This is the mail archive of the
mauve-patches@sourceware.org
mailing list for the Mauve project.
Re: RFC: Harness with no busy-waiting
- From: Mark Wielaard <mark at klomp dot org>
- To: Anthony Balkissoon <abalkiss at redhat dot com>
- Cc: mauve-patches at sources dot redhat dot com
- Date: Tue, 27 Jun 2006 10:41:14 +0200
- Subject: Re: RFC: Harness with no busy-waiting
- References: <1151350073.3361.12.camel@localhost.localdomain>
Hi Anthony,
On Mon, 2006-06-26 at 15:27 -0400, Anthony Balkissoon wrote:
> This removes the busy waiting from the Harness and replaces it with
> blocking read() calls using Sockets with specified timeouts.
>
> This is RFC because I'd like it to be tested to see if it speeds things
> up as much as Mark's workaround before it is checked in.
I don't really like this design. It puts even more stuff inside the
RunnerProcess which really should be as simple as possible so you can
easily run and debug it. Otherwise you might be trying to debug the
infrastructure and not the test itself. And by adding Sockets you a also
make it very hard/impossible to run single tests without the Harness to
really debug them. Please just use normal streams for communication
between the RunnerProcess and whatever calls it. And add a old-fashion
timeout with Object.wait()/notify() to the Harness, or if you want
something more fancy there add a nio.channel.Selector. But please don't
make the RunnerProcess depend on too much infrastructure.
Cheers,
Mark