[Buildroot] [PATCH] berkeleydb: accommodate threads presence/absence

[Buildroot] [PATCH] berkeleydb: accommodate threads presence/absence

[Alexey Brodkin]

In BerkeleyDB the most convenient implementation of mutexes on Linux is
via POSIX mutex, and that requires pthread library.

Still it is possible to build (and hopefully use) BerkeleyDB without
mutexes. For this we pass "--disable-mutexsupport" during configuration
of the package.

Fixes:
http://autobuild.buildroot.net/results/717f3b37600a56262badc6f7cb64d7949fdacb67/
http://autobuild.buildroot.net/results/80ebf0382992b277fd94743815bbf0c7426a3654/

Signed-off-by: Alexey Brodkin <abrodkin@synopsys.com>
Cc: Yann E. MORIN <yann.morin.1998@free.fr>
Cc: Peter Korsgaard <peter@korsgaard.com>
---
 package/berkeleydb/berkeleydb.mk | 1 +
 1 file changed, 1 insertion(+)

diff --git a/package/berkeleydb/berkeleydb.mk b/package/berkeleydb/berkeleydb.mk
index 30fc518..e5c11bd 100644
--- a/package/berkeleydb/berkeleydb.mk
+++ b/package/berkeleydb/berkeleydb.mk
@@ -40,6 +40,7 @@ define BERKELEYDB_CONFIGURE_CMDS
 		$(SHARED_STATIC_LIBS_OPTS) \
 		--with-pic \
 		--enable-o_direct \
+		$(if $(BR2_TOOLCHAIN_HAS_THREADS),--enable-mutexsupport,--disable-mutexsupport) \
 	)
 endef
 
-- 
2.4.3

[Buildroot] [PATCH] berkeleydb: accommodate threads presence/absence

[Thomas Petazzoni]

Dear Alexey Brodkin,

On Mon,  6 Jul 2015 17:48:04 +0300, Alexey Brodkin wrote:
> In BerkeleyDB the most convenient implementation of mutexes on Linux is
> via POSIX mutex, and that requires pthread library.
> 
> Still it is possible to build (and hopefully use) BerkeleyDB without
> mutexes. For this we pass "--disable-mutexsupport" during configuration
> of the package.
> 
> Fixes:
> http://autobuild.buildroot.net/results/717f3b37600a56262badc6f7cb64d7949fdacb67/
> http://autobuild.buildroot.net/results/80ebf0382992b277fd94743815bbf0c7426a3654/

These build failures can no longer occur today: our toolchain
configuration for ARC does not explicitly disable threads. But they
were disabled at a specific point in time due to uClibc-ng being chosen
instead of uClibc-ARC, and with uClibc-ng, no thread implementation
could be selected for ARC, so it was default to no threads.

That being said, this isn't a reason to not fix the problem. However,
testing with an ARM no-thread toolchain
(http://autobuild.buildroot.org/toolchains/configs/br-arm-full-nothread.config),
I can't reproduce the problem. The build spits out many warnings:

#   warning requested reentrant code, but thread support was disabled

But it does build until the end, even though the mutexes fail also fails:

configure:20510: checking for mutexes
configure:20603: /home/thomas/projets/buildroot/output/host/usr/bin/arm-linux-gcc -o conftest -D_LARGEFILE_SOURCE -D_LARGEFILE64_SOURCE -D_F
ILE_OFFSET_BITS=64   -Os   -D_LARGEFILE_SOURCE -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -D_REENTRANT  conftest.c  -lpthrea
d >&5
In file included from /home/thomas/projets/buildroot/output/host/usr/arm-buildroot-linux-uclibcgnueabi/sysroot/usr/include/stdlib.h:25:0,
                 from conftest.c:43:
/home/thomas/projets/buildroot/output/host/usr/arm-buildroot-linux-uclibcgnueabi/sysroot/usr/include/features.h:209:5: warning: #warning req
uested reentrant code, but thread support was disabled [-Wcpp]
 #   warning requested reentrant code, but thread support was disabled
     ^
conftest.c:44:21: fatal error: pthread.h: No such file or directory
 #include <pthread.h>

Aah, but then it does:

checking for mutexes... ARM/gcc-assembly

so it falls back to using some internal mutex implementation. Which
explains why it doesn't build on ARC, but builds on ARM.

> diff --git a/package/berkeleydb/berkeleydb.mk b/package/berkeleydb/berkeleydb.mk
> index 30fc518..e5c11bd 100644
> --- a/package/berkeleydb/berkeleydb.mk
> +++ b/package/berkeleydb/berkeleydb.mk
> @@ -40,6 +40,7 @@ define BERKELEYDB_CONFIGURE_CMDS
>  		$(SHARED_STATIC_LIBS_OPTS) \
>  		--with-pic \
>  		--enable-o_direct \
> +		$(if $(BR2_TOOLCHAIN_HAS_THREADS),--enable-mutexsupport,--disable-mutexsupport) \
>  	)
>  endef

This doesn't get rid of the -D_REENTRANT definition that configure.ac
unconditionally add, but oh well, that's berkeleydb mess.

Conclusion: patch applied, thanks!

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