List Info

Thread: LAM: liblam.so is not linked to libutil on Linux
LAM: liblam.so is not linked to libutil on Linux
user name
 2006-07-06 12:48:35 
> -----Original Message-----
> From: lam-bounceslam-mpi.org 
> [mailto:lam-bounceslam-mpi.org] On Behalf Of
Dennis van Dok
> Sent: Thursday, July 06, 2006 8:19 AM
> To: General LAM/MPI mailing list
> Subject: Re: LAM: liblam.so is not linked to libutil on
Linux
> 
> >The LIBS here should already have
"-lutil" in it because 
> it was added
> >during the same AC_CHECK_LIB as when
LIBLAM_EXTRA_LIBS got "-lutil".
> >  
> Yeah, well, right below it says:
> 
> LIBS="$libs_SAVE"

Right, duh.

[Jeff reaches for more caffeine]
 
> Confirmed. I think you don't want it in there in
general, 
> because LIBS 
> is global for the entire build.

You're absolutely right -- I just wasn't awake enough to
see the
ramifications.

What threw me was how this could have worked at all without
your patch.
Digging into the LAM source a little more, I see that we
added
LIBLAM_EXTRA_LIBS to every place that we
also linked liblam.  But I
can see how this would not help external packages that do
not use the
LAM wrapper compilers.

Is there a reason that you did not use the wrapper compilers
to create
Rmpi, btw?  Using the wrapper compilers should automatically
add -lutil
for you.  Alternatively, you can use the --showme option to
extract the
flags that the wrappers add and invoke the underlying
compiler yourself,
etc.

-- 
Jeff Squyres
Server Virtualization Business Unit
Cisco Systems

_______________________________________________
This list is archived at http://www.l
am-mpi.org/MailArchives/lam/
LAM: liblam.so is not linked to libutil on Linux
user name
 2006-07-06 13:23:23 

  


  

    
Jeff Squyres (jsquyres) wrote:

xmb-rtp-215.amer.cisco.com"
 type="cite">
  
Right, duh.

[Jeff reaches for more caffeine]
  



Hey, it's still the July 4th aftermath 




xmb-rtp-215.amer.cisco.com"
 type="cite">
  

    
Confirmed. I think you don't want it in there in general, 
because LIBS 
is global for the entire build.
    

  

  

You're absolutely right -- I just wasn't awake enough to see the
ramifications.

What threw me was how this could have worked at all without your patch.
Digging into the LAM source a little more, I see that we added
LIBLAM_EXTRA_LIBS to every place that we also linked liblam.  But I
can see how this would not help external packages that do not use the
LAM wrapper compilers.
  



Exactly. I think I just added -lutil myself at an earlier stage.


xmb-rtp-215.amer.cisco.com"
 type="cite">
  
Is there a reason that you did not use the wrapper compilers to create
Rmpi, btw?



Well, that has to do with the way Rmpi is distributed. It is a source
tarball with a configure script, but you are supposed to feed it to R
by saying 'R CMD INSTALL Rmpi_0.5-2.tar.gz
--configure-args="--with-mpi=/opt/lam"' and it will build and install
it for you as a local R package.



I suppose I could pass the hint to the Rmpi maintainer to work this
into the configure script.



My point is that dynamic linking of shared libraries should be used, if
supported. I believe that e.g. Solaris doesn't.



Regards,



Dennis


-- 
D.H. van Dok :: Software Engineer :: www.nikhef.nl :: www.vl-e.nl




  

  
  
   
     
    






 [1-2]











   
 





about | contact  Other archives ( Real Estate discussion Medical topics )