release v1.2

优化高分屏适配,对程序的部分界面进行调整

Co-Authored-By: 肥宅水水呀 <91878264+happycola233@users.noreply.github.com>
This commit is contained in:
wuziqian211
2023-08-19 23:13:55 +08:00
parent ce20b1d11e
commit a080f1ff3e
2 changed files with 154 additions and 0 deletions
Binary file not shown.
+154
View File
@@ -0,0 +1,154 @@
# 版本:1.2
# 导入相关库
import tkinter as tk
from tkinter import ttk, messagebox, filedialog
import threading
import requests
import os
import json
import pyperclip
import win32print,win32gui,win32con,win32api,ctypes
# 定义解析URL的函数
def analyze(url):
try:
# 提取URL中的contentId
for q in url[url.find('?') + 1:].split('&'):
if q.split('=')[0] == 'contentId':
contentId = q.split('=')[1]
break
response = requests.get(f"https://s-file-1.ykt.cbern.com.cn/zxx/ndrv2/resources/tch_material/details/{contentId}.json")
data = json.loads(response.text)
for item in list(data["ti_items"]):
if item["lc_ti_format"] == "pdf": # 找到存有PDF链接列表的项
pdf_url = item["ti_storages"][0].replace("-private","") # 获取并构建PDF的URL
break
return pdf_url, contentId
except:
return None, None # 如果解析失败,返回None
# 获取默认文件名
def get_default_filename(contentId):
response = requests.get(f"https://s-file-1.ykt.cbern.com.cn/zxx/ndrv2/resources/tch_material/details/{contentId}.json")
try:
data = json.loads(response.text)
return data["title"] # 返回教材标题
except:
return None
# 下载文件的函数
def download_file(url, save_path):
response = requests.get(url, stream=True)
with open(save_path, 'wb') as file:
for chunk in response.iter_content(chunk_size=8192): # 分块下载
file.write(chunk)
messagebox.showinfo("完成", f"文件已下载到:{save_path}") # 显示完成对话框
# 解析并复制链接的函数
def analyze_and_copy():
urls = [line.strip() for line in url_text.get("1.0", tk.END).splitlines() if line.strip()] # 获取所有非空行
pdf_links = []
failed_links = []
for url in urls:
pdf_url, _ = analyze(url)
if not pdf_url:
failed_links.append(url) # 添加到失败链接
continue
pdf_links.append(pdf_url)
if failed_links:
failed_msg = "以下链接无法解析:\n" + '\n'.join(failed_links)
messagebox.showwarning("警告", failed_msg) # 显示警告对话框
if pdf_links:
pyperclip.copy("\n".join(pdf_links)) # 将链接复制到剪贴板
messagebox.showinfo("提示", "PDF链接已复制到剪贴板")
# 下载PDF文件的函数
def download():
urls = [line.strip() for line in url_text.get("1.0", tk.END).splitlines() if line.strip()] # 获取所有非空行
failed_links = []
if len(urls) > 1:
messagebox.showinfo("提示", "您选择了多个链接,将在选定的文件夹中使用教材名称作为文件名进行下载。")
dir_path = filedialog.askdirectory().replace("/","\\") # 选择文件夹
if not dir_path:
return
else:
dir_path = None
for url in urls:
pdf_url, contentId = analyze(url)
if not pdf_url:
failed_links.append(url) # 添加到失败链接
continue
if dir_path:
default_filename = get_default_filename(contentId) or "download"
save_path = os.path.join(dir_path, f"{default_filename}.pdf") # 构造完整路径
else:
default_filename = get_default_filename(contentId) or "download"
save_path = filedialog.asksaveasfilename(defaultextension=".pdf", filetypes=[("PDF files", "*.pdf"), ("All files", "*.*")], initialfile=default_filename).replace("/","\\") # 选择保存路径
if not save_path:
return
threading.Thread(target=download_file, args=(pdf_url, save_path)).start() # 开始下载
if failed_links:
failed_msg = "以下链接无法解析:\n" + '\n'.join(failed_links)
messagebox.showwarning("警告", failed_msg) # 显示警告对话框
scale = round(win32print.GetDeviceCaps(win32gui.GetDC(0), win32con.DESKTOPHORZRES) / win32api.GetSystemMetrics(0), 2) # 获取屏幕缩放比例
# GUI
root = tk.Tk()
#----------高DPI适配start---------
#获得当前的缩放因子
ScaleFactor=round(win32print.GetDeviceCaps(win32gui.GetDC(0), win32con.DESKTOPHORZRES) / win32api.GetSystemMetrics (0), 2)
#调用api设置成由应用程序缩放
try: # 系统版本 >= win 8.1
ctypes.windll.shcore.SetProcessDpiAwareness(2)
except: # 系统版本 <= win 8.0
ctypes.windll.user32.SetProcessDPIAware()
#设置缩放因子
root.tk.call('tk', 'scaling', ScaleFactor/0.75)
#----------高DPI适配end---------
root.title("国家中小学智慧教育平台 电子课本解析") # 设置窗口标题
# root.geometry("900x600") # 设置窗口大小
# 创建一个容器框架
container_frame = ttk.Frame(root)
container_frame.pack(anchor='center',expand='yes', padx=int(20*scale), pady=int(20*scale)) # 容器的中心位置放置,允许组件在容器中扩展,水平外边距40,垂直为边距40
title_label = ttk.Label(container_frame, text="国家中小学智慧教育平台 电子课本解析", font=("微软雅黑", 16, "bold")) # 添加标题标签
title_label.pack(pady=int(5*scale)) # 设置垂直外边距(跟随缩放)
description = '''请在下面的文本框中粘贴一个或多个课本原网址(支持批量每个URL一行)。
例如:
https://basic.smartedu.cn/tchMaterial/detail?contentType=
assets_document&contentId=b8e9a3fe-dae7-49c0-86cb-d146f88
3fd8e&catalogType=tchMaterial&subCatalog=tchMaterial
点击下载按钮后,程序会解析并下载所有PDF文件。'''
description_label = ttk.Label(container_frame, text=description, justify="left") # 添加描述标签
description_label.pack(pady=int(5*scale)) # 设置垂直外边距(跟随缩放)
url_text = tk.Text(container_frame, width=70, height=12) # 添加URL输入框,长度和宽度不使用缩放!!!
url_text.pack(padx=int(15*scale), pady=int(15*scale)) # 设置水平外边距、垂直外边距(跟随缩放)
download_btn = ttk.Button(container_frame, text="下载", command=download) # 添加下载按钮
download_btn.pack(side="left", padx=int(40*scale), pady=int(5*scale), ipady=int(5*scale)) # 设置水平外边距、垂直外边距(跟随缩放),设置按钮高度(跟随缩放)
copy_btn = ttk.Button(container_frame, text="解析并复制", command=analyze_and_copy) # 添加“解析并复制”按钮
copy_btn.pack(side="right", padx=int(40*scale), pady=int(5*scale), ipady=int(5*scale)) # 设置水平外边距、垂直外边距(跟随缩放),设置按钮高度(跟随缩放)
root.mainloop() # 开始主循环