# File:		sgf/README

# Author:	Ken Friedenbach
# Copyright, 2009-2014. 
# All rights reserved
 
	
The SGF - FF[4] file format is defined in:
	http://www.red-bean.com/sgf/index.html

The sgf definition (with extensions) in this project is contained in:
	sgf_properties_spec.txt

The sgf package supports:
	reading and writing SGF files
	building, writing, and reading pattern libraries:
		fuseki (whole board) patterns
		joseki (corner) patterns
		side, quadrant, and center patterns
	
In addtion to normal SGF tree structures, this package supports:
	extensions to record wins in dictionaries:
		WB - Wins Black
		WW - Wins White
		WO - Wins Other (jigo, ?, Void (Left Unfinished), etc.)
		WC - Win Continue (TODO: used to point to continuation of games)
// TODO:
	extensions for Acyclic Directed Graphs (ADGs)
	extensions for very large trees/ADGs stored in multiple files

The package consists of the following files:
    findPatterns.go - walk SGF game trees and record patterns 
	game.go			- supports the data structures for storing a game
	interface.go	- defines the interfaces to the Parser
	parser.go		- implements a Parser for SGF files
	printer.go		- supports the writing of SGF files
	scanner.go		- implements a Scanner for SGF files
	sgf.go			- reads sgf_properties_spec.txt file and builds theProperties
	token.go		- defines tokens in SGF files
	tree.go			- defines the Nodes for SGF trees and ADG's

Notes on implementation:
	Mode 1: read and write the files in sgfdb Database, 
		and compare for equality (ignoring white space). (Compare time to JOSEKILIB)
	Mode 2: read sgfdb Database and write:
		whole board dictionaries
		corner dictionaries
		side dictionaries
		quadrant dictionaries
		center dictionaries
		player histories
	Mode 3: support of a player:
		read pieces of dictionaries
		maintain self and opponent histories

TODO: implement and remove comments below:
Thoughts on very large dictionaries:
	Need to add properties for:
		RP - root path (needed to read blocks not in memory, but not written out)
		DP - directory path (relative to root)
		FN - file name (for continuations)
		BL - block link node (for reference to lower block)

TODO: finish implementation and remove comments below:
Parser currently had variable sized slices of Nodes, Properties, and Strings.
	Sizes are: 
		Property:		- 48 (less than 100)
		TreeNode:		- 16 (64K per blcok)
		PropertyValue	- 24 (only a few, if ignore comments, times, etc.)
		String:			- 16 + len(str) (only a few, mostly setup and Game Info)
		... ( more printed by temp.go )

TODO: implement and remove comments below:
Algorithm for splitting a full block:
	Oldest children are the first on children lists.
	Walk the tree, depth-first, oldest first, marking tree Nodes.
	If an ADG link is found (it is a leaf) mark it and go to target.
		(push target)
		Mark parents of target until a marked ancester is found.
		(push any unmarked children)
	Pop
		do depth first search
	Whenever leaf is reached, Pop Stack.
		If stack is empty, check if about 50% is marked.
			If so, stop and split into marked and unmarked.
				If not, back track up tree until an unmarked Sibling is found. 
					Re-Start depth first from there.
Before splitting block, unmark from root down to first node with two marked children.