[Buildroot] [arc-buildroot] [autobuild.buildroot.net] arc build results for 2015-07-29

[Buildroot] [arc-buildroot] [autobuild.buildroot.net] arc build results for 2015-07-29

[Alexey Brodkin]

Hi Thomas,

On Thu, 2015-07-30 at 08:30 +0200, Thomas Petazzoni wrote:
> Those results are limited to the arc architecture.
> 
> Build statistics for 2015-07-29
> ===============================
> 
>         success : 4  
>        failures : 6  
>        timeouts : 0  
>           TOTAL : 10 
> 
> Classification of failures by reason
> ====================================
> 
>              libselinux-2.1.13 | 2 
>                    sudo-1.8.13 | 1 
>                       atf-0.21 | 1 
>                   bonnie-1.03e | 1 
>                   boost-1.58.0 | 1 
> 
> Detail of failures
> ===================
> 
>          arc |                       atf-0.21 | NOK | 
> http://autobuild.buildroot.net/results/1a5861da3fb6628c872f3635118bd1f5c7707893/
>          arc |                   bonnie-1.03e | NOK | 
> http://autobuild.buildroot.net/results/c2af14c4096424e41449c094666ffa0262f4e9f2/

Still pre-fixes toolchain is used:
BR2_TOOLCHAIN_EXTERNAL_URL="
http://autobuild.buildroot.org/toolchains/tarballs/br-arcle-hs38-full-2015.05-1014-g5ccde01.tar.bz2"

>          arc |                   boost-1.58.0 | NOK | 
> http://autobuild.buildroot.net/results/ea427b22dc541aafcefae09710d9b30c0fa5fae2/

Waiting for reply from Boost guys on how should we fix this properly.
Probably that's not a question to you but we may change uClibc's __GLIBC_MINOR__ from 2 to anything >= 8
and see how it goes. At least in boost proper eventfd() will be used.

>          arc |              libselinux-2.1.13 | NOK | 
> http://autobuild.buildroot.net/results/9c2f09fb7b2a594c7d22fe413e8bae843e8d227c/
>          arc |              libselinux-2.1.13 | NOK | 
> http://autobuild.buildroot.net/results/3835eeaec5b81885d5b99dcc8c0df2b30acd9ac8/

Hm, I was under impression we had that fixed:
------------------------->8--------------------
/tmp/ccucQq2P.s: Assembler messages:
/tmp/ccucQq2P.s:475: Error: internal error: fixup not contained within frag
------------------------->8--------------------

Will reiterate this one.

>          arc |                    sudo-1.8.13 | NOK | 
> 
> http://autobuild.buildroot.net/results/1b77539d08ffd950317a58cef8e2a4ce624c4710/
------------------------->8--------------------
/home/test/autobuild/instance-0/output/host/usr/arc-buildroot-linux-uclibc/sysroot/usr/lib/libc.a(getenv.os): In
function `__GI_getenv':
getenv.c:(.text+0x0): multiple definition of `getenv'
env_hooks.o:env_hooks.c:(.text+0x11c): first defined here
------------------------->8--------------------

I believe something similar to the issue Vincente fixed for Bash, see
http://git.buildroot.net/buildroot/commit/?id=9623ff651f83a350819719f10705552da9793205

Will look into that and prepare a patch.

-Alexey

[Buildroot] [arc-buildroot] [autobuild.buildroot.net] arc build results for 2015-07-29

[Alexey Brodkin]

Hi Thomas,

On Thu, 2015-07-30 at 12:06 +0300, Alexey Brodkin wrote:
> Hi Thomas,
> >          arc |                    sudo-1.8.13 | NOK | 
> > 
> > http://autobuild.buildroot.net/results/1b77539d08ffd950317a58cef8e2a4ce624c4710/
> ------------------------->8--------------------
> /home/test/autobuild/instance-0/output/host/usr/arc-buildroot-linux-uclibc/sysroot/usr/lib/libc.a(getenv.os): In
> function `__GI_getenv':
> getenv.c:(.text+0x0): multiple definition of `getenv'
> env_hooks.o:env_hooks.c:(.text+0x11c): first defined here
> ------------------------->8--------------------
> 
> I believe something similar to the issue Vincente fixed for Bash, see
> http://git.buildroot.net/buildroot/commit/?id=9623ff651f83a350819719f10705552da9793205
> 
> Will look into that and prepare a patch.

Looks like there's no simple way to resolve this issue with static building on "sudo" against uClibc.
There's no autoconf option to disable internal implementation of "getenv" and somehow "sudo" could be
built as a static binary on my x86 host (obviously with glibc).

I'd say that this has something to do with uClibc (how it differs in implementation of "getenv")
compared to glibc but this is out of my scope of knowledge.

Probably Waldemar might be interested in that issue as well.

And as a dirty hack we may disable "sudo" in static builds but I don't like this approach personally
knowing that could be done with another libc implementation.

And indeed this is arch-independent problem with uClibc.

-Alexey

[Buildroot] [arc-buildroot] [autobuild.buildroot.net] arc build results for 2015-07-29

[Waldemar Brodkorb]

Hi,
Alexey Brodkin wrote,

> Hi Thomas,
> 
> On Thu, 2015-07-30 at 12:06 +0300, Alexey Brodkin wrote:
> > Hi Thomas,
> > >          arc |                    sudo-1.8.13 | NOK | 
> > > 
> > > http://autobuild.buildroot.net/results/1b77539d08ffd950317a58cef8e2a4ce624c4710/
> > ------------------------->8--------------------
> > /home/test/autobuild/instance-0/output/host/usr/arc-buildroot-linux-uclibc/sysroot/usr/lib/libc.a(getenv.os): In
> > function `__GI_getenv':
> > getenv.c:(.text+0x0): multiple definition of `getenv'
> > env_hooks.o:env_hooks.c:(.text+0x11c): first defined here
> > ------------------------->8--------------------
> > 
> > I believe something similar to the issue Vincente fixed for Bash, see
> > http://git.buildroot.net/buildroot/commit/?id=9623ff651f83a350819719f10705552da9793205
> > 
> > Will look into that and prepare a patch.
> 
> Looks like there's no simple way to resolve this issue with static building on "sudo" against uClibc.
> There's no autoconf option to disable internal implementation of "getenv" and somehow "sudo" could be
> built as a static binary on my x86 host (obviously with glibc).
> 
> I'd say that this has something to do with uClibc (how it differs in implementation of "getenv")
> compared to glibc but this is out of my scope of knowledge.
> 
> Probably Waldemar might be interested in that issue as well.

I can reproduce it and will push it on my TODO list. Thanks for
reporting this.

best regards
 Waldemar

[Buildroot] [arc-buildroot] [autobuild.buildroot.net] arc build results for 2015-07-29

[Waldemar Brodkorb]

Hi,
Waldemar Brodkorb wrote,

> Hi,
> Alexey Brodkin wrote,
> 
> > Hi Thomas,
> > 
> > On Thu, 2015-07-30 at 12:06 +0300, Alexey Brodkin wrote:
> > > Hi Thomas,
> > > >          arc |                    sudo-1.8.13 | NOK | 
> > > > 
> > > > http://autobuild.buildroot.net/results/1b77539d08ffd950317a58cef8e2a4ce624c4710/
> > > ------------------------->8--------------------
> > > /home/test/autobuild/instance-0/output/host/usr/arc-buildroot-linux-uclibc/sysroot/usr/lib/libc.a(getenv.os): In
> > > function `__GI_getenv':
> > > getenv.c:(.text+0x0): multiple definition of `getenv'
> > > env_hooks.o:env_hooks.c:(.text+0x11c): first defined here
> > > ------------------------->8--------------------
> > > 
> > > I believe something similar to the issue Vincente fixed for Bash, see
> > > http://git.buildroot.net/buildroot/commit/?id=9623ff651f83a350819719f10705552da9793205
> > > 
> > > Will look into that and prepare a patch.
> > 
> > Looks like there's no simple way to resolve this issue with static building on "sudo" against uClibc.
> > There's no autoconf option to disable internal implementation of "getenv" and somehow "sudo" could be
> > built as a static binary on my x86 host (obviously with glibc).
> > 
> > I'd say that this has something to do with uClibc (how it differs in implementation of "getenv")
> > compared to glibc but this is out of my scope of knowledge.
> > 
> > Probably Waldemar might be interested in that issue as well.
> 
> I can reproduce it and will push it on my TODO list. Thanks for
> reporting this.

It is indeed uClibc specific. The reason seems to be some
differences in include/libc-symbols.h between uClibc and glibc.
In uclibc getenv object file contains a hidden symbol for
__GI_getenv, but in GNU libc it is a hiddden symbol getenv.

Sudo implements it's own getenv, and overwrites getenv() from C
library via adding "default" visibility. This works fine for glibc,
because getenv symbol is hidden. In uClibc it is supposed to be
hidden, too, but unfortunately there a alias is used, which indeed
it hidden. 

We could set getenv weak like with this:
diff --git a/libc/stdlib/getenv.c b/libc/stdlib/getenv.c
index d5db178..9b04d0f 100644
--- a/libc/stdlib/getenv.c
+++ b/libc/stdlib/getenv.c
@@ -27,4 +27,4 @@ char *getenv(const char *var)
     }
     return NULL;
 }
-libc_hidden_def(getenv)
+libc_hidden_weak(getenv)

This works, but I am not sure it is correct.
Any opinions on this?

best regards
 Waldemar