Date: 2001-03-19 18:50:26
--- In boost_at_y..., williamkempf_at_h... wrote:
> I didn't get the feeling that everyone agreed on this count, but if
> they do I feel a lot better :). For the short term I'll change the
> other error checks to standard C asserts, though I probably won't
> this done until tomorrow (there will be another upload later
> with other changes, however). I do think that a configurable
> assertion would be appropriate for Boost libraries in general, but
> don't want to tackle this idea specifically for Boost.Threads.
I've uploaded the latest changes. This time around I've implemented
Alexander's "algorithm 8/a" for Win32 conditions. His test case does
illustrate problems with the "traditional" ACE implementation of
condition variables once you know how to run it right (there's a
tt.bat that illustrates what's needed in the zip right now... I'll
consider drafting a more portable test before formal submission).
Using his algorithm the test case does not exhibit deadlocks or
spurious wakeups while also still passing the tests and the other
examples. The algorithm is quite complex with some subtle
implementation details that, frankly, I'm still not 100% sure about,
but I trust the test results and his knowledge.
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk