当前位置:首页>办公设备>考勤机>

中控考勤机ua100怎么样(中控考勤机ua100无法开机)

中控考勤机ua100怎么样(中控考勤机ua100无法开机)

更新时间:2022-03-22 13:26:36

今天刷到了一个这样的短视频,我寻思我是不是也可以写一个类似的上课点名程序,想法经不起等待,说写就写~

一.准备工作

私信小编01即可获取大量python学习资源

1.Tkinter

Tkinter 是 python 内置的 TK GUI 工具集。TK 是 Tcl 语言的原生 GUI 库。作为 python 的图形设计工具,它所使用的 Tcl 语言环境已经完全嵌入到了 python 解释器中。

我们使用Tkinter开发GUI界面。

2.PIL

PIL(Python Image Library)库是Python语言的第三方库,需要通过pip工具安装。安装PIL库的方法如下,需要注意,安装库的名字是pillow。

PIL库支持图像储存、显示和处理,他能够处理几乎所有图片格式,可以完成对图像的缩放、剪裁、叠加以及向图像添加线条、图像和文字等操作。

使用PIL中的Image,ImageTk处理、引入一张图片,可以使用下面代码安装一下。

pip install pillow

二.预览1.启动

双击打开后,进入软件主界面,所有功能一目了然。程序会自动识别软件目录下的names.txt,将里面的名字导入。

2.开始点名-顺序点名

选择顺序点名后,点击开始,屏幕上就开始滚动出现人名,人名出现的概率是相同的,点击停止,人名就停止滚动,点名结束。

3.开始点名-随机点名

点击随机点名,程序就会进行随机点名,人名出现的概率是随机的。

4.手动加载人名单

可以自己手动选择人名单,前提是人名单格式为txt,且每个名字占一行。

5.开始点名-顺序点名-Pyqt5版本

用Pyqt5也写了一个版本,实现逻辑与TK版本相同,界面可能更好看了一些,但是文件大了许多,大家可以在后面总结部分自取。

三.思路1.整体实现思路

2.点名实现思路

四.源代码point_names-GUI.py(主程序GUI)

import random import re import time import threading from tkinter import * from tkinter import ttk from base64 import b64decode from PIL import Image,ImageTk from tkinter import messagebox from tkinter.filedialog import askopenfilename """" 2021-11-10点名/抽奖程序 主要亮点: 1.两种模式: ①顺序点名 ②随机点名 2.自动识别人名单 3.支持手动导入人名单 4.人名单导入校验 5.人名显示位置自动矫正 6.最多显示五个大字 """ imgs=['./point_name.png'] class APP: def __init__(self): self.root = Tk() self.running_flag=False #开始标志 self.time_span=0.05 #名字显示间隔 self.root.title('Point_name-V1.0') width = 680 height = 350 left = (self.root.winfo_screenwidth() - width) / 2 top = (self.root.winfo_screenheight() - height) / 2 self.root.geometry("%dx%d %d %d" % (width, height, left, top)) self.root.resizable(0,0) self.create_widget() self.set_widget() self.place_widget() self.root.mainloop() def create_widget(self): self.label_show_name_var=StringVar() self.label_show_name=ttk.Label(self.root,textvariable=self.label_show_name_var,font=('Arial', 100,"bold"),foreground = '#1E90FF') self.btn_start=ttk.Button(self.root,text="开始",) self.btn_load_names=ttk.Button(self.root,text="手动加载人名单",) self.lf1=ttk.LabelFrame(self.root,text="点名方式") self.radioBtn_var=IntVar() self.radioBtn_var.set(1) self.radioBtn_sequence=ttk.Radiobutton(self.lf1,text="顺序点名",variable=self.radioBtn_var, value=1) self.radioBtn_random=ttk.Radiobutton(self.lf1,text="随机点名",variable=self.radioBtn_var, value=2) self.label_show_name_num=ttk.Label(self.root,font=('Arial', 20),foreground = '#FF7F50') paned = PanedWindow(self.root) self.img = imgs img_=b'iVBORw0KGgoAAAANSUhEUgAAALQAAAB4CAIAAADUhU qAAAACXBIWXMAAAsTAAALEwEAmpwYAAAgAElEQVR4nO196XNbx5Vvd9 LfSU2EgD3fRNJbZRkyVJseY9sP7scOy ZTCqpVKXm8/wp TA1X2amJjWZSXksy1Ik27IsWXIsRpK1kCIpLgBJkAAXrMQO3K3fh0O0rkBKkaONzvOJSwGBu3b/ uznNKaUoh/oyZB6bDHG2 RSD0/807mNmth7ql/ymbz8EyL2LvdbeH/bC245bk UHgQOSik8x8O85P0eHb5XX4pSSghRX7Pq nAwxhi 3zwWlFJFURBC7Dr3G68tv3/0UYYnZC91P8IYw3NSSjmO23zTv/okbPTYaz7NlXMXHJtf AHTw05hn9kxVaewi9AKVX2z Sz1uQ 4KaVUlmVUGTIGvs2j/52e/6/SZlg/4PqKokiSVPWQD39TgFfVuD1tzgGvUbWg2dPcj7b8VT09DAQIIVmWy VyqVSSZVl9I/bmWq1Wr9drNJrNs7uZtbAVqYbFlkey5atmXewF1S/7kC oZlqMC255uqIoq6urgUDAYrF0dnaaTKa/OqRVd69aP9/p9Eene8CBNjE6NUNDD4FZOB5YKLsmQkiW5ZWVlatXrwaDQUIIx3HqSVUURRAEo9F44MCBoaEhQgicxR4JnoHN8fr6 urqqt1udzqdWq2WUipJEntCRVFEUczn85IkAWthj8Fux54NY2w0Gu12u06ne/DbMSjALWRZhj8VRaliCer5m5mZ dd//Ve/3/9P//RPzc3NVQdseRYDgSzLmUwmm80aDAan0wnDAndnQ/RXRdujEM eEu4hy3I4HI7H436/v6amplQqlUolk8lEKZ2enl5eXsYYGwwGs9lss9lqamqsVqtWqyWEwKip51uNelEU79y58/vf/x5j/NZbbxmNRhhZODifz1 7du369euyLPf29vI8rx6pbDa7vLxMKYX5SyQSly5dmp2dfemllw4ePGg2m3meRwhxHMdmLhwOX7p0aXV1tVgsptPpQqFgNBoNBgOs8mKxuL6 Til1OBw2m627u/vIkSN1dXVMLWCkZoFAsiwrilIqlTDGOp1uM7tSLwmMsc1mk2X55s2bx48fr6 v12g0Wq3W4/F0dHTU1NSoeR4ATpZleLxIJBIMBgOBQKlU2r9//0svvWSxWAAW6oX6RAXNXc4BA5fNZr/44ouRkZF33323r69vZGQklUodPXrU4/Gk0 lIJCIIQrlczmazuVzOaDQODg4 99xzdXV1VaKkimRZTqfTsVisr6/v9ddft9vtsDgURVEUJZlMrq6ufv3118B11FeQJGlpaenMmTPz8/OCIFBKE4nE3NwcAGJubs5ut1utVrPZ3NnZ2dTUBG9hNBqbmppsNlsikRgbGwsEAkePHu3r6wM2Mz8//5e//CWfz//iF78YGhpyu916vf5hBgseeGlp6ZtvvnE4HLt373Y4HADKKimJKuzKZDLZbLY7d 4EAgGEEMZYFMVoNOrxeOx2OzA2WCSCIMzPz09PTweDweXl5WKxuLKyMj8/73K53G53Z2dnS0uLRqPhOE69eJ4o3eUc8D6ZTObOnTuxWIzn Xw /5e//GVxcbG/v7 5ufm5557bvXu3LMuFQiGVSt25c fEiRNXrlyRZfntt9 G8QUeK4pisVhkiphWq5VlGd4KY8zzPMgRjUbD8zxw5mw2azQanU4nzLpau6yrq3v efNZvPJkydnZ2e7urqOHTtmMpkY4HieN5lMdrvd5/NptVqO4zwej9PpxBgvLS19 23Kysr /bte/PNNzUajaIoo6OjIyMjq6urO3fuPHz4MDCMzeIcljKsVIbjRCJx4sSJTz755NVXX 3t7c3lcuVyGdgSE6aCICwtLYVCIUqpKIrALVwu1wsvvACMhOM4h8Oh5p3AsD/77LNwOOxyufbs2ePz QghoVAI Mfvfve7/v7 5557rrOz02KxMGXriSqnPCAdYC7Lci6XW19fNxqNNptNr9frdLpcLlcoFBBCWq1Wq9UqimK1Wl0ulyAIBoNBPXzwrIqiBAKB8 fPz8zMZDIZs9m8c fOgwcPMsgXCoXx8fErV6709/cfOnTIYDCUSqVUKlVTUwMciC0 JuaLxeLY2FipVHr33XePHTvW2dnJ83ypVCqXy8BpCSE1NTUg4EDnKBQKiqKsr68XCgVRFDOZTDKZBJgmEolisSgIwurqajAY1Gg0NpvNbrczjg0THAqF1tbWmpuba2pqQqHQ vp6bW3t3Nzc2bNnLRbL0NCQ0WgcGxv78ssv3W73Sy 91NTUBCtEEITLly9/ OGH5XLZ7XZnMhmtVjs2NpbNZmEAGxsbDx8 3NPTA4ISyO12/ hHPyoUCjU1NbW1tSB5BwcHs9lsMBgcGRn54osvbt68 bOf/ezgwYNwIwYOtRbyRMCRTqeTyeTk5GQkEiGEzMzMxGKxVCqVSqVGR0fr6 ubm5vhmSRJCoVCJ0 eDAQChw8f3rVrF8wKu6jH4xkcHIxGo fPn3e73Tt37uQ4Tq/XG41GhFCxWLx9 /bFixeNRuP /fthtqLRqN1ud7vdbJnCjdbW1r766quPP/74xo0b3d3dr7/ t69e3U6nSzLqVRqbGzM7/f39PQwQwAoHA5fvnw5n8 vra0tLCwkk8k///nPqVSK4zhBEBYXFyORSC6X /TTTycnJ81m8969e/fv328wGBgo8/n8119/febMmXfeeWdoaOjMmTMTExNHjhwJhULhcPjIkSM6nW5hYQGMkdOnTweDwX/8x3/s6ekBvlgqlRKJRGtr67vvvutyuUADK5fL4XD42rVrH330UTabra2tBfYAg2az2Twez6VLl65cudLU1NTa2gpqUKFQkCRJo9EQQnK5XC6Xo/c6jZ4c/9iQXuVy dtvvx0ZGZmcnJyfn8cYnzp1ymazTU1NJZPJa9euNTY2ulwurVYrimIul5uamrp165Yoih6PBzCOVBaNyWRqaGjw /0cx1mt1vr6ep7nLRZLR0cHsNNYLAbcGCEkiuLKykoymRwYGHA6neyFZVkOhULHjx /fPmyLMtOp1NRlHw v7CwUCwWC4XCyMjIf//3fw8PD7///vs nw ku8FgQAjpdDqbzYYxFgTBZrOtra3BrxqNRpIkURTNZjPG PDhw3v37gWRBBKHUppMJuPx MrKytTU1NTU1Llz55aWlm7dujU/P6/VakG3jUQi//Vf/wUMRpIkr9e7vLwcDAYbGhosFgvHcUajked5h8MBAgLmT5KkfD7v8/ng4Gg0Wltby8AB4 Z0Om/evHnu3Dmj0djf328ymRYXF2OxGMdx /btO3ToUF9fH7wjw/ETtFbg0hzHeb3ePXv21NbWZrNZhND777/f3Nx8 vTpCxcuvP7660ePHrVarblc7saNG3Nzc7lczuFwzM3N/fGPf4zH4729vRazubGxsaW1VafTLS8vf/HFF fPn0 lUqIonjhx4sUXX9y9e/evf/1rWFgrKyvZbHZmZubrr7/u7 /3eDwvv/zy0NAQQAd4gCAIIyMjExMTP/rRj3p7ez/88MNAILC8vLywsDA9PV0qlSKRSCQS fOf/5zP510uV0tLy8svv9zS0sLzvM/nq62tlWUZpiGZTB4 fPiNN97gOE5RlOnp6YmJiVAo1NbW1tPTo9FoUEXnyGazN2/evHz5cjabDYfDAFCHw6HRaNLpdDQaffnll996662VlZVgMGgymfx v9fr1ev1oih2dXUZDAYQnbW1tVarNZ1OF4tFeCNCiEajsVgsra2tPp9vbW1tbW0NTDOEkCRJYJ9LkuRwOBBCV65cuXLlik6nczqd3d3dnZ2dHo8nHo HQiFYaVVOqScCDkCuXq/v7e3t6OhYXFycnp5eXFx0Op1tbW319fUwauA5YI4HvV4/MDDgcrlu3rz5xRdfpNNpn88nK0ptXZ1Op9PpdPX19fX19aCIDQwM P1 i8Xi9/sppePj4 l0OpPJnD59emxs7J// Z93797d3d1tMBhAJINmp9Vqd 3a1dLS0tbWJsuy3W4XBEFRlAMHDuzevbtcLo Pj6 urur1 h07dvT09FgsFqvVyngsaLvlcjkajWq1WqfTqdfreZ6XJAmGFXQg8LigilNLr9d3d3e73W6tVjs1NVUqlfx //vvvw 8ze/3HzhwwO/3nz59enJy0u12cxyXyWSA c3MzLz99tsdHR2gQNTW1q6urk5MTGg0GngYEJe5XA6U/XQ6DUquLMtra2tff/317OysJEnRaHRtbc1msx0 fNjj8Wg0Gp1Ox/P8zMzM7du329rafvvb3/b09IC7CD2Er/JvBwf8H8YYzCS3293a2jo Pj41NQUTJknSyMjI8vJya2vr0aNHDx06dOjQIUmSFEUJhUKlUmlxcfGNN97o6 szGo1ms5njOJfLNTw8nEqlzp8/jxCCN4/FYsPDw0ajMRqNFotFl8uVyWRisVg n9dqtSaTSRTFeDyOMXY6nSBie3t7YdoSiQSYJ6IoNjY22mw2hJDFYjl37lwqlWppaXnxxRfV3g4QEKIoRiKRZDJps9lcLhcbSkmSACJgPdGKFxVcOK2tra2traIoUkq9Xm8mk4lGo7Is63S6 fn5CxcutLW1LS8v5/P5I0eOvPbaazqdbnFx8d/ 7d mp6Z6enqampp4nrfZbH6//9q1a7/73e98Ph9wFPDUlUqlmZkZvV7PfKxge3d1dbW2tlqt1tu3bweDQYvF8qtf/aqrq4t5X0ZGRkZHR6enp9PpNHoqTvS7sRUYHbPZ/MILL4RCoa //jqVSoXD4UQiMTU11dra2tLSYjabdTodQkir1ebz cXFxUQiMTg42N3dDYYGjPXCwsJnn3129epVm83mcDhyuZzVarXZbDqdDlZ8oVDYv39/Op0eHR0dHR11OBxmsxlu6vF4PvjgA7/fD7CAaeN53mq18jwP2lkqlcpkMrlczm63B4PB27dvd3R06HQ6kNnMaQE pba2tra2Nr/fD5NBCNHr9bW1tcDhgOEjlebPpIDL5err67t69eqlS5f0ej3GOB6PX79 nRnboiiy4zUaDa/RwEV4nvd4PPv37x8ZGZmfn9fpdJ2dnT6fb3Jykuf5Xbt2uVwuq9Xa1NSEKoaGxWIZGBhACImiGAwGZVnmeR6sRWYDAv8ol8ssovT0wAEPodVqu7q6fvOb39y4cSOVSpXL5ebm5ra2tnfeeae5uVmn08G6RAjl8/loNLpjx47XXnvN7/fDkMFg2e12cI04nU63222z2YxGI3D1VCpls9kOHDjw4osvajSazz//PBKJnDhxQqvV5nK5UqnU1dWl1WoRQozhcxyn1Wqbm5v7 vpAHl /fv3y5csgm9xu9/T09H/8x39oNJru7u6XXnoJTAAAwd69e9va2sxmM3AOWKYtLS2//e1vFUVpb28H5lFlCsIE19XVffDBBy 99BLP8xqNZmVlJRAI Hy zs7O69evLywsRKPRU6dOgaXtcDiGhobggoSQcrm8vLyczWYHBgZEUVxaWgLlZmVlZXh42O12g2UuSdLg4CAgj90aWFoul7t9 3Y8HgcuKEnSxMRENpt1u92wPmExM0/Jk8DKFuIKpGC5XBYEIRaLTU5OwigbjUZYyvBAYK2BEFEHJsClI0kSyHUWRmHWaTKZlCTJZrNxHJdOp7PZLPBbSilo GBWqC8IZyWTSavV6nQ6V1dXo9EoOwtECULIarV6vV6j0QgPCTIePqvHDiQ93O4Bg8scX7gSWCiXy2zmkslkuVxmV9BoNCaTyWKxaDQaWZbHxsb 5V/ ZWFh4Ze//GU2m71y5cquXbs4jvvqq6/27dt38ODB27dvX7p0ac ePT/72c9cLhd7BlmWZ2dnjx8/HggETCaT2l9MKbXZbLt27Tp48CCoO zVnh44mL0AKBEEAdimeo5hyOBINkDsdOb7U487VUUQUEWNYj4chBAsdzhXrYqzuYSzwOKAU AzXAQcssAJ4GrqkNiWY4e3CoLTewMl7AnxvdFz9qv6 nBAqVT67LPP/vCHPwCLslqt4XAYjNvl5eWamprGxkZJku7cuWM2m7u7u3U6HZzIUJjNZiF8zV4KdGcQNGzASYX lpl/CHqQoqtUiL2/2nfJFCV1yK1qTNXfqFcAUs0BG2L1WVjlcq1awVX/0k1BjaoDHgCOhyf1i7Mvq8wE9rlcLs/Ozs7MzNTV1Q0NDRkMBoYqptAghGRZhqVFK/kSjE wF1G/kXpA2E2rFKbHSw8Ch3pG1V y2CB9aA//lpfa oE2gWzLc7dcLpsB97iGTP0M92Mz6i/B1Qbakppxbj5LPfFMxn3Xx3sG4NiS1GP0hESd l7qsXvIZ3uij/QwpA7Zsye/H6S 0ws Zfpb/CdbrqHHSA9eo3/1yydNm6d28wGM56kF5ZZH/r2B4wf6/4SeQWnCD/R9oR/AsUHPREJtc/q gqPKfVIlvDdbffCBmQNVJiK61y79AR9A32Odo8rEpSrfHVJ5ZTZbVZvdKux79YcfILLtOMfDeETAfwqJ4DDNUIUgiqIgCMVisVgslstliADQihcfYvTg/4YCGfhGq9VCEBi8q1gVimOO2vs90oONke87bS9wqN2I6i RSgQACEqlUiaTgbKORCKRSCQgfSadTq vr2ezWcCHOr0b0m00Go3BYDAYDOCHtlgsDofD6XTW1tZ6vV6IDxuNRl2F1EKKPdKWjGd7mqOPQk8JHA92T7HVqV6m8CeUE0I6ezqdTqVSsVgsEoksLS0tLS2trq6ur69nMplCoVAul0VRZC5/BrIqxQJjzBI7gDcAaFgljtPpdDqdDofD5/M1NTXV1dXV1NRARA1ydqquqX6FB7zj95Geks7xAHCoMcE KIpSKBTi8fja2losFltZWYlEIgsLC8vLy7FYjOWUU1U1Inj0WX0AhC02ByPQvbIAq8rX1F9SSvV6PYNIW1ubz fz Xxer9flcpnNZnXuJ7syk0ePe/yeDT1VcKhNBnQvhwCCFNxwODw9PT09PT07Ozs/Px Px1OpVKFQgEgVURGsZnUsF67MYrZqnXTjhVUEDIYF9xmCIRYNUgkSA3Q6XU1NjdfrbWxs7Orq6u/vb29v93q9ZrOZZRcwjvJ3o4g8PXAoqtJntfUIOkQ6nQ6Hw5OTk2NjY1NTUwsLC6lUCtJhGFcA5VGtNqpVAfahittvZhub52wzdwFxxnI8gbWABqPVat1ud3Nzc39//759 3bs2FFXVwcYxZWslKpHQt9PoDw9cFTpE7BS8/l8JBIZHx /du3a Pj4/Px8IpEolUqQGgi1YjqdDvI2UCWfgyXvsOVOt4pvoXs51t13fuA8MX1CnWUCP0mSJAhCPp8XRRFjbDAYIOX4hRdeGBgYqK2t1ev17PEYPa4xfPr0VMEBUwhmJ5SUXbhw4csvv5yamorH46Iogj1JCAHzknEIlrZTBYItwbHZ0vlO78jAodY6Ie2KQRwUZKi3g9T2Xbt2vfbaa/v37/f7/ZA8jCqZLo9tEJ86PVlwMCFCVbXFhUIhEAhcunTp3Llzo6OjkKEPyU46nQ6K59RzXGXKbvmnOhWoKi0NfXdwMN2Iqlxq6rxDpq9AiVepVNJoNF6vd /evceOHTt48KDL5WL55eh7q4U8EXCohQjod2zBhcPhkZGRs2fPfvvtt5DyD3UZzE lXm14q0wf9ZfsJ7Wl thfZzNhVRolyJpyuSxJksFgaG5uPnLkyEsvvQQ1WizrVi1lvi8QebLggBUGg5jJZMbHxz/55JNz585FIhFJkiBRCmqZ2IlVrHh7ggNIEARo1AESUBAEMKksFkt/f/ rr7766quvtre3QzKYmh3 AI67imexWITi9LNnz46OjhYKBY1GA5UKLD2dqhL71GYn2kqdfLbgYEIHmhKUSiVCiMFg0Gg0oiiC257n dra2ueee 7999/fs2cPy6f//x0c6mRxSmkikbhx48aJEycuXLgQjUYppcAt9Hq9eo6pKomySqw84F6bwfF43 V N6WUAqahy1mpVEIIsYZmxWIRHHRGo3FoaOi999578cUXvV4vyyetUkG2LVYev/scxg70jFgsdurUqQ8//HBiYiKXy0FxB0uuZ1pkleyomvKHoac8vozVsYgdRPtkWTYYDFarVZblfD5fKBT 8pe/rK vJxKJt99 u7GxEfDB3mubmzOPHxysBHRlZeX48eO///3vg8EgpdRgMAAyUKV6Rb2A2JD9DWbnUyZ2U1YwAWVU0HwMIQRCk O4UqmUy UmJibS6XQul3v//ffb29s3e1S3LT2SWKnyMQApilIulxcWFj7 OM//vGP8/PzCCGdTmc0GmFcmBC53zX/hiG7H7a2tCEfiyTdzN4KhQK0MISGegaDAWOczWbT6TSl1O/3Hzt27B/ 4R/6 vqYC7XKNHv0p3q89KicQ236w6yXy XJyckTJ06cOHEiHA5Dzara6EcPHIinOUZqlGx53y1hVIVvdgx0o4MuWeBXhc6qoJKvra2dPHmS47hf/epXLS0tIF 2ISDU9HjECq2QKIqzs7P/ Z//eebMmbW1NRgdNTLUjOF K/gxDtnDM4ktj/xOPAbKFaEDR6lUAvag0WiAZRYKhVgsdvLkSZ1O9/Of/7y1tRWaxmxnemw6h6IogiBMT0// 7// 6lTp LxOIgSQAbop5tnffPoP97FdL lDx/IVs1xthRMm90tainGGCewSVmWS6VSsViEZDOMMSQN5XK5lZWVjz/ 2GAw/OQnP2lqagIHz7blH48EDqa0g3CJRCKnTp06c bM6uoqJFwB81RU/aDVA6HW7DbzWKwK3iKVuQu/stlSSzR6b8pWVXWy grs4iwtSG1 q/UJdU6y gGgtJUlkbDvIW9IURTweQB7IITodDrIWgqHwx999JHJZHrvvffq6uoQQuoclG0FlEcCB630d1YUJZPJXLx48fTp06urq0zP4DgOrJItPRZVFp2iakKt9nRVfUaq9cqOB bEDoaZZuwKnNzMV0tVyR/M88ZyRNiUq OxVJX2wS4OYX12QUAJM26FCoEuAmMCCSvBYPDkyZN1dXWvvvqqyWRij01UHS62Az2qWGGtRaanp8 ePTs9Pa0oitFohN4SbL63tFCqEMMmHgLiwJPRvQFS9UXUmIPUYvhJFEXIF2SzDudCEEer1UJ7KnDEwQej0WixWEAIwjE6nQ4yidit4ToQSYEOaevr68lkcmVlJRQKLS0t5XI5xkUEQWAZQ8A84MmBf0CzzT/84Q9ut3t4eJg16nwsZtRjpMdjrayvr1 fPn27dvgOQZRCksZ/bWEcjWrB8UFOvCZTCboRclOhw Q38VycNgHSAsCcWaz2aCVCnSftVqtFoulpqampqbGZrPZbDaLxQI4AI8cBISBZzDxB09VFROhlZQfEBylUimdTl 7du1//ud/rl27Bq3G4DFQpUGZWt7xPA/dTguFwuXLlyGlubOzk3W/2T5sAz06ODDGoigGAoGRkZFYLAbjwlL30AM9xFXSnR0DXYFY5hjwbZALMMQABYhoAJeyWCwWi8Vut7tcLq/X63Q6AQeAMMg1BxyoY7/qLA3gWMwdDqwCDgaWAy FEFKzE5vN5nQ6i8ViQ0PD6Ogo9OyGmQYPOrsXrURkQMqAC/XChQvd3d0ul8vn8z3iRDwJegycI5fLQR5XsViEQGvVnikPfym1CwFIkiRIFgTpjjE2GAw1NTUej6erq6u7u9vj8bjd7pqaGnA9GY1G6KGD7i0gqEIqmzB2QKlUCoVC4 PjgUAgGo2CkII2h06ns6Ghoampqba21m63w 4LbIkTQiwWS21trcFggBauEBxgUpXeu5cN/Ar3XVtb yzz7q7u6HP3bbSRtFjMWXj8fjo6GgqlWJyvapDzWZt436k5gqCIMBYI4TMZrPT6YT874aGhtbW1vr6 traWuh3zhgVVrX/YrcGUVWVekhUbR5JZZ NeDweCATGx8ej0SjEWqHZvslkKpfLmUwGmBCIITWCrVarz ez2WyxWEwdFiCqhmPMGgIC/lEqlWZnZ69cudLT09PQ0LCtZAp6LAppNBoNBoOgKNBNDtOHvI7adoCLQD8/6ALe1NTU1NTU0NDgcrlMJhPYQcCcSaVXK0MDaKOwu8P6 jqrcykWi9Bltb6 3uFwWCwWZmnD7bxeb3t7e6lUcrvdlFLYFaWlpcXj8cCuLiBi1KYysBCLxdLc3Oz1eufn55nxglTsSm3MQ6tdjuN0Op0oiolE4vLly7t37/Z4PCy/cJvQo5qy5XJ5aWlpbW1NqfTtgwbW35VDssJGUPdqampefPHF9957r7u7GzpVgvHCxh0OYxCBPWwymUwikYjFYlDwEg6Hw Hw8vLy vo6tF8G7Q 6RXd3dzc2NkJ7U8AHx3E2m62hoaG vt7pdDY1NXk8HuiuCZov7NehromCzxqNxu/3t7a2jo6OZjKZcrnMSuXg7Uilby7bywyCc4CPQCBw9erVXbt21dXVPTyXfQr0qE6wUqkUjUYLhQKllOd5sAbBFv1Ol6IqQghBg y vr7GxkY1c1Yzc1Rp/5hKpRYXF8fHx8fGxoLBYCqVAqiB0wnylh0OB8i7WCxWLBYjkcjq6urw8HBXV5fVagUDlRAC/f8zmQwhpFgsjo6Owr5VhULBYrH09PT09vbW1dVBUA0eAFhXbW0tdLTNZDKgezEFlj28WpGCOl7IeMpkMhMTE Fw2O12/11xDsj/A1bJfFlqww/d3z1c5bFACCmKwnqWJ5NJwBy618BjAw0FL6urq9evX7958 bS0pIkST6fDwpJ3G437JEgCAKq8HbYeAXCpMDD4WnBwjQajYqiQN9mqJoJBAJjY2OFQgG60zO7CfgWqL0AaKvVOjQ0NDQ0tLi4mM1mwUhR7wvGlF w5sAMhlXEcdzq6mokEunv72cZk9uBHoNCqlYGlcruRg pbahxA8MECmaxWAyHw4 bDmAAACAASURBVLFYrK2tbXMyOlY5v3U6ncfj2bFjx/DwsNPptNvtFosFskZgu7FoNBqLxZLJJM/zDQ0NdXV1ICMcDgdIDbgsiAz43NvbK4piNpsFC1lRlNra2paWlpaWFtgmZ/Pz8Dzf2Ng4PDx869at8fFxCLyBAcJgATIRNGLwkjEMQThGXXX36PPy6PSo4NBoNE6n02Qy5XI5Zimo7YWHpCpZC/vWLC0t7dy5E3oCk03dTuFGBoOhs7Ozs7MTDovH41NTU0tLSysrK6B8JJNJ2FSqra0NqukdDgdAgakCaovXZDLBhialUsnr9Q4ODoL1pNfroZya2WKsagGopqZm//79d 7ciUQi6 vrYO8w4QJcFlpCsNazvb29O3bskGUZNGUwgraPZHlUnUOj0djtdlhMyr2bj37XSzGFA3xTsNFfOp02mUxM28f3JvXQiq8sm82y3fPi8Ti40aDuobOzs62trbW1tbGxsa6uDnYJVRs4SOX/oJUdbhVFgZxhthkb CewqsiKldqCSNXpdE1NTa urq6ipsnAip1LhSqw2YAGvf4XDU1tYeO3bs9ddfhxvV1NSw9svbhB4DOGADUbaemNqhZrz3O33zASBZoOEC7CkGvdXZYmUYAmYOOzFEo9Hl5WVZlsEtZrFYEEKyLJvN5traWpAFMLvqqlfGMNQ6rxoBoiiur6 D9QGTCs5WtVLM3gVjbDQaoUl8S0vLuXPn5ubmQDk1Go0 n89ut9tsNnDVNDY2NjQ0gJwCeUrvrQXfDvSoCinP801NTX19faFQKJfLiaJYlfSFHk4hZfONK1m75XJ5ampqdHQURhBVJo/VziOEoK4aNtPo6OiAdBsQ4eBuh1/ZogeWDu0b4CzmzlK7IpgUUxSlVCqBthiLxQwGQ0dHR09PD3T1V88lU8PNZvPg4KDP5 vp6Tl16tTly5cLhUJ7e/sHH3ywe/duu90OHn3wpDH3HcPl348pC2D3 /1Hjx4dHx /c cOOKDYxjNUFVTb8vQtP4PtANsfXb9 fXBw0G63M WOSXr2AKDDovv3g2PMP5FILC0tZTIZm80GW64wqIEXlW16yjayBINTq9Wurq4uLS1NTEwkk8nh4WG2RwdAhMWfIeLj9/tfe 21pqYmh8Nx9uzZ1dXVmZmZvXv3 v1 9aMSVQ4p2Wpz62dLj8Faga2HgXlAr3/GtNG9 TJVJ96Pu4CaWS6Xi8Xi9evXe3t7YVMw9SJD944sAwSqjDuYJOzKoiguLi5evnw5GAy2tLQMDg6CWpNOp PxeDQajUajsMcPJAmDFSMIgt1uHxoa6unpaW1thT1HYbdHOIb5ZvCm9AOTyTQ4OAj29oULF06fPq3X6yGBlPEbgMK2EiVqelSdAyEEfu7BwUHYoRjcf1XsUT15D3llrVYLNsvIyAjbN/p wV41k4Bv1BMgSVI6nZ6ZmQmFQrAd6eeff14ulxOJxPLycjweTyQS0FgBdBqTyQQNfTo6Orq7u3t7exsbGwEKzFBX /vZjaruDvvPvfXWW2tra998881nn33W2dlZV1fHEnyqjLvtwzOAHgM4EEJGo3FgYKC1tTUcDoP6xlRIvFUO2MNcGZxF Xx ampqbGwMHESKajP3KgJkqJtqMI4FHLu vt7r9fI8H4/HJyYmoIOUIAgcx9ntdo/HA8kfZrPZ7XZDTAd0W9BAYWstfO/WPgwijKvB90pl6yeTyTQwMDA8PDw Ph4Khc6dO7dr167u7u7Nhsl2QwZ6LIE3UDLa29v3798/NTW1trZWLBbBxFcHWTavjypuXGU7wAbY4J6/fft2JBKxWq0PyNimlRw FMdnoX4qtVqBWulpaWlr68vnU5DphZ02qCUQpC9XC5DDRLstCWKIjQd9Pl87e3tPp8PfN7qBFX1C7I7Ap4IITabDRhGPB6fn58PBAItLS1MEX7E8X i9BiSfeCDy U6fPhwIBD49NNP8/k82HVsBDf3blNzZlqJ5dJKihetJGlyHFcul2dmZkZHR/1 v8vlesCYqsV/1WdoOYoqsTpI1Ein02CGLC8vg8IBmT6SJEGoL5vNwrbTsizDnn6vvPLKwMBATU3NZlaxJVBQRUWFBLBisZjL5cDhsd0M1830GMBBCIHl1dvb /bbby8sLFy/fr1YLMJkQIUgeAhALjCUABTAnBNFERyIsOUi7PSMENLr9ZDk91e1FrWLAr6hqrgMqaQHw5 ZTGZ2dvbSpUtXr16NRqMIIbvdXldX5/f7YY9tqDQJBoOLi4uwB2w6nQbllKpiBezWVWhg6EQIgT0M7wVenKo2JNuWHtWUhQ ksv/evn37Pvjgg3w Pzc3JwgC2LTgckCV VMqu8SBAgg5eZAhDJ5NCGe73W6/39/Z2dnV1dXb29vb22uz2dhwq /LiAVHtnSvMZYjCMLc3NyHH354 vTpRCIB0XnAXzwehwZUoihCb1OolNdoND6fb8 ePXv37rXb7YAGdVxty2HBGINBPjk5CUn5fr /oaFhGyZ9bUmPqpCqzUVCiMvl vGPf6zT6T799NObN2 m02nw9hgMBo7jCoUCqAXAFZj CLoecAiXywWuw7a2tpaWFr/fX1NTYzKZmHuAblXfUcVUqjChni24aTabBd0I8Ap7i0IOOoRnQTOFDHVQMuBXyKq/nzeCYRdXtvGen5//9NNPv/zyy1QqVVtbu3///ra2NrZv4f3gtU3oUftzqKU7fCNJUjabnZub KLLz7//PPZ2VlFUQYHBx0Ox8zMzPLyMlQIQjhmZWVlfX0dIdTQ0PDiiy/u378fIiA1NTVGo5H1gqrSatFDZ2nTe/MUYUpkWQ6Hw9euXRsbG4vH4z6f7 DBg01NTYBg5rhU36JcLufzeZ7nISy3uZmkWpcCEQmtzz755JNPP/10aWnJZrMdO3bsl7/8ZU9PD/OusutsT3oM4Kj6hgWZ0ul0MBj89ttvY7HYgQMHGhoaIpFIPB43m80Oh8NgMAiC8Pnnn//v//4vbKj5xhtv/PznP /r6wMHVJXFWCWkH5Itq8FBVWVtIM7C4fDMzEw n2cpg BoJ5XSJsgRL5VKOp0OcsRramogN4xuCp9C9ii0 wGH2/nz52/cuJFOp30 35tvvvnee 9B9Fi99 ffPvRPnh6btQJEK EPQojb7XY4HP39/YIggFzo7Oxk7BREEuwwferUqVAo9OWXX0KGVVdXF6tYATsZbfIRqV1hVXfHqoQj9XMy5gHzqtPpzGazJEm3b9 em5tbX18HyxPqblgBgV6vr6 vHxgYgC3p1d01mMCilX7cqVRqYWHh1q1bIyMjt2/fjsVier3 wIEDP/7xj19 eX6 nomHL8XOscTafuENgU41OyXVvyMCCHYyvvChQvHjx8fHx83Go3Q0HPfvn1erxeSJ1hKxOYBvR9cqvxvbJUDV2MBd0EQotEopJpCx0tI4GPuf8hWaW5uhvRBk8kEBhet5KwLgpDL5RKJxOLi4uTk5OTkJGSTFAoFo9HY2tp65MiRI0eO9PT02O32zUbKNofI4weH2opjS3mzwkhVsdBMJvPtt99 OGH0Desra3thRdeOHTo0ODgIITHIPMblNmqkMrGa6jiupvzFKkqvsPAwXgMQgjsKRaUYVXXwEjg7qhibbG89ng8HgqFABPBYHBhYSGbzRoMBqipef75559//vmOjg6TycRuVxVm2uZi5Ym3mkSbPOhq84FNXrlcnpub /LLL8 cOTMzM0Mp9Xq9AwMDO3bs6OjoaGxsdDqdZrOZlZXez1jAWzUmrGIq7AHY3KjRgO7NGgFmA2WPqVQKGnzNz88vLCwsLS1BaAYKYh0OR2Nj486dO4eGhjo6OhoaGiAlQI2Mbc4qquhJdTBWr n7/cr pRXPdyqVmpiYOH/ /OXLlxcWFiDn2 v11tfXd3V19fX1NTU1gWUL4n/z7ifk3r7jiqpyn5Hay07vLcMHWwN6vUGpdCqVgmrpcDi8urqaSCSy2Szs8ALZYn6/v729vbW1tb29HWxvKHcglYpLSCoAJfd NvD2pGe5xxubIeZlB56/vr4eCARu3Lhx8eLF2dlZyG7HGIMXBCpE/H6/3W6vqakBPwSkdABW2Ae6KbgPsIDKEUgihIIX4Acw6 l0GjZ5gc/QpgfanAMXgQxhjUbjdrvb2tp6e3v9fj ltFgsQtowHMxxHGzwU1dXNzAw4Ha7N3fu3ub0LMGhVvWRitkghCAkBsl/oVBoenp6YmIChDpMEttQAdwSsCjhXzA3ABywcNntYHbBGwtKMThDIfwGn6GpBvhqmYQilR3gEEJwa47jWHUupChADMViscAmLFB/cOfOHY/H89vf/nbv3r3qPRWexXh/Z3qWThimkTD9gMFFr9dDJ4XW1tbh4eFcLheLxUKhUCQSuXnz5qVLl9bX1x0Oh8PhKBQKEDZj9SlUlZ1FK W7jJEw7xZYK6AuEEJYFQk8FVhJcC5jTtDPA1L9AAfgMzWbzWazGf6E jyEUKFQuHjx4o0bN7RarcViYYLm 4IMtE02AFSjRP09rWzsCBsl f3 8fHxW7duWSyWgYEB6DxfLBZnZ2dHR0enpqYKhQLHcZIkaTSa2tpa0CJ9Pl9/f7/NZgNWAb1ZMMaQwBwMBmGC19bWeJ5vaWlxuVyAA jp4HA4oBaGlfBDsS5AhxWegDlDKoVMkiRNTk7evn27WCz29fVBHgna9rZrFW0LcDyA1Fbf6urqn/70p6tXr3Z3dz///PMulwsqPhoaGjo6Os6ePXv9 vVMJgNthP1 fz6fFwShrq5u3759bW1tVNVXCWMci8Ugd7BcLkNOcm9v769//es9e/ZArjlwL9aTmvk2mJ8DRA9W fQYvrPZLJRner3e3bt322y2ZzmIfyttd3Cw4c5kMl999dWlS5ccDsfOnTs9Hg9WZfQ3NDT86Ec/0ul0t27dWltbg1o3pHJ7sPxCiPnhyj5LDocjEolAgQnkBTY0NMDqZ8YFraSboEpZHlXV1zCDGfJ3wN4JhULXr1/PZrNvvPFGT0/PtipyfHja1uBg4y7LcjAYvHTpUqFQOHDgQH19Pax1pHKHNzc3Q0D14sWLqVQqHo9DNRR4t9hhpFJzoNFoPB6Pw EIh8OiKDocjvb2dqfTWeWQYOgkqpIqIBZTZWYRHJDP52/cuHHr1q2GhoZDhw45nU7mQ2PHo DiNnW4GBe1PX19YsXL05NTTU3N3d1dbEERMbJwXT0 XyDg4Ng1IA7nFIKtS2QWQJ2LLQdy fzKysrCCHoLEsphVBtU1OTz dzuVzqMnms8sCyZ0OV7GW4O6udn5mZuXLliiRJR48e7e3thetsLjugf/dpgk aGJe evVquVxubGy02WwsFMf4B6tmA4OC53kwUBFCRqMRAiKlUimZTAaDwWAwCC14QKWAclyNRhOPx3// 9 Lojg0NPTuu /u2LFDr9fDY D7pEkzcZNIJMbHxycnJ6G54Pj4eGtr6759 0Db2NIX94BXZp fLXq2OzgQQoVCATaYhdZb6m50qKJVQCeIQCBw586d2dlZSZJgXiHfDIoPZmZmIMBmNpu7uro6OzvBoQlObrjRjRs3Pvroo3PnzkFpgro tsqri1SRmlQqdfbs2U8 WR5edlgMBQKhUwmk8vlwLOuzp t8sxWvWlVVAHd3457OrTdwaEoCuR7xuPxzs5OyOyFsnfQJzKZTDgcXlxcXFhYWFxchEYrkMoFIdOlpaXPP/ 8UCiApfrKK6/s3LkTTFamNED0ZH5 PhaLpdNpcHmBmKjywaB7VzYE4a5evXr8 PHFxcUDBw4cPHiwUCh88cUXU1NTf/7zn7u7u/1 P/N9YZWnX63WsBgCUqWgQgb/5tzHp0bbHRwIIUEQkslkqVSCPrIIoVKpBNH2UCi0uLgI5kZ9ff1zzz0XDodHR0dhcKHCMRwOZ7PZrq6uvXv3dnR01NXVabXaUCg0NTWVTqfz TxCSJKkaDQ6NjZ28 bN9fV1v98/Ojoai8VAuaniFlV6Q6FQmJycvHXrlslkSiQSExMTkO Ty VGRkYEQQCFVB2AVV TsSWQj8xHhzH2 /2wxRNrcwunP7VY7rYGB8wxtGex2WwrKytXrlyBbZHS6XQikYCKgSNHjjQ3N/f39 t0ujNnzgQCgWQyCXiClkOgh4LjAaZWlmVRFIrFUsWvipi3tLaullJlfPz2xMQ4pagKHBhjhDBGCCYKYyTLiiAIZrNFo FnZmaDwTlCSLlc1uv1gA/WCERt9YBJrK7GJgQTwhFCeJ7DmMTjMZvN5vf71dh6ysxjW4MDRsTpdL7wwgsQ0IdNnzQaTVtb2/Dw8MDAABQaQSZwLBYDTzmgQal0 zMYDA0NDU6nU1EUQjDHcxhhhBAmGCOECYHWtaz/OcdxGl7DcYRSJMuKWp3cEC8UUXSP2qgoCpjAGyCgiFKKMFVPKlVltzB84LsVFUij0RCOSJJcLpWvXbsaDAahghBqXqpCRU BtjU4aKXPQltbW39//8WLFyHB3 v1Hjx48Be/ AUrWockrrW1tUgkQgipra0FVdRisUCHrtra2t7eXowxQpSQu/mkhOMopRgjjnBsquDuGGNFoQwc96xaCrN/d6NT4E9qbwdVVWqpfWVqYxhXYj0YY4TultAV ILBYBAEMZlMVvUreJq0rcGBKkMZj8eDwSB0lgJuDIEPFlGTZXlychJ2Ddu1a9f6 vrCwkJtbW1zc/PCwnwkEgmFFjweD89z5XKZIkUUhEKhIMkSx/EYphPRyrUJQjDfFFGkti8qHzawAn8TNmcYKYqCUUV7xRj4E8KVzDd0t5cm8AtJlqiCEIKJp4qiUIR4jisUCgsLC VyiTX1/kGsbE2SJEUikcnJyVKpBKJBFMV0Og2CA2MMbV4 /vjjUqn0yiuvxOPxixcvms3mvr5evV4XCs2LohAIzApCWZKkTDYtioKiKPAnQhv6A7NHMCYcR0A0YMwpLCNJoQhRihClEkIIIUwVRVYU9XxVFBQM800wRymGKa1WXDDGGCmKguiGEgqCiGBMKS2WSoV83uv11tXVPcPyhe0ODkppuVyG7Bu9Xu/1ekVRTKVSsVgMKnIRQuDdamlpaW9vX11dPXv2bDQaHRgYsFqts7PTmcx6W1trY2OjxWqVJanGYaNUqYRF4BbsH4QwospdHwMhPCgXVFEUqiCKKFIolWFeRVGSJHHjOTfgpSCECCYVQBCMCMIbHaqAIVUOr9ySUsLBPhOYKgrhuPVUKrS4qOH5AweeGx4eBjfaD2IFoU2dPCBUEQ6H19fXoSZRkqRMJrO6uhqLxVpaWgghZrN5z549/f39U1NTZ8 enZuba2ho8Hg8c3PB2dkZq83a399X5/UihCgFua4gjHBlqpgRQQjBBCOKZGVjjjHmKuoFopRihBBGmCiKIsOf96ZNq/4HTIJyGBGKKELq4jaqABOiCOONPziOx4jIspxIJqNra3ab7bVXX33vJz9paWn523YZeCy07cBRRZIkQUJvNpuF6llKaTAYXF5eXlpaGhoagmQws9m8srL82Weffnv9W7PZ3NTUlMlkpqYmJVms9/uczhqCqUIVghDhEMwyAiAiShFGMEmYxxQzoYAJYmwEY0Spgja4DQW9kxCsggJCGBGEN2acUkQRIRzBXMW9JQGUKFUQRZg19CUbqglFaHllZXp6GiH042PH3nzzrabGRlZA9UwGfzuCA6sSgyVJisViS0tL4ISApFGz2by6ujoxMXH48GGPx6MoSjKZ OqrC19/fQlj1NHRRqkcDM7m8/m29lavz8txBFGZIIqoQmTMIUwRoggpWKGIKmhjSjFCiBKYdISRJMtUpkwLlGWFUgVjgtDG7mCEq zLoVCkILrh5yYEY46QDewoCkaKQmWCFIARABMjSilBGGPCKzLN54uRyPLMzIzdbjt27M033njD5/NpnvX2xNsRHIzAXZFIJBKJBEIIYOFyuZqamtbW1m7fvj09PW2xWERR/Oabb06ePJnJZro6u x2 8zMTDQa9fv97R3tZrNZUWTME0QpRQqiGGGyISA21AyE4F9ZQQQRniiYlErlTCaTSWcLhYIsy2ybQKZdqrxYGCHEczxHePgTqq7NZrOW4xUqUyqTDbxt3An4AMdxBBFJVvL5wtzcwuzsbEtLywcffAB5TCxl9RkGb7c1OBBCoihCvIPneWAb0GU8nU7Pz89DnW2xWPz87Gdz88HmpmaPx7W4GJqbD9pttr7 HrfbRTAmBFOqIKRgTBFGFAGzAGRUTFKqKJRiRJBM05n09PRMMDhXLBb1Op1Wq6VgiSBEFYrxPT0LEUKIUow2SmYkUZQUubG obevz 10IKpU1AuKMORRYwX8Y5yGUhqLxWdnA/F4cufOnT/96U Hh4etVmuV7foDOLYgGO5isVgqlRBCuVxufn4eagVsNlsgEPjTn/70zTffKIqSy2fq632NTfWJZDw4F9Bo Na2FqfTwXOEUgUAgVGlhzXzfW/MKcYYUYplpIglIZ5Mzs4Gp2dmi6ViV2fnzp27YPsVWZEhWR8jAvoppVQB5UKhiixIolAoFOfn5ycmJhKxeClfQDU1oFHA3FKFYowIxphwCGFRFKOxxMTklCzTV1555Z133unp6YFtoNQZAs8war/dwaFUmnkQQqAnH6TcQU2RJEnhcBgh1NRc39XViTGemwsKQrmrq7O5ucloNIBKSTCquBoowohihCmhCsWUEMwhiqksI0xlQVpeWRkfnwyFw7ls3ma3tba0Dg4MOBwOQRQ2HFwVJRVcm3hDq0VYUTBCgigYDIZQaEGWJUWRCSFoo8qSoooTjHAc5rhcLh9aikxNz5rNtrffefvNN9/yer2ssAU9a54BtK3BAasTahWhptnpdEKqZk1NjdPphO73Lpdr186dJpN fHw8k0k3Nzd1dLTbbBZMsKLICLHIyIbSSDFCSAEdFFOMEUcRSqfTwbm56ZlALJFQKOU4TqPRarVaRJEoCJRSOI0qCqUywghTosiV3vgIE0ooohwiOo0GUySLElIoTwjCGCFF2WADFCGqyPL6enp6ZmZxKdLe3vl/f/aLvcP7nA7Xs8XBlrRNwVEVFqeV7mHQWAe yeVymUxGq9U2Nze7PZ5QaH5lZc3hdHZ2dbhcLl7DI0oJ5hSKKHNsEYoxRQhTRBBCiqwQTBVFTqVSd6YmZgKBYrHkrfPqDIalcIQjHEGEYIwoQuwCoLZghMDfVfmSYIwQkRSJIxzBnCAIpWK5LAg8x8HJHMEUEYxQPJmanJqOxuIDAzv/8Ze/3Lt3v9FoAvBst7rq7QgOpp rwwoQ2aKVjJhSqZRIJAqFAqR8ppLpYGDBoDd3tHe7nB5COEWhBBGCYHNQGWNQQqlMFSyDE4MQjiuVhLWV6PTUTGhxnnBcV3dPW3t7PJlcXlnZCLRiDiGMqCp xupy0YaDAkMghSKiYII5QjiFIlGWJVnGPMdxPFUUGWNJkuLx2HQgIIjKj4/9nzffequ7q8dgMAG0cIWe6djfQ9sUHOwDrnSr5VR7wKJKGBayPSRJWgovra v9/X1Njc16TZ62WIFKSALEMaUIoUqCFGCOQSBekxEQQqHw2Ojt6NrMYfD3tHZ2dzSYjSZ0pkMpYgQwm3kXhCEN1RQhO8J1qOKuAB9hiKk1Wo1Wi0qFSmihOM5jqcIEY4IghAORwKBgMVm 8kH7xw9 lJdnVer0YInfrvBAmjbgYNFt2mlah7Sx1lnC7zhadpom7SxA2F0FXZ/NZlMFMkbyp MKFXAXKCUEKzBhIKFwnN8PleYnZ6dmZ7NZfL19f7OznZfvd9gMHIaXm/UEQ4jpGCCKzZJxXC911eJCYgphBSkIIow0uq0Or0OZZGsKETDcxqNJEnZXG5hYWFtba2jo vd997dMzxstdp5jseYwBs//XF GNp24ABiGQwcx0FhaiKRUEfPYfcMjPH6 rogCFqdtr293el0kY2tfilC4P/esCckScEEaQjPc5ws0Xg8FZgNzNyZVhTa0dHR0d5e47BptBpMEMdhjBRCKCF3NybbMElA/1BPJQROKIJbwhNjQkRJyubz8UTSaDQUCsVIJJJKpnbt3vPOO/9nx CAwWAkhMMIE8IxcDxDZ9f9aNuBQ 1copTq9XqoSwuFQmwjBEopNPajlEI cGtrS0NDA8/zkigKYlkUBUWREYRJKFVkWZQkEFdiWcyms3Pzc8vhsF5r2NHf39HeYbVawCmOEJUkQZQkQjDHV0quqYIqYRU1MpjyseEWJ1iW5bJQFkUxXywsLCxkczmtVsfxnM1m /Fbb73x utNjU0anQbjStjt3jTjpzfKD0fbDhxALO8BXNHqglU4AFcy/WHj6kgkUiwWMMGyLApCWRDKkiQi8GgoiqIgCjYLReWSIJSEXC6HqNLQ1dTS0mKzWTDaiLESgiRFliRBkiUNr E2Evg21B10/wAY4TBFGINTDGOzxbJv/77DR15wOJ1GgwHaioAdjjBFSAFHyfbDwz20fcHB0KDO0sOqZDudTud0OsvlMiEknU4XiwVwN2GCNvySGLKBOUjZ5TieEGIySIVcTiiXFEXR63VanQ5hQpGCMMIE8jAwVagiyYpGBl0DUYoJ2UDmRt7HXQ6CN0QXUjBBBBGeIxyx2Wv27tv/5ptvQi3FhpKEQUWhCBFFAfm0rdGxTcHB1E8Q SD1wY/O6k4tFktnZyeqTBomd1M0IDMDV0YfY9VO91Qp5rOiKCZTqbIkirJMCUaIbPjKMEIEs5CqJEkQc0NMHwWzZCN8BjAFryumWEF4I8LLcbxep9cbDBqtFt8bVsWIIIQJR7c5MtA2BAe d 8SjuNcLldXV9f8/Dx0aLFarUzQgJVLKbi27s4BBmBQ8GdsXBacrRsxUYIpVURJguQfqigbwFIwVTCi7D/KQ2FBRaYQQiCDi6WEKJRyGJKE0AZQRNv3QQAAAltJREFUcWXDF4DjfV70CQ7iY6JtBw6kSuRHCMFOLr/5zW/a2to uijYDAIO5WKogh7U8iyLMsSxxFWcAC5EhgBMu5axRCpoYokiUI2l5Oh9SBG4NveiI3Rjf8UiiAddKP3hiyrVEe04TlBaMNcIVihgBfgOkSW5YoI2n5GyEPTdgQHqvAP GwwGPr6 sxm8 zsbLFYfPvttw8dOiRJUjKZhPaPqVQym81kstlcLifLcj6Xg5xTntfY7TWsaq3iPpEJRyD7AvQDBVG8oVXArQnHaTjMUVXquVrF2Ajxb5guuFLKokD8n c1Gl4jy/SutlkxTL53tE3BgVSmHbdRWrLRVKOxsXHPnj1msxlav0FFa6lUgPYHYNyeOXPm PHjDQ2NP/3p/21pacGVRmGUUlmRyuXihfPnT548KYiiJMuE48CbhTFRFAVRmSM84TRUoRhKCDjursigaMMEraSQYoIRogBnrFS2tpRkpNIzvqfsY/uCQ02KohSLxWw2C WNqFJUyAqUFUVWqKzICrjFRkZGYH9oaPh0b48eqVgqrK6unjt3ThCFjU3/FBlTTAiliAIaONAzMKaKQplOitBdpRepvKWVD4QQjVajN hz QL6XqgVDyRekqVn/Qz3pQ2GQZWyUM7msphgm92m0fIKVWRZ3ghzKAj0BJD iqIIolgWBKvNZnfUEI5QpFBEMMZwGCYYEwVhiRIZUyQpgkJFTBBVkILljZCKBhENorKMiCwjUZAo MEQYlFacJZTvKFXQEiQEEJ4LTEYdbwWK1SQZFGSZXiXeznH9wM2/w8z07TIub6ABQAAAABJRU5ErkJggg==' the_img = b64decode(img_)#将图片硬编码到GUI paned.image = ImageTk.PhotoImage(data=the_img) self._img = Label(self.root, image=paned.image,background='black') def set_widget(self): default_name_="会是谁?" self.label_show_name_var.set(default_name_) self.label_show_name_adjust(default_name_) self.btn_start.config(command=lambda :self.thread_it(self.start_point_name)) self.btn_load_names.config(command=self.load_names) init_names=self.load_names_txt("./names.txt") self.root.protocol('WM_DELETE_WINDOW',self.quit_window) self.root.bind('<Escape>',self.quit_window) if init_names: self.default_names=init_names #1.文件存在但是无内容。2.文件不存在 self.label_show_name_num.config(text=f"一共加载了{len(self.default_names)}个姓名") else: self.btn_start.config(state=DISABLED) self.label_show_name_num.config(text=f"请先手动导入人名单!") def place_widget(self): self.lf1.place(x=300,y=160,width=250,height=50) self.radioBtn_sequence.place(x=20,y=0) self.radioBtn_random.place(x=150,y=0) self.btn_start.place(x=300,y=220,width=100,height=30) self.btn_load_names.place(x=450,y=220,width=100,height=30) self._img.place(x=90, y=165, height=120, width=180) self.label_show_name_num.place(x=300,y=260) def label_show_name_adjust(self,the_name): if len (the_name)==1: self.label_show_name.place(x=280, y=10) elif len(the_name) == 2: self.label_show_name.place(x=180, y=10) elif len(the_name) == 3: self.label_show_name.place(x=120, y=10) elif len(the_name) == 4: self.label_show_name.place(x=80, y=10) else: self.label_show_name.place(x=0, y=10) def start_point_name(self): """ 启动之前进行判断,获取点名模式 :return: """ if len(self.default_names)==1: messagebox.showinfo("提示",'人名单就一个人,不用选了!') self.label_show_name_var.set(self.default_names[0]) self.label_show_name_adjust(self.default_names[0]) return if self.btn_start["text"]=="开始": self.btn_load_names.config(state=DISABLED) self.running_flag=True if isinstance(self.default_names,list): self.btn_start.config(text="就你了") if self.radioBtn_var.get()==1: mode="sequence" elif self.radioBtn_var.get()==2: mode="random" else: pass self.thread_it(self.point_name_begin(mode)) else: messagebox.showwarning("警告","请先导入人名单!") else: self.running_flag=False self.btn_load_names.config(state=NORMAL) self.btn_start.config(text="开始") def point_name_begin(self,mode): """ 开始点名,点名主函数 :param mode: :return: """ if mode == "sequence": if self.running_flag: self.always_ergodic() elif mode=="random": while True: if self.running_flag: random_choice_name=random.choice(self.default_names) self.label_show_name_var.set(random_choice_name) self.label_show_name_adjust(random_choice_name) time.sleep(self.time_span) else: break def always_ergodic(self): """ 一直遍历此列表,使用死循环会造成线程阻塞 :return: """ for i in self.default_names: if self.running_flag: self.label_show_name_var.set(i) self.label_show_name_adjust(i) time.sleep(self.time_span) if i==self.default_names[-1]: self.always_ergodic() else: break def load_names(self): """ 手动加载txt格式人名单 :return: """ filename = askopenfilename( filetypes = [('文本文件', '.TXT'), ], title = "选择一个文本文件", initialdir="./" ) if filename: names=self.load_names_txt(filename) if names: self.default_names=names no_Chinese_name_num=len([n for n in names if not self.load_name_check(n)]) if no_Chinese_name_num==0: pass else: messagebox.showwarning("请注意",f'导入名单有{no_Chinese_name_num}个不是中文名字') self.label_show_name_num.config(text=f"一共加载了{len(self.default_names)}个姓名") default_name_ = "会是谁?" self.label_show_name_var.set(default_name_) self.label_show_name_adjust(default_name_) self.btn_start.config(state=NORMAL) else: messagebox.showwarning("警告","导入失败,请检查!") def load_names_txt(self,txt_file): """ 读取txt格式的人名单 :param txt_file: :return: """ try: with open(txt_file,'r',encoding="utf-8")as f: names=[name.strip() for name in f.readlines()] if len(names)==0: return False else: return names except: return False def load_name_check(self,name): """ 对txt文本中的人名进行校验 中文汉字->True 非中文汉字->False :param name: :return: """ regex = r'[u4e00-u9fa5] ' if re.match(regex,name): return True else: return False def thread_it(self,func,*args): t=threading.Thread(target=func,args=args) t.setDaemon(True) t.start() def quit_window(self,*args): """ 程序退出触发此函数 :param args: :return: """ ret=messagebox.askyesno('退出','确定要退出?') if ret: self.root.destroy() if __name__ == '__main__': a=APP()五.总结

本次使用Tkinter开发了一款上课点名程序,此程序可以用于点名、抽奖…代码不到200行,程序简单又实用,主要有以下六个亮点:

1.两种模式:

  • 顺序点名
  • 随机点名

2.自动识别人名单

3.支持手动导入人名单

4.人名单导入校验

5.人名显示位置自动矫正

6.最多显示五个大字

,