pthread_cancel()

[Joel Sherrill]

I have been thinking more and it really isn't as bad as I 
made it out to be.

Joel Sherrill wrote:
> 
> Pattara Kiatisevi wrote:
> >
> > Hi,
> >
> > Is pthread_cancel working fine with RTEMS? It doesn't seem to cancel :)
> > I also can't find it here:
> > http://www.oarcorp.com/rtemsdoc-4.5.1-pre3/rtemsdoc/html/posix_users/posix_users00326.html
> 
> It is there but my reading is that it doesn't work like the Linux man
> page.

There is a man page although it is basically a shell.  POSIX called this
Thread Cancellation and made it a separate chapter.  This is why the
manual has an index. :)

http://www.oarcorp.com/rtemsdoc-4.5.1-pre3/rtemsdoc/html/posix_users/posix_users00362.html

> And the Linux man page does not look anything like the functionality
> that
> was in the draft spec this was originally implemented against.  Based
> upon
> the Linux man page, I would tend to treat this as a special test in
> the calls listed on the Linux man page.

And I still tend to think that's the right way to go.  And I think
the RTEMS implementation is mostly there. 

> I am sorry to say this but the semantics of the RTEMS implementation
> do not appear to be what any modern program using them would expect.
> 
> This raises the question.. how critical are they to the app you are
> porting?  If they are important, would you be willing (with advice/help)
> to fix the implementation?

I will make you an offer.  I will file a PR to track this with.  You
start 
by writing some test programs that fully exercise the functionality of 
the cancelation manager.  These should work on Linux and I will adapt
the
configurery to RTEMS.  That should really be enough to identify what's
deficient.

Then identify precisely the points in the FSU and Linux pthreads 
implementation where a cancellation actually occurs inside the 
services listed.  Submit those and let me review what's broken.  I 
suspect that fixing it is not really hard and mostly a matter of
inserting
a cancellation test at the right point in pthread_join, sigwait, etc.

And while I am fixing it, you can fill in the manager documentation
since
you will actually understand it then. :)

As an aside, I also remember that condition variables and cancellation
were
very difficult to figure out how they really worked in the draft POSIX 
specification we implemented against.  And cancellation points were not 
required to make the GNU Ada run-time work which is why we initially 
implemented POSIX threads.  This is not a defense of their status, only
some background.

> > Thanks,
> > Pattara
> >
> > --
> > Please avoid sending me Word or PowerPoint attachments.
> > See http://www.fsf.org/philosophy/no-word-attachments.html
> > ----------------------------------------------------------------------
> > Ott Pattara Kiatisevi                              T L W G
> > M.Sc. INFOTECH Student, Stuttgart, Germany      http://linux.thai.net/
> > ----------------------------------------------------------------------
> 
> --
> Joel Sherrill, Ph.D.             Director of Research & Development
> joel at OARcorp.com                 On-Line Applications Research
> Ask me about RTEMS: a free RTOS  Huntsville AL 35805
> Support Available                (256) 722-9985

-- 
Joel Sherrill, Ph.D.             Director of Research & Development
joel at OARcorp.com                 On-Line Applications Research
Ask me about RTEMS: a free RTOS  Huntsville AL 35805
   Support Available             (256) 722-9985