[LTP] Support for static builds

[LTP] Support for static builds

[Alex =?unknown-8bit?q?Benn=C3=A9e?=]

Hi,

I'm in the process of trying to package up an LTP build for QEMU's
various target architectures. For ease of testing I'd like to build all
the testcases as static binaries to avoid having to mess about with
chroots during the actual testing. However I've been running into
problems. I run configure as:

  ./configure CFLAGS=-pthread LDFLAGS=-static

This seems to avoid most of the linking problems I have (undefined
reference to `pthread_mutex_foo' etc). However it seems the convention
of passing CFLAGS down to sub-makes isn't uniform in the code base.

For example building on PowerPC I ran into this:

  gcc -pthread -g -O2 -fno-strict-aliasing -pipe -Wall -W -Wold-style-definition  -I/ltp.git/testcases/kernel/include -I/ltp.git/testcases/network/tcp_cmds/sendfile/../include -I../../../../inc
  lude -I../../../../include -I../../../../include/old/ -static  -L../../../../lib  testsf_c.c   -lltp -o testsf_c
  gcc -pthread -g -O2 -fno-strict-aliasing -pipe -Wall -W -Wold-style-definition  -I/ltp.git/testcases/kernel/include -I/ltp.git/testcases/network/tcp_cmds/sendfile/../include -I../../../../inc
  lude -I../../../../include -I../../../../include/old/ -static  -L../../../../lib  testsf_s.c   -lltp -o testsf_s
  gcc -static  -L../../../../lib  testsf_c6.o   -lltp -o testsf_c6
  gcc -static  -L../../../../lib  testsf_s6.o   -lltp -o testsf_s6
  ../../../../lib/libltp.a(safe_macros.o): In function `safe_getpwnam':
  /ltp.git/lib/safe_macros.c:119: warning: Using 'getpwnam' in statically linked applications requires at runtime the shared libraries from the glibc version used for linking
  testsf_s6.o: In function `main':
  /ltp.git/testcases/network/tcp_cmds/sendfile/testsf_s.c:59: warning: Using 'getaddrinfo' in statically linked applications requires at runtime the shared libraries from the glibc version used
   for linking
  ../../../../lib/libltp.a(tst_res.o): In function `tst_res__':
  /ltp.git/lib/tst_res.c:153: undefined reference to `pthread_mutex_lock'
  /ltp.git/lib/tst_res.c:199: undefined reference to `pthread_mutex_unlock'
  /ltp.git/lib/tst_res.c:199: undefined reference to `pthread_mutex_unlock'

And when I looked at my host built x86_64 build which completed fine it
hadn't actually done what I asked:

  17:10 alex@zen/x86_64  [sendfile/master@origin] >file testsf_s6
testsf_s6: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), dynamically linked, interpreter /lib64/ld-linux-x86-64.so.2, for GNU/Linux 2.6.32, BuildID[sha1]=17a6b13234c805f7a5499acc714ebf5197a7c4ea, not stripped

So my questions:

  - is this the correct way to build a static-linked LTP?
  - is the testsf_06 failure because its not using the common build machinery?

Cheers,

--
Alex Bennée

[LTP] Support for static builds

[Cyril Hrubis]

Hi!
> I'm in the process of trying to package up an LTP build for QEMU's
> various target architectures. For ease of testing I'd like to build all
> the testcases as static binaries to avoid having to mess about with
> chroots during the actual testing. However I've been running into
> problems. I run configure as:
> 
>   ./configure CFLAGS=-pthread LDFLAGS=-static
> 
> This seems to avoid most of the linking problems I have (undefined
> reference to `pthread_mutex_foo' etc). However it seems the convention
> of passing CFLAGS down to sub-makes isn't uniform in the code base.
> 
> For example building on PowerPC I ran into this:
> 
>   gcc -pthread -g -O2 -fno-strict-aliasing -pipe -Wall -W -Wold-style-definition  -I/ltp.git/testcases/kernel/include -I/ltp.git/testcases/network/tcp_cmds/sendfile/../include -I../../../../inc
>   lude -I../../../../include -I../../../../include/old/ -static  -L../../../../lib  testsf_c.c   -lltp -o testsf_c
>   gcc -pthread -g -O2 -fno-strict-aliasing -pipe -Wall -W -Wold-style-definition  -I/ltp.git/testcases/kernel/include -I/ltp.git/testcases/network/tcp_cmds/sendfile/../include -I../../../../inc
>   lude -I../../../../include -I../../../../include/old/ -static  -L../../../../lib  testsf_s.c   -lltp -o testsf_s
>   gcc -static  -L../../../../lib  testsf_c6.o   -lltp -o testsf_c6
>   gcc -static  -L../../../../lib  testsf_s6.o   -lltp -o testsf_s6
>   ../../../../lib/libltp.a(safe_macros.o): In function `safe_getpwnam':
>   /ltp.git/lib/safe_macros.c:119: warning: Using 'getpwnam' in statically linked applications requires at runtime the shared libraries from the glibc version used for linking
>   testsf_s6.o: In function `main':
>   /ltp.git/testcases/network/tcp_cmds/sendfile/testsf_s.c:59: warning: Using 'getaddrinfo' in statically linked applications requires at runtime the shared libraries from the glibc version used
>    for linking

This warnings are a known glibc problem, you cannot link certain
functions (nss) statically with glibc.

>   ../../../../lib/libltp.a(tst_res.o): In function `tst_res__':
>   /ltp.git/lib/tst_res.c:153: undefined reference to `pthread_mutex_lock'
>   /ltp.git/lib/tst_res.c:199: undefined reference to `pthread_mutex_unlock'
>   /ltp.git/lib/tst_res.c:199: undefined reference to `pthread_mutex_unlock'

Hmm, this should be covered by glibc pthread_mutex_* stubs, that are
used when there is no -pthread in CFLAGS. But in this case you passed
-pthread in cflags for build and it's missing from linking, which is
most likely the reason why you get these kind of warnings.

Looks like CFLAGS are not propagated correctly there for some reason.
But first of all you shouldn't be passing -pthread to the top-level
configure at all.

> And when I looked at my host built x86_64 build which completed fine it
> hadn't actually done what I asked:
> 
>   17:10 alex@zen/x86_64  [sendfile/master@origin] >file testsf_s6
> testsf_s6: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), dynamically linked, interpreter /lib64/ld-linux-x86-64.so.2, for GNU/Linux 2.6.32, BuildID[sha1]=17a6b13234c805f7a5499acc714ebf5197a7c4ea, not stripped
> 
> So my questions:
> 
>   - is this the correct way to build a static-linked LTP?

Passing -static in LDFLAGS should do, unless the test uses functions
from NSS, although as far as I can tell, static build is not tested at
all.

>   - is the testsf_06 failure because its not using the common build machinery?

I doubt so.

-- 
Cyril Hrubis
chrubis@suse.cz