Well, not exactly synchronous, but more synchronous, at least.
On 10/18/07, Evan Weaver <[EMAIL PROTECTED]> wrote:
> 1. Ah, I see. Try mongrel trunk again; it's synchronous now. I don't
> really see why it wasn't before, since a server can only have one
> @acceptor anyway.
>
> What kind of system were you getting the error on, since I can't
> reproduce it myself?
>
> 2. Filed, as you are probably aware.
>
> 3 & 4; still working.
>
> Evan
>
> On 10/18/07, Charles Oliver Nutter <[EMAIL PROTECTED]> wrote:
> > Evan Weaver wrote:
> > > All right, here goes:
> > >
> > > 1. I can't reproduce this. Is the problem that closing the acceptor
> > > thread doesn't block on JRuby? Or is it that the thread dies, but the
> > > system still takes a few moments to release the socket?
> >
> > There are multiple layers of threads here. First off, the acceptor
> > thread must be killed from outside. That's accomplished asynchronously
> > in the HttpServer#stop call, as follows:
> >
> > def stop
> > stopper = Thread.new do
> > exc = StopServer.new
> > @acceptor.raise(exc)
> > end
> > stopper.priority = 10
> > end
> >
> > There's no guarantee when this thread will run, even with a high
> > priority. Since execution continues in the calling thread (the one
> > running test_configurator) the calling thread's attempt to make a
> > connection could succeed (if the raise call hasn't been reached yet), be
> > connecting and then fail unexpectedly (if the acceptor is in the midst
> > of raising), or fail outright (if the raise has completed successfully
> > and the thread has terminated).
> >
> > At any rate, the call to stop is certainly not guaranteed to kill the
> > target server before it returns.
> >
> > This one will vary from machine to machine depending on speed, number of
> > cores, and OS-level threading characteristics, but I'm positive the test
> > is not thread-safe.
> >
> > > 2. FileUtils#mkdir_p causes test suites to not register if the mkdir_p
> > > argument already exists. Easily worked around but seems like a bug.
> > > Here's a minimal example:
> > >
> > > require 'test/unit'
> > > require 'fileutils'
> > > FileUtils.mkdir_p "some_dir"
> > >
> > > class TrueTest < Test::Unit::TestCase
> > > def test_true; assert true; end
> > > end
> > >
> > > It works as long as "some_dir" doesn't already exist, mkdir_p is not
> > > called, or mkdir_p is called from within a TrueTest instance method
> > > rather than at load time.
> >
> > That's really weird, but it matches my observation. Can you file this as
> > a bug in JIRA along with this simple test case?
> >
> > http://jira.codehaus.org/browse/JRUBY
> >
> > > 3. You have my JDK versions already... any more tips on how to debug
> > > this? I'm using a 32bit system, for what it's worth.
> >
> > puts?
> >
> > Debugging this stuff is still a little difficult. Hopefully jruby-debug
> > will come along soon.
> >
> > > 4. HttpServer is in the Mongrel module, so somehow it's not finding
> > > that reference appropriately. I haven't looked at this in detail yet.
> >
> > This also could be a bug. If you look into it and can confirm it, file
> > it in JIRA.
> >
> > - Charlie
> >
> > ---------------------------------------------------------------------
> > To unsubscribe from this list please visit:
> >
> > http://xircles.codehaus.org/manage_email
> >
> >
>
>
> --
> Evan Weaver
> Cloudburst, LLC
>
--
Evan Weaver
Cloudburst, LLC
---------------------------------------------------------------------
To unsubscribe from this list please visit:
http://xircles.codehaus.org/manage_email
