[654] Integrated parallel testing into Test::Harness.

[Andy Armstrong]

On 1 Oct 2007, at 18:50, Andy Armstrong wrote:
> Multiplexing the input isn't nearly as fast as the fork based
> experiment in branches/paa - but the fork based version is very hard
> to wire up to an output formatter.

...because the events you want to generate output for are happening  
in multiple processes.

I haven't verified this but I assume the fork based approach is  
faster because it offloads TAP parsing to the individual processes.  
The multiplexed approach does all the parsing in a single process -  
so if that gets CPU bound it defines the overall runtime.

-- 
Andy Armstrong, Hexten