[Boost-testing] Testing & autoconfiguration failure

From: Vladimir Prus (ghost_at_[hidden])
Date: 2007-01-24 15:25:17

Hi,
I've just spend some time with Martin on IRC, trying to
diagnose a failure in his tests. What we ended up was:

1. Tests are run with:

("/b/r/rc/boost/tools/jam/src/bin.linux/bjam" --v2 
"-sBOOST_BUILD_PATH=/b/r/rc" 
"-sBOOST_ROOT=/b/r/rc/boost" 
"--toolset=gcc-3.4.5_linux_x86_64" 
-d2 --dump-tests  "--build-dir=/b/r/rc/results" >>"/b/r/rc/results/bjam.log" 2>&1).

2. The bjam.log says that tests are compiled with gcc-3.4.6. 3.4.6 happens to be
the version that's invoked just by 'gcc'.

3. When Martin reverts regression.py to revision 1.63.2.6, where regression.py
does not use the --toolset option, the problem disappears and tests are built
with the right compiler.

I think that Martin can post his user-config.jam or output of 
--debug-configuration, for anybody to debug this, but it's not the
first case where we have a problem with --toolset option. 

Should we stop passing it from regression.py?

- Volodya