Re: [Chicken-users] Segfault - a hard one

[Vincent Manis]

On 2008-Aug-7, at 06:04, felix winkelmann wrote:

> It must be added that `thread-terminate!' is a "problematic"
> function: terminating a thread forcefully destroy an execution
> context that doesn't necessarily expect to be terminated and
> may leave any kind of system in an inconsistent state.
> In fact, `thread-terminate!' shouldn't be used at all. That is
> not meant as an excuse - chicken provides it, so it should
> work, but it is a dangerous device in general.

I'd be a bit stronger than Felix on this...fine if thread-terminate  
can be
made to work, but it's almost impossible to design correct software that
allows one thread (or process) to terminate another. Basically, the
victim thread might be in any state at all, and therefore it's more or  
less
impossible to ensure that whatever global invariants must be preserved
are indeed preserved when the thread actually dies.  Testing is also  
close
to impossible in the presence of thread termination.

Of course, this doesn't apply if the victim thread is allowed to  
intercept
the terminate request, and execute a shutdown handler, but then it  
really
isn't a true termination (the killer thread can't be sure that the  
victim is
in fact dead).

I'd prefer to have the manual document that to the best of our knowledge
thread-terminate works `correctly', but that its use is fraught with  
design
and testing problems, and therefore it should be avoided.

-- v