[RFC] Fix PosixProcess by porting VMProcess from Classpath...

Wed Jul 14 16:14:00 GMT 2004

Andrew Haley wrote:
> Bryce McKinlay writes:
> > Andrew Haley wrote:
> > 
> > >Bryce McKinlay writes:
> > > > David Daney wrote:
> > > > 
> > > > >It is said that:
> > > > >
> > > > >http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11801
> > > > >
> > > > >Could be fixed by using VMProcess from Classpath instead of the current
> > > > >PosixProcess.
> > > > >
> > > > >There are several other problems that I have been experiencing related
> > > > >to not reaping terminated or failed Processes that would also be fixed.
> > > > > 
> > > > >
> > > > >Q1: Does this seem like a good idea?.
> > > > > 
> > > > >
> > > > 
> > > > Yes. If the general opinion is that Classpath's VMProcess is better than 
> > > > ours, then I am in favour of switching to/merging with the classpath 
> > > > version.
> > >
> > >Well, obviously that's true. However, we might also break things with
> > >this change, so we need really to be convinced that it is better in
> > >all respects. Having a single base implementation seems like a good
> > >this, certainly.
> > >
> > Our Process implementation is broken by design (see PR 11801), so
> > its pretty clear that a rewrite is required. Given that, we might
> > as well start with an existing implementation - and if this means
> > another class can be merged with Classpath, then all the better.
>> I agree. However, I'd be interested to know if Classpath's
> implementation of Process is stable and if anyone has ever used it for
> anything much. Forewarned is forearmed, as they say.

One thing I don't like about Classpath's VMProcess is that it polls for
terminated processes.
I was initially thinking about having the process forking thread be
blocking in waitpid() most of the time. Whenever a new Process was
needed, a signal would be sent to the forking thread to terminate the
waitpid() call.
The issues I had in my mind about this approach were:
1) Which signal to use.
2) I know I could make it work under the Linux kernel, but cannot test
on other Posix environments.
3) One would have to be careful about avoiding race conditions.
David Daney.