笔划输入法查找算法示例（Lua实现）

公司同事最近利用谷歌拼音输入法实现了自己的拼音输入法，经过了解，最核心的就是一个trie（词典树）的构造和检索（这里不太介绍trie树了，google一搜一大把），于是今天就想实现了一个笔划输入法。大概的想法是：

1. 找一个所有汉字或者一、二级汉字的笔顺数据库
2. 用Lua将数据库读进来，构造一个trie树
   1. 每一个节点存一个笔划
   2. 每一个节点带一个子节点集合
   3. 每一个节点带一个汉字集合，表示到这一级时所有笔划组成的完整汉字
3. 检索时根据用户输入的笔划，检索到一个节点，然后按笔划顺序遍历子树
   1. 遍历子树可以给出所有以这些笔划开始的所有汉字，但是总不能一下显示出来吧，所以需要一个迭代器，每调用一次给出一个可能的值，这个迭代器用C实现比较复杂，但是用Lua实现简直就是小意思，直接将遍历子树的函数封装到一个coroutine中，每找到一个汉字就 yield(汉字) 即可

笔顺数据库

CSDN上可以下载到 http://download.csdn.net/detail/yyjlan/3766691

下载的mdb格式，我不太喜欢，Lua也不太喜欢。由于luasql支持odbc，所以可以将mdb文件加入到odbc数据源，然后载入后转成sqlite3的格式，方便以后使用，转换代码如下

require "luasql.odbc"
require "luasql.sqlite3"

odbc_env = luasql.odbc()

-- 将Access文件在控制面板->管理工具->数据源 中增加到用户DSN，名称是hzbs
odbc_conn = odbc_env:connect("hzbs")
odbc_cur = odbc_conn:execute("SELECT * FROM hzbs;")

sqlite_env = luasql.sqlite3()
sqlite_conn = sqlite_env:connect("hzbs.sqlite3.db")
sqlite_conn:execute("CREATE TABLE hzbs (id INTEGER primary key, hanzi TEXT, stroke_number INTEGER, stroke_order TEXT, unicode TEXT, gbk TEXT);")
sqlite_conn:setautocommit(false) -- start transaction

record = {}
while odbc_cur:fetch(record, "n") do
	local id = record[1]
	local hanzi = record[2]
	local stroke_number = record[3]
	local stroke_order = record[4]
	local unicode = record[5]
	local gbk = record[6]
	sqlite_conn:execute("INSERT INTO hzbs(id, hanzi, stroke_number, stroke_order, unicode, gbk) VALUES(" .. id .. ",\'" .. hanzi .. "\'," .. stroke_number .. ",\'" .. stroke_order .. "\',\'" .. unicode .. "\',\'" .. gbk .. "\');")
end

sqlite_conn:commit() -- commit the transaction
sqlite_conn:close()

odbc_cur:close()
odbc_conn:close()
odbc_env:close()

构造子树与检索

多的不说，直接看代码吧。代码写得有点乱，不过凑合看是没什么问题的。要运行代码必须要先安装 LuaForWindows

require "luasql.sqlite3"
require "wx"


function _T(s)
	return s
end

-- enum stroke_t {
local stroke_root = 0 -- for trie root, not a valid stroke
local stroke_heng = 1
local stroke_shu = 2
local stroke_pie = 3
local stroke_na = 4
local stroke_zhe = 5
local stroke_max = 5
local stroke_text = {_T"一", _T"丨", _T"丿", _T"丶", _T"乛"}
-- }

function new_node(stroke)
	return {stroke=stroke,  -- see stroke definition
		subnodes = {},  -- next strokes
		hanzis={} -- two or more hanzi could have the same stroke order
	}
end

function new_trie()
	return new_node(stroke_root)
end

-- insert hanzi and create the trie
function insert_hanzi(node, stroke_order, hanzi)
	local stroke, not_found_index
	for i = 1, #stroke_order do
		stroke = tonumber(stroke_order:sub(i,i))
		if node.subnodes[stroke] then
			node = node.subnodes[stroke]
		else
			not_found_index = i
			break
		end
	end
	if not_found_index then
		for i = not_found_index, #stroke_order do
			stroke = tonumber(stroke_order:sub(i,i))
			node.subnodes[stroke] = new_node(stroke)
			node = node.subnodes[stroke]
		end
	end
	table.insert(node.hanzis, hanzi)
end

-- 看看strokes数组组成的笔划顺序的节点是否存在，如果存在则返回节点
function find_node(root, strokes)
	local node = root

	if #strokes < 1 then
		return nil
	end

	for i, stroke in ipairs(strokes) do
		if node.subnodes[stroke] then
			node = node.subnodes[stroke]
		else
			return nil
		end
	end
	return node
end

function db_to_trie(db_name)
	local env = luasql.sqlite3()
	local conn = env:connect(db_name)
	local cur = conn:execute("SELECT hanzi,stroke_order FROM hzbs;")
	local trie = new_trie()

	record = {}
	while cur:fetch(record, "a") do
		insert_hanzi(trie, record.stroke_order, record.hanzi)
	end

	cur:close()
	conn:close()
	env:close()

	return trie
end

function get_hanzi_enumerator(root)
	local traverse

	traverse = function(node)
		for i = 1, #node.hanzis do
			coroutine.yield(node.hanzis[i])
		end

		for stroke = 1, stroke_max do
			if node.subnodes[stroke] then
				traverse(node.subnodes[stroke])
			end
		end
	end
	local co = coroutine.create(function () traverse(root) end)

	return (function ()
		local ret, hanzi = coroutine.resume(co)
		if not ret then -- already stopped
			return nil
		elseif hanzi == nil then -- the last call, no yield and no return value
			return nil
		else
			return hanzi
		end
	end)
end

---------------------------------------------------------------
-- GUI
local new_id = (function ()
	local id = wx.wxID_HIGHEST
	return (function ()
		id = id + 1
		return id
	end)
end)()

dialog = wx.wxDialog(wx.NULL, wx.wxID_ANY, _T"Lua笔划输入法演示",
	wx.wxDefaultPosition, wx.wxDefaultSize)
panel = wx.wxPanel(dialog, wx.wxID_ANY)
local main_sizer = wx.wxBoxSizer(wx.wxVERTICAL)

-- 横竖撇捺折 按钮
local stroke_label = wx.wxStaticText(panel, new_id(), _T"可选笔划")
local heng_button = wx.wxButton(panel, stroke_heng, stroke_text[stroke_heng])
local shu_button = wx.wxButton(panel, stroke_shu, stroke_text[stroke_shu])
local pie_button = wx.wxButton(panel, stroke_pie, stroke_text[stroke_pie])
local na_button = wx.wxButton(panel, stroke_na, stroke_text[stroke_na])
local zhe_button = wx.wxButton(panel, stroke_zhe, stroke_text[stroke_zhe])

local button_sizer = wx.wxBoxSizer(wx.wxHORIZONTAL)
button_sizer:Add(stroke_label, 0, wx.wxALIGN_LEFT+wx.wxALL, 5)
button_sizer:Add(heng_button, 0, wx.wxALIGN_LEFT+wx.wxEXPAND+wx.wxALL, 5)
button_sizer:Add(shu_button, 0, wx.wxALIGN_LEFT+wx.wxEXPAND+wx.wxALL, 5)
button_sizer:Add(pie_button, 0, wx.wxALIGN_LEFT+wx.wxEXPAND+wx.wxALL, 5)
button_sizer:Add(na_button, 0, wx.wxALIGN_LEFT+wx.wxEXPAND+wx.wxALL, 5)
button_sizer:Add(zhe_button, 0, wx.wxALIGN_LEFT+wx.wxEXPAND+wx.wxALL, 5)

main_sizer:Add(button_sizer, 0, wx.wxALIGN_LEFT+wx.wxEXPAND+wx.wxALL, 5)

-- 输入笔划列表
local input_label = wx.wxStaticText(panel, new_id(), _T"输入笔划")
local input_textctrl = wx.wxTextCtrl(panel, new_id(), "",
	wx.wxDefaultPosition, wx.wxDefaultSize, wx.wxTE_READONLY)
local input_backspace_button = wx.wxButton(panel, new_id(), _T"退格")
local input_clear_button = wx.wxButton(panel, new_id(), _T"清除")

local input_sizer = wx.wxBoxSizer(wx.wxHORIZONTAL)
input_sizer:Add(input_label, 0, wx.wxALIGN_LEFT+wx.wxALL, 5)
input_sizer:Add(input_textctrl, 1, wx.wxALIGN_LEFT+wx.wxEXPAND+wx.wxALL, 5)
input_sizer:Add(input_backspace_button, 0, wx.wxALL, 5)
input_sizer:Add(input_clear_button, 0, wx.wxALL, 5)
main_sizer:Add(input_sizer, 1, wx.wxALIGN_LEFT+wx.wxEXPAND+wx.wxALL, 5)

-- 备选汉字
local candidate_label = wx.wxStaticText(panel, new_id(), _T"备选汉字")
local candidate_sizer = wx.wxBoxSizer(wx.wxHORIZONTAL)
candidate_sizer:Add(candidate_label, 0, wx.wxALIGN_LEFT+wx.wxALL, 5)

local candidate_number = 5
function create_candidate_btn(num)
	local textctrls = {}
	for i= 1, num do
		textctrls[i] = wx.wxButton(panel, new_id(), "")
		candidate_sizer:Add(textctrls[i], 1, wx.wxALIGN_LEFT+wx.wxALL+wx.wxEXPAND, 5)
	end
	textctrls.start_id = textctrls[1]:GetId()
	textctrls.end_id = textctrls.start_id + candidate_number - 1
	return textctrls
end
local candidate_textctrls = create_candidate_btn(candidate_number)
main_sizer:Add(candidate_sizer, 1, wx.wxALIGN_LEFT+wx.wxALL+wx.wxEXPAND, 5)

-- 选择输出的汉字
local output_textctrl = wx.wxTextCtrl(panel, new_id(), "", wx.wxDefaultPosition,
	wx.wxSize(0, 100), wx.wxTE_MULTILINE)
local output_sizer = wx.wxBoxSizer(wx.wxHORIZONTAL)
output_sizer:Add(output_textctrl, 1, wx.wxALIGN_LEFT+wx.wxEXPAND+wx.wxALL, 5)
main_sizer:Add(output_sizer, 0, wx.wxALIGN_LEFT+wx.wxEXPAND+wx.wxALL, 0)

main_sizer:SetSizeHints(dialog)
dialog:SetSizer(main_sizer)

-- 必须加，否则不能正确退出程序
dialog:Connect(wx.wxEVT_CLOSE_WINDOW,
	function (event)
		dialog:Destroy()
		event:Skip()
	end)

-- 读入笔划数据库
local trie = db_to_trie("hzbs.sqlite3.db")

-- 输入的stroke数组
input_strokes = {}
get_next_candidate = nil

function update_candidate()
	if get_next_candidate == nil then
		for _,textctrl in ipairs(candidate_textctrls) do
			textctrl:SetLabel("")
		end
	else
		for _,textctrl in ipairs(candidate_textctrls) do
			local hanzi = get_next_candidate()
			if hanzi then
				textctrl:SetLabel(hanzi)
			else
				textctrl:SetLabel("")
			end
		end
	end
end

function update_input()
	local text = {}
	for _,stroke in ipairs(input_strokes) do
		table.insert(text, stroke_text[stroke])
	end

	input_textctrl:SetValue(table.concat(text, " "))
end

function insert_stroke(stroke)
	table.insert(input_strokes, stroke);
	local node = find_node(trie, input_strokes)
	if node == nil then
		table.remove(input_strokes) -- 删除不合法的输入
		-- BEEP
	else
		get_next_candidate = get_hanzi_enumerator(node)
		update_input()
		update_candidate()
	end
end

function remove_stroke()
	table.remove(input_strokes)
	local node = find_node(trie, input_strokes)
	if node == nil then
		get_next_candidate = nil
        else
                get_next_candidate = get_hanzi_enumerator(node)
        end

	update_input()
	update_candidate()
end

function clear_stroke()
	input_strokes = {}
	get_next_candidate = nil
	update_input()
	update_candidate()
end

dialog:Connect(wx.wxID_ANY, wx.wxEVT_COMMAND_BUTTON_CLICKED,
	function(event)
		local id = event:GetId()
		if id <= stroke_max then
			insert_stroke(id)
		elseif id >= candidate_textctrls.start_id and id <= candidate_textctrls.end_id then
			output_textctrl:AppendText(candidate_textctrls[id-candidate_textctrls.start_id+1]:GetLabel())
			clear_stroke()
		elseif id == input_backspace_button:GetId() then
			remove_stroke()
		elseif id == input_clear_button:GetId() then
			clear_stroke()
		end
	end)

dialog:Centre()
dialog:Show(true)
wx.wxGetApp():MainLoop()

打包下载

转好的数据库文件和源代码文件，可以去我的资源上下载（CSDN的上传太垃圾，我已经搬到cnblogs上了，文件上传到那里了，链接http://files.cnblogs.com/windtail/ime-win32.rar，再一看，居然CSDN上传文件延迟了一天才上来，悲剧http://download.csdn.net/detail/windtailljj/4476402），先来张截图~~