You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
beaker-blog/app.py

109 lines
3.8 KiB

5 years ago
from flask import Flask, flash, make_response, render_template, request, redirect
import datetime
from datetime import datetime
import os
from os import path
import pathlib
app = Flask(__name__)
website = 'localhost:5000'
@app.errorhandler(404)
def page_not_found(e):
return render_template('error.html', title='Error 404', errorcode='404'), 404
@app.route('/')
@app.route('/index.html')
def index():
content = gen_index_string()
print('content is: ', content)
return render_template('index.html', title='Blog', content_string=content)
@app.route('/archive')
@app.route('/archive.html')
def blog_archive():
content = gen_arch_string()
return render_template('archive.html', title='Blog Archive', content_string=content)
@app.route('/feed.xml')
@app.route('/rss.xml')
def feed():
content = get_rss_string()
rss_xml = render_template('rss.xml', content_string=content)
response = make_response(rss_xml)
response.headers['Content-Type'] = 'application/rss+xml'
return response
def gen_arch_string():
path_ex = 'templates/entry'
if path.exists(path_ex):
name_list = os.listdir(path_ex)
full_list = [os.path.join(path_ex,i) for i in name_list]
contents = sorted(full_list, key=os.path.getctime)
content_string = ''
for file in contents:
filename = pathlib.PurePath(file)
title = open(filename).readline().rstrip('\n')
filename = filename.name
if filename[0] != '.':
filename = filename.split('.',1)[0]
content_string += '<a href="' + '/index.html#' + filename + '">' + title + '</a><br>\n'
return content_string
def gen_index_string():
path_ex = 'templates/entry'
content_string = ''
if path.exists(path_ex):
name_list = os.listdir(path_ex)
full_list = [os.path.join(path_ex,i) for i in name_list]
contents = sorted(full_list, key=os.path.getctime)
for file in contents:
filename = pathlib.PurePath(file)
title = open(filename).readline().rstrip('\n')
text = open(filename).readlines()[1:]
filename = filename.name
if filename[0] != '.':
filename = filename.split('.',1)[0]
content_string += '<div class=\'entry\'>\n'
content_string += '<h2 id=\'' + filename + '\'>' + title + '</h2>\n'
for line in text:
content_string += line
content_string += '<br><small>' + datetime.fromtimestamp(os.path.getctime(file)).strftime('%Y-%m-%d') + '</small>'
content_string += '</div>'
return content_string
def get_rss_string():
path_ex = 'templates/entry'
if path.exists(path_ex):
name_list = os.listdir(path_ex)
full_list = [os.path.join(path_ex,i) for i in name_list]
contents = sorted(full_list, key=os.path.getctime)
content_string = ''
for file in contents:
filename = pathlib.PurePath(file)
title = open(filename).readline().rstrip('\n')
text = open(filename).readlines()[1:]
filename = filename.name
if filename[0] != '.':
filename = filename.split('.',1)[0]
content_string += '<item>\n'
content_string += '<title>' + title + '</title>\n'
5 years ago
content_string += '<guid>' + website + '/index.html#' + filename + '</guid>\n'
5 years ago
content_string += '<pubDate>' + datetime.fromtimestamp(os.path.getctime(file)).strftime('%Y-%m-%d') + '</pubDate>\n'
content_string += '<description>'
for line in text:
content_string += line
content_string += '</description>\n'
content_string += '</item>\n'
return content_string
if __name__ == '__main__':
app.run(host='0.0.0.0')