Re: [gnugo-devel] paul_3_15.7

[Arend Bayer]

> But I now think that is not true. Inge has pointed out
> that my method of timing is flawed. I just timed it a
> third time and found no appreciable slowdown. (The
> first timing can be ignored because I was running
> gnome at the time.)

Two comments:
1. If you use "time", and the "user" time doesn't agree with the "real"
time (within 0.something percent), then you can throw the timing away.
Really. (That's when you have stuff running in background.) I always
time my regressions runs out of habit, and this is consistently true.

2. For a patch such as Paul's it is much less effort to use node counts
instead (as it obivously can't have a measurable performance impact
otherwise), i.e. Gunnar's script. Pike probably doesn't come with most
distributions, but it is no trouble to install it either.
If we'd all switch over, we could also always keep current node counts in
BREAKAGE, or extend the script to produce useful summaries itself,
comparing the current run with the last one etc.

Arend