Re: [Boost-testing] Problems with debug info on Intel 9

From: Martin Wille (mw8329_at_[hidden])
Date: 2005-08-24 07:42:21

John Maddock wrote:
>>While we're on that ... is there any elegant way how we could suppress
>>linking/running some special test cases. I am still having the linker
>>problems with icc 9. These cause the need for several *manual*
>>interventions on *every* regression run. Affected are all *_lib tests of
>>Boost.Thread and all Boost.Wave tests.
> 
> 
> I believe those are all <runtime-link>static tests?  They all fail to link 
> on my machine because there are no static runtime libraries available, I'm 
> sure something similar is occurring for you, although why they hang rather 
> than just fail I don't know.  What happens if you add <runtime-link>dynamic 
> to your BUILD environment variable?  Those tests should all be skipped then.

You're right, all affected tests are <runtime-link>static. However, I 
can't tell whether there are any passing <runtime-link>static tests.

Static runtime libraries are available here (libcprts.a libcxa.a 
libcxaguard.a libguide.a libguide_stats.a libimf.a libirc.a libirc_s.a 
libompstub.a libsvml.a libunwind.a). I don't see any non-static 
libraries on the linker command line of, say, http://tinyurl.com/cdkk9
(librt.a and libpthread.a exist).

Is there a way to add <runtime-link>dynamic to my toolset definition 
instead of adding it to the commandline or the BUILD environment 
variable? (I'm only running tests for one toolset at a time)

All the failing tests also have <threading>multi. Could combining MT 
with static libraries cause the problem for icc9? A very simple test 
program using pthreads apparently works with -openmp -static (with and 
without -g).

Regards,
m
Send instant messages to your online friends http://au.messenger.yahoo.com