Re: Patch: initialize ProcessManager early

[Andrew Haley <aph at redhat dot com>]

Tom Tromey writes:
 > >>>>> "David" == David Daney <ddaney@avtrex.com> writes:
 > 
 > David> 1) Programs that never directly or indirectly do Runtime.exec() are
 > David> now burdened with the weight of an extra thread.
 > David> 2) Startup time for all programs will now be slower because this extra
 > David> thread is created sooner that it used to be.
 > 
 > Yeah, good points.
 > 
 > I'm not sure what else to do though.
 > 
 > My Plan B was to rearrange startup so that the user thread is not the
 > primary thread.  Then libgcj worker threads would not descend from the
 > user thread, circumventing this problem.
 > 
 > But for this approach to work, we have to start another new thread
 > anyway.
 > 
 > Hmm, maybe I could make the ProcessManager thread a child of the
 > finalizer thread.  That might work.

You have to wonder whether Java's Process / Streams interface is
really a sensible way to spawn a process that does name lookup for
stack traces .  An earlier version of this name lookup code was
written in C++ and didn't use Process at all.  Much more lightweight,
smaller, faster, etc.  And, most importantly, you don't want a
diagnostic to have side-effects on the system you're diagnosing.

Andrew.