[lsb-discuss] pthread_atfork

[Robert Schweikert]

On Tue, 2006-09-19 at 07:23 -0700, Wichmann, Mats D wrote:
> >Potential inconsistency?
> >
> >Architecture: x86-64
> >
> >Running appchck on the Intel openMP library (libguide.so) produces the
> >following output:
> >
> >Checking binary /opt/intel/cce/9.1.039/lib/libguide.so
> >Section .text1: Not recognized by name. Checking as type SHT_PROGBITS
> >Symbol syscall used, but not part of LSB_Modules
> >Symbol pthread_getattr_np used, but not part of LSB_Modules
> >
> >Notable here is that appchck does not complain about pthread_atfork
> >although this interface is used in the library:
> >
> >-> nm -Bg /opt/intel/cce/9.1.039/lib/libguide.so | grep pthread_at
> >                 U pthread_atfork
> >                 U pthread_attr_destroy@@GLIBC_2.2.5
> >                 U pthread_attr_getstack@@GLIBC_2.2.5
> >                 U pthread_attr_getstacksize@@GLIBC_2.2.5
> >                 U pthread_attr_init@@GLIBC_2.2.5
> >                 U pthread_attr_setdetachstate@@GLIBC_2.2.5
> >                 U pthread_attr_setstacksize@@GLIBC_2.2.5
> >
> >According to a database search pthread_atfork is an IA32 interface.
> 
> pthread_atfork is a very weird case.  It ends up getting
> resolved at link-edit time against libpthread_nonshared.a
> and so it's considered a source-only interface: it's fine
> appearing in source code, but should not appear in an
> executable binary as a symbol to be resolved by dynamic
> linking, as it will already have been resolved by static
> linking.

This would indicate a problem with the Intel library, as nm clearly
shows the symbol as unresolved and thus left to be resolved by the
loader. libpthread.so does have the symbol, if it's source only and
resolved at static link time the symbol would technically not be needed
in libptherad.so correct?

> 
> It may be the online database query tool ("dbadmin") isn't
> representing this state quite correctly - it's hard for me to
> go look at that right now but I'll try to take a look later.
> It is correct in the actual database: Interface.Iname is SrcOnly
> for this entry.

I also noticed this on IA64.

> 
-- 
Robert Schweikert                   MAY THE SOURCE BE WITH YOU
(Robert.Schweikert at abaqus.com)                 LINUX
ABAQUS Inc.
Phone : 401-276-7190
FAX : 401-276-4408