A few years ago Noel (I believe) did some work on locking of the output from a boost test run so that the compile and run output doesn't get interleaved between tests in the results. This never really worked for us on AIX. I'm not sure if anyone else has the same issue.

We have been running our tests non concurrently to work around this problem. However, we are starting to find that the large numbers of new tests added to Boost CVS has caused a spike in the time required to complete a full test run (we also run non incremental). We have a renewed interest in getting this to work properly in order to get the testing times under the 24+ hours it currently takes. Can anyone point us in the right direction of where and what we should be looking at? Any help would be appreciated.

Chris Cambly
XL C++ Compiler Development