forked from github/pelican
362 lines
12 KiB
Python
Executable file
362 lines
12 KiB
Python
Executable file
#!/usr/bin/env python
|
|
|
|
# -*- coding: utf-8 -*-
|
|
from __future__ import unicode_literals, print_function
|
|
import argparse
|
|
try:
|
|
# py3k import
|
|
from html.parser import HTMLParser
|
|
except ImportError:
|
|
# py2 import
|
|
from HTMLParser import HTMLParser # NOQA
|
|
import os
|
|
import subprocess
|
|
import sys
|
|
import time
|
|
|
|
from codecs import open
|
|
|
|
from pelican.utils import slugify
|
|
|
|
|
|
def wp2fields(xml):
|
|
"""Opens a wordpress XML file, and yield pelican fields"""
|
|
try:
|
|
from bs4 import BeautifulSoup
|
|
except ImportError:
|
|
error = ('Missing dependency '
|
|
'"BeautifulSoup4" and "lxml" required to import Wordpress XML files.')
|
|
sys.exit(error)
|
|
|
|
xmlfile = open(xml, encoding='utf-8').read()
|
|
soup = BeautifulSoup(xmlfile, "xml")
|
|
items = soup.rss.channel.findAll('item')
|
|
|
|
for item in items:
|
|
|
|
if item.find('status').string == "publish":
|
|
|
|
try:
|
|
# Use HTMLParser due to issues with BeautifulSoup 3
|
|
title = HTMLParser().unescape(item.title.contents[0])
|
|
except IndexError:
|
|
continue
|
|
|
|
content = item.find('encoded').string
|
|
filename = item.find('post_name').string
|
|
|
|
raw_date = item.find('post_date').string
|
|
date_object = time.strptime(raw_date, "%Y-%m-%d %H:%M:%S")
|
|
date = time.strftime("%Y-%m-%d %H:%M", date_object)
|
|
author = item.find('creator').string
|
|
|
|
categories = [cat.string for cat in item.findAll('category', {'domain' : 'category'})]
|
|
# caturl = [cat['nicename'] for cat in item.find(domain='category')]
|
|
|
|
tags = [tag.string for tag in item.findAll('category', {'domain' : 'post_tag'})]
|
|
|
|
yield (title, content, filename, date, author, categories, tags, "html")
|
|
|
|
def dc2fields(file):
|
|
"""Opens a Dotclear export file, and yield pelican fields"""
|
|
try:
|
|
from bs4 import BeautifulSoup
|
|
except ImportError:
|
|
error = ('Missing dependency '
|
|
'"BeautifulSoup4" and "lxml" required to import Dotclear files.')
|
|
sys.exit(error)
|
|
|
|
|
|
in_cat = False
|
|
in_post = False
|
|
category_list = {}
|
|
posts = []
|
|
|
|
with open(file, 'r', encoding='utf-8') as f:
|
|
|
|
for line in f:
|
|
# remove final \n
|
|
line = line[:-1]
|
|
|
|
if line.startswith('[category'):
|
|
in_cat = True
|
|
elif line.startswith('[post'):
|
|
in_post = True
|
|
elif in_cat:
|
|
fields = line.split('","')
|
|
if not line:
|
|
in_cat = False
|
|
else:
|
|
# remove 1st and last ""
|
|
fields[0] = fields[0][1:]
|
|
# fields[-1] = fields[-1][:-1]
|
|
category_list[fields[0]]=fields[2]
|
|
elif in_post:
|
|
if not line:
|
|
in_post = False
|
|
break
|
|
else:
|
|
posts.append(line)
|
|
|
|
print("%i posts read." % len(posts))
|
|
|
|
for post in posts:
|
|
fields = post.split('","')
|
|
|
|
# post_id = fields[0][1:]
|
|
# blog_id = fields[1]
|
|
# user_id = fields[2]
|
|
cat_id = fields[3]
|
|
# post_dt = fields[4]
|
|
# post_tz = fields[5]
|
|
post_creadt = fields[6]
|
|
# post_upddt = fields[7]
|
|
# post_password = fields[8]
|
|
# post_type = fields[9]
|
|
post_format = fields[10]
|
|
# post_url = fields[11]
|
|
# post_lang = fields[12]
|
|
post_title = fields[13]
|
|
post_excerpt = fields[14]
|
|
post_excerpt_xhtml = fields[15]
|
|
post_content = fields[16]
|
|
post_content_xhtml = fields[17]
|
|
# post_notes = fields[18]
|
|
# post_words = fields[19]
|
|
# post_status = fields[20]
|
|
# post_selected = fields[21]
|
|
# post_position = fields[22]
|
|
# post_open_comment = fields[23]
|
|
# post_open_tb = fields[24]
|
|
# nb_comment = fields[25]
|
|
# nb_trackback = fields[26]
|
|
post_meta = fields[27]
|
|
# redirect_url = fields[28][:-1]
|
|
|
|
# remove seconds
|
|
post_creadt = ':'.join(post_creadt.split(':')[0:2])
|
|
|
|
author = ""
|
|
categories = []
|
|
tags = []
|
|
|
|
if cat_id:
|
|
categories = [category_list[id].strip() for id in cat_id.split(',')]
|
|
|
|
# Get tags related to a post
|
|
tag = post_meta.replace('{', '').replace('}', '').replace('a:1:s:3:\\"tag\\";a:', '').replace('a:0:', '')
|
|
if len(tag) > 1:
|
|
if int(tag[:1]) == 1:
|
|
newtag = tag.split('"')[1]
|
|
tags.append(
|
|
BeautifulSoup(
|
|
newtag
|
|
, "xml"
|
|
)
|
|
# bs4 always outputs UTF-8
|
|
.decode('utf-8')
|
|
)
|
|
else:
|
|
i=1
|
|
j=1
|
|
while(i <= int(tag[:1])):
|
|
newtag = tag.split('"')[j].replace('\\','')
|
|
tags.append(
|
|
BeautifulSoup(
|
|
newtag
|
|
, "xml"
|
|
)
|
|
# bs4 always outputs UTF-8
|
|
.decode('utf-8')
|
|
)
|
|
i=i+1
|
|
if j < int(tag[:1])*2:
|
|
j=j+2
|
|
|
|
"""
|
|
dotclear2 does not use markdown by default unless you use the markdown plugin
|
|
Ref: http://plugins.dotaddict.org/dc2/details/formatting-markdown
|
|
"""
|
|
if post_format == "markdown":
|
|
content = post_excerpt + post_content
|
|
else:
|
|
content = post_excerpt_xhtml + post_content_xhtml
|
|
content = content.replace('\\n', '')
|
|
post_format = "html"
|
|
|
|
yield (post_title, content, slugify(post_title), post_creadt, author, categories, tags, post_format)
|
|
|
|
|
|
def feed2fields(file):
|
|
"""Read a feed and yield pelican fields"""
|
|
import feedparser
|
|
d = feedparser.parse(file)
|
|
for entry in d.entries:
|
|
date = (time.strftime("%Y-%m-%d %H:%M", entry.updated_parsed)
|
|
if hasattr(entry, "updated_parsed") else None)
|
|
author = entry.author if hasattr(entry, "author") else None
|
|
tags = [e['term'] for e in entry.tags] if hasattr(entry, "tags") else None
|
|
|
|
slug = slugify(entry.title)
|
|
yield (entry.title, entry.description, slug, date, author, [], tags, "html")
|
|
|
|
|
|
def build_header(title, date, author, categories, tags, slug):
|
|
"""Build a header from a list of fields"""
|
|
header = '%s\n%s\n' % (title, '#' * len(title))
|
|
if date:
|
|
header += ':date: %s\n' % date
|
|
if author:
|
|
header += ':author: %s\n' % author
|
|
if categories:
|
|
header += ':category: %s\n' % ', '.join(categories)
|
|
if tags:
|
|
header += ':tags: %s\n' % ', '.join(tags)
|
|
if slug:
|
|
header += ':slug: %s\n' % slug
|
|
header += '\n'
|
|
return header
|
|
|
|
def build_markdown_header(title, date, author, categories, tags, slug):
|
|
"""Build a header from a list of fields"""
|
|
header = 'Title: %s\n' % title
|
|
if date:
|
|
header += 'Date: %s\n' % date
|
|
if author:
|
|
header += 'Author: %s\n' % author
|
|
if categories:
|
|
header += 'Category: %s\n' % ', '.join(categories)
|
|
if tags:
|
|
header += 'Tags: %s\n' % ', '.join(tags)
|
|
if slug:
|
|
header += 'Slug: %s\n' % slug
|
|
header += '\n'
|
|
return header
|
|
|
|
def fields2pelican(fields, out_markup, output_path, dircat=False, strip_raw=False, disable_slugs=False):
|
|
for title, content, filename, date, author, categories, tags, in_markup in fields:
|
|
slug = not disable_slugs and filename or None
|
|
if (in_markup == "markdown") or (out_markup == "markdown") :
|
|
ext = '.md'
|
|
header = build_markdown_header(title, date, author, categories, tags, slug)
|
|
else:
|
|
out_markup = "rst"
|
|
ext = '.rst'
|
|
header = build_header(title, date, author, categories, tags, slug)
|
|
|
|
filename = os.path.basename(filename)
|
|
|
|
# option to put files in directories with categories names
|
|
if dircat and (len(categories) > 0):
|
|
catname = slugify(categories[0])
|
|
out_filename = os.path.join(output_path, catname, filename+ext)
|
|
if not os.path.isdir(os.path.join(output_path, catname)):
|
|
os.mkdir(os.path.join(output_path, catname))
|
|
else:
|
|
out_filename = os.path.join(output_path, filename+ext)
|
|
|
|
print(out_filename)
|
|
|
|
if in_markup == "html":
|
|
html_filename = os.path.join(output_path, filename+'.html')
|
|
|
|
with open(html_filename, 'w', encoding='utf-8') as fp:
|
|
# Replace newlines with paragraphs wrapped with <p> so
|
|
# HTML is valid before conversion
|
|
paragraphs = content.splitlines()
|
|
paragraphs = ['<p>{0}</p>'.format(p) for p in paragraphs]
|
|
new_content = ''.join(paragraphs)
|
|
|
|
fp.write(new_content)
|
|
|
|
|
|
parse_raw = '--parse-raw' if not strip_raw else ''
|
|
cmd = ('pandoc --normalize --reference-links {0} --from=html'
|
|
' --to={1} -o "{2}" "{3}"').format(
|
|
parse_raw, out_markup, out_filename, html_filename)
|
|
|
|
try:
|
|
rc = subprocess.call(cmd, shell=True)
|
|
if rc < 0:
|
|
error = "Child was terminated by signal %d" % -rc
|
|
exit(error)
|
|
|
|
elif rc > 0:
|
|
error = "Please, check your Pandoc installation."
|
|
exit(error)
|
|
except OSError as e:
|
|
error = "Pandoc execution failed: %s" % e
|
|
exit(error)
|
|
|
|
os.remove(html_filename)
|
|
|
|
with open(out_filename, 'r', encoding='utf-8') as fs:
|
|
content = fs.read()
|
|
if out_markup == "markdown":
|
|
# In markdown, to insert a <br />, end a line with two or more spaces & then a end-of-line
|
|
content = content.replace("\\\n ", " \n")
|
|
content = content.replace("\\\n", " \n")
|
|
|
|
with open(out_filename, 'w', encoding='utf-8') as fs:
|
|
fs.write(header + content)
|
|
|
|
|
|
def main():
|
|
parser = argparse.ArgumentParser(
|
|
description="Transform feed, Wordpress or Dotclear files to reST (rst) "
|
|
"or Markdown (md) files. Be sure to have pandoc installed.",
|
|
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
|
|
|
|
parser.add_argument(dest='input', help='The input file to read')
|
|
parser.add_argument('--wpfile', action='store_true', dest='wpfile',
|
|
help='Wordpress XML export')
|
|
parser.add_argument('--dotclear', action='store_true', dest='dotclear',
|
|
help='Dotclear export')
|
|
parser.add_argument('--feed', action='store_true', dest='feed',
|
|
help='Feed to parse')
|
|
parser.add_argument('-o', '--output', dest='output', default='output',
|
|
help='Output path')
|
|
parser.add_argument('-m', '--markup', dest='markup', default='rst',
|
|
help='Output markup format (supports rst & markdown)')
|
|
parser.add_argument('--dir-cat', action='store_true', dest='dircat',
|
|
help='Put files in directories with categories name')
|
|
parser.add_argument('--strip-raw', action='store_true', dest='strip_raw',
|
|
help="Strip raw HTML code that can't be converted to "
|
|
"markup such as flash embeds or iframes (wordpress import only)")
|
|
parser.add_argument('--disable-slugs', action='store_true',
|
|
dest='disable_slugs',
|
|
help='Disable storing slugs from imported posts within output. '
|
|
'With this disabled, your Pelican URLs may not be consistent '
|
|
'with your original posts.')
|
|
|
|
args = parser.parse_args()
|
|
|
|
input_type = None
|
|
if args.wpfile:
|
|
input_type = 'wordpress'
|
|
elif args.dotclear:
|
|
input_type = 'dotclear'
|
|
elif args.feed:
|
|
input_type = 'feed'
|
|
else:
|
|
error = "You must provide either --wpfile, --dotclear or --feed options"
|
|
exit(error)
|
|
|
|
if not os.path.exists(args.output):
|
|
try:
|
|
os.mkdir(args.output)
|
|
except OSError:
|
|
error = "Unable to create the output folder: " + args.output
|
|
exit(error)
|
|
|
|
if input_type == 'wordpress':
|
|
fields = wp2fields(args.input)
|
|
elif input_type == 'dotclear':
|
|
fields = dc2fields(args.input)
|
|
elif input_type == 'feed':
|
|
fields = feed2fields(args.input)
|
|
|
|
fields2pelican(fields, args.markup, args.output,
|
|
dircat=args.dircat or False,
|
|
strip_raw=args.strip_raw or False,
|
|
disable_slugs=args.disable_slugs or False)
|