Re: Boost.Threads draft library submission

From: williamkempf_at_[hidden]
Date: 2001-03-17 11:56:09

--- In boost_at_y..., williamkempf_at_h... wrote:
> --- In boost_at_y..., <boost_at_y...> wrote:
> > 
> > Hello,
> > 
> > This email message is a notification to let you know that
> > a file has been uploaded to the Files area of the boost 
> > group.
> > 
> >   File        : /threads/threads.zip 
> >   Uploaded by : williamkempf_at_h... 
> >   Description : Boost.Threads draft library submission 
> > 
> > You can access this file at the URL
> > 
> > http://groups.yahoo.com/group/boost/files/threads/threads.zip 
> > 
> > To learn more about file sharing for your group, please visit
> > 
> > http://help.yahoo.com/help/us/groups/files

This file has been updated once again.  Incorporated are fixes for 
all issues addressed so far (I hope).  Also incorporated was stronger 
error checking when calling the Win32 API.  The pthreads 
implementation has not been done in this regard because it is still 
not fully working correctly in the first place.  I'm still hoping 
someone volunteers to help with this, though if I find the time I'll 
try and at least diagnose the race condition this weekend.

I've also added a detail namespace in thread.hpp that includes some 
common functions needed by the examples and test programs.  These 
include thread creation, joining on all created threads and 
sleeping.  These are in the detail namespace because they will be 
replaced by more appropriate interfaces in Phase 2 and are not meant 
for general use in Phase 1.  As such they are poorly implemented and 
have some major design holes, but can be safely used by the examples 
and tests provided at this time.  So, unless a bug is found that 
prevents use for the specific examples given, these details should 
not be evaluated at this time.

Also added were two new examples.  One example is "starvephil" and 
illustrates thread starvation caused by condition variables, taken 
directly from Communicating Sequential Processes with Java.  The 
other is "tennis", an example provided by Alexander that I believe 
was meant to illustrate the problems found with the ACE 
implementation of condition variables, which both Boost.Threads and 
pthread-win32 were based on.  The problem is, I can not get this 
example to produce any "spurious wakeups" with the current 
Boost.Threads implementation, let alone stolen signals or deadlock as 
Alexander claims can occur.  When implementing the Boost.Threads 
condition type for Win32 I did discover a race condition in the ACE 
implementation which I fixed (and sent a report to Doug Schmidt, but 
he's not reachable at the moment to comment on my findings).  Maybe 
this race condition is what causes the problems addressed by 
Alexander and the fix I applied means that the Boost.Threads 
implementation is yet another algorithm that will fix the problems.  
Or maybe I'm mistaken about the tennis example being an example meant 
to illustrate the problems.  I don't know at this point and I am 
trying to converse with Alexander to diagnose any possible issues 
here.

Bill Kempf