Re: [gnugo-devel] 5% increase in owl nodes

[bump]

> (Note that if you do cvs update -D ... you first have to convert
> 2004-04-23 2:00h into your local time zone. Whatever you do, afterwards
> CVS/Tag carries the CVS time stamp in UTC, so it must show above date
> and hour to obtain this version. (Yes, CVS sucks.))

This is very puzzling.

With GNU Go 3.5.6 I get the following node counts for nngs1,
one of the files for which you reported a difference:

10000 48705290
10001 66901
10002 326063
10003 65992223

I then got the version that you describe with
cvs update -D "2004-04-22 18:00h"
(The time difference to utc from here is -8 hours according
to the chart at http://setiathome.ssl.berkeley.edu/utc.html.)

With this version I got these node counts:

10000 47224984
10001 63620
10002 323214
10003 64419460

I then applied the patch large_scale_5_1.1e to the
older version. (You must run autoconf after applying
the patch.) got the following node counts:

10000 47224984
10001 63620
10002 323214
10003 64419460

This SEEMS to show that large_scale_5_1.1e is not the culprit.
Yet an examination of the patch obtained with:

cvs diff -u -D "2004-04-22 18:00h" -r rel-3-5-6 >mpatch

seems to show nothing else besides the large scale patch
that could account for this.

Finally, I backed out large_scale_5_1.1e from the older
version then applied mpatch created as above. This should
produce a version identical to 3.5.6, and indeed the node
counts for nngs1 are identical to 3.5.6.

So what is there in mpatch that is not in the large
scale patch?

Dan