRESQL_CONFIG_SCRIPTS = pg_config
 
+# unsupported platform, overwrite manually
+ifeq ($(BR2_bfin),y)
+POSTGRESQL_CONF_OPTS += --with-template=linux
+endif
+
 ifeq ($(BR2_TOOLCHAIN_USES_UCLIBC),y)
 # overwrite IPV6 detection
 ifeq ($(BR2_INET_IPV6),y)
@@ -31,7 +36,7 @@ ifneq ($(BR2_TOOLCHAIN_HAS_THREADS),y)
 	POSTGRESQL_CONF_OPTS += --disable-thread-safety
 endif
 
-ifeq ($(BR2_microblazeel)$(BR2_microblazebe)$(BR2_nios2),y)
+ifeq ($(BR2_bfin)$(BR2_microblazebe)$(BR2_microblazeel)$(BR2_nios2),y)
 	POSTGRESQL_CONF_OPTS += --disable-spinlocks
 endif
 
-- 
2.1.4

[Buildroot] [PATCH v3 3/8] postgresql: enable bfin compile

[Thomas Petazzoni]

Dear Peter Seiderer,

On Tue, 10 Mar 2015 20:59:39 +0100, Peter Seiderer wrote:

> +# unsupported platform, overwrite manually
> +ifeq ($(BR2_bfin),y)
> +POSTGRESQL_CONF_OPTS += --with-template=linux
> +endif

Do we really need to make this conditional? I believe
--with-template=linux is fine in all cases since Buildroot only targets
Linux.

So what about having --with-template=linux unconditionally in
POSTGRESQL_CONF_OPTS instead?

If that's OK for you, I can adjust this when applying.

Thanks,

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com

[Buildroot] [PATCH v3 3/8] postgresql: enable bfin compile

[Thomas Petazzoni]

Dear Peter Seiderer,

On Tue, 10 Mar 2015 20:59:39 +0100, Peter Seiderer wrote:
> - overwrite platform support with linux template
> - disable spinlock support
> 
> Signed-off-by: Peter Seiderer <ps.report@gmx.net>

Applied, after making --with-template=linux used unconditionally, as we
discussed.

Thanks!

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com

[Buildroot] [PATCH v3 3/8] postgresql: enable bfin compile

[Thomas Petazzoni]

Dear Peter Seiderer,

On Tue, 10 Mar 2015 20:59:39 +0100, Peter Seiderer wrote:
> - overwrite platform support with linux template
> - disable spinlock support
> 
> Signed-off-by: Peter Seiderer <ps.report@gmx.net>

postgresql uses fork(), which is not available on Blackfin:
http://autobuild.buildroot.org/results/c9c/c9c93156fe64e89f35627aa592c47d8c5181ed7d/build-end.log.

So what was the point of enabling postgresql on Blackfin ?

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com