1
0
Fork 0
mirror of https://github.com/marty-oehme/scripts.git synced 2024-12-22 07:58:08 +00:00
qownnotes-scripts/inbox/inbox.py
2017-09-14 22:25:44 +03:00

352 lines
14 KiB
Python

#!/usr/bin/env python
import os
import re
import sys
import time
import argparse
import collections
import multiprocessing.dummy
import md_link
import md_convert
import safe_path
try:
import watchdog.events
import watchdog.observers
except ImportError:
pass
File_attrs = collections.namedtuple('File_attrs', 'file_path folder_dir_path output_dir_path output_file')
"""
A named tuple which functions use to pass input data - data of files to be processed
:param file_path: full absolute path to the file to process
:param folder_dir_path: full absolute path to directory where 'media' and 'attachment' directories are
:param output_dir_path: full absolute path to directory where resulting text file will be stored
:param output_file: empty for new standalone text file with mtime in the name,
'*no mtime*' for or new standalone text file without mtime in the name
or full absolute path to the text file which will be appended with a new entry
"""
Note_attrs = collections.namedtuple('Note_attrs', 'input_file_path output_file_path text mtime title')
'''A named tuple which functions use to pass output data - data of notes to be written.
:param input_file_path: full absolute path to the file which was processed to this tuple
:param output_file_path: full absolute path to the output text file which should be written
:param text: content of the text file which should be written
:param mtime: modification time of input file as markdown headline to optionally prepend a text
:param title: title of a input file as markdown headline to optionally prepend a text'''
def text_to_md(file_attrs, topic_marker):
"""
This will process specified text file getting its topics and replacing urls with favicons and titles where possible
:param file_attrs: File_attrs named tuple
:param topic_marker: symbol(s) which start the 'topic' word, if such word present in text, it will go to 'topic.md'
:return: list of Note_attrs named tuple
"""
filename = os.path.splitext(os.path.basename(file_attrs.file_path))[0]
mtime = time.localtime(os.path.getmtime(file_attrs.file_path))
try:
with open(file_attrs.file_path, 'r') as text_file:
text = text_file.read()
except UnicodeDecodeError:
return
topics = re.findall(topic_marker + '(\w*)', text)
text = re.sub(topic_marker + '\w*[ ]?', '', text).strip()
if re.match('^http[s]?://[^\s]*$', text):
is_bookmark = True
else:
is_bookmark = False
for link in re.findall('(^|\s)(http[s]?://.*)(\s|$)', text, re.MULTILINE | re.IGNORECASE):
url = md_link.URL(link[1], file_attrs.folder_dir_path)
text = text.replace(link[1], url.md)
if is_bookmark:
bookmark_title = url.title
if file_attrs.output_file and file_attrs.output_file != '*no mtime*':
output_files = [file_attrs.output_file]
headline_title = ''
elif topics:
output_files = [topic + '.md' for topic in topics]
headline_title = ''
elif is_bookmark:
headline_title = '# {}\n'.format(bookmark_title)
if file_attrs.output_file == '*no mtime*':
output_files = [bookmark_title + '.md']
else:
output_files = [time.strftime('%m-%d %H:%M', mtime) + ' ' + bookmark_title + '.md']
else:
headline_title = '# {}\n'.format(filename)
if file_attrs.output_file == '*no mtime*':
output_files = [filename + '.md']
else:
output_files = [time.strftime('%m-%d %H:%M', mtime) + ' ' + filename + '.md']
output = []
for output_file in output_files:
output.append(Note_attrs(input_file_path=file_attrs.file_path,
output_file_path=file_attrs.output_dir_path + os.sep + safe_path.filename(output_file),
text=text,
mtime='**{}** \n'.format(time.strftime('%x %a %X', mtime)),
title=headline_title))
return output
def html_to_md(file_attrs, pandoc_bin='pandoc', pandoc_ver=''):
"""
This will move specified convert specified html file to markdown and move all in-line images to sub-folder at media directory
:param file_attrs: File_attrs named tuple
:return: Note_attrs named tuple
"""
html_file_name_noext = os.path.splitext(os.path.basename(file_attrs.file_path))[0]
mtime = time.localtime(os.path.getmtime(file_attrs.file_path))
md_text = md_convert.saved_html(file_attrs.file_path, file_attrs.folder_dir_path,
pandoc_bin=pandoc_bin, pandoc_ver=pandoc_ver)
if not md_text:
return
return Note_attrs(input_file_path=file_attrs.file_path,
output_file_path=file_attrs.output_dir_path + os.sep + safe_path.filename(html_file_name_noext + '.md'),
text=md_text,
mtime='**{}** \n'.format(time.strftime('%x %a %X', mtime)),
title='')
def file_to_md(file_attrs, media_dir_name):
"""
This will move specified file to media_dir_name and put note with a reference to that file instead
:param file_attrs: File_attrs named tuple
:param media_dir_name: name of sub-directory in folder_dir_path where file will be moved (for non-text files)
:return: Note_attrs named tuple
"""
mtime = time.localtime(os.path.getmtime(file_attrs.file_path))
new_filename = str(time.mktime(mtime))[:-2] + '_' + os.path.basename(file_attrs.file_path)
new_path = os.path.join(file_attrs.folder_dir_path, media_dir_name, new_filename)
try:
os.rename(file_attrs.file_path, new_path)
except OSError:
pass
file = md_link.File(new_path, file_attrs.folder_dir_path, os.path.splitext(os.path.basename(file_attrs.file_path))[0])
if file_attrs.output_file == '*no mtime*':
output_file = file.title + '.md'
elif file_attrs.output_file:
output_file = file_attrs.output_file
else:
output_file = time.strftime('%m-%d %H:%M', mtime) + ' ' + file.title + '.md'
return Note_attrs(input_file_path=file_attrs.file_path,
output_file_path=file_attrs.output_dir_path + os.sep + safe_path.filename(output_file),
text=file.md,
mtime='**{}** \n'.format(time.strftime('%x %a %X', mtime)),
title='# {}\n'.format(file.title))
def make_flat_list(mixed_list, target_item_type=tuple):
"""
Make a list that has lists and 'target_item_type' as items flat, not recursive.
:param mixed_list: list to make flat
:param target_item_type: type of items in the flat list
:return: flat list of 'target_item_type'
"""
flat_list = []
for obj in mixed_list:
if type(obj) == list:
for item in obj:
if type(item) == target_item_type:
flat_list.append(item)
elif type(obj) == target_item_type:
flat_list.append(obj)
return flat_list
def process_by_path(file_path):
"""
Checks if the file is valid for processing and returns File_attrs tuple depending on its path
:param file_path: Absolute file path
:return: File_attrs named tuple
"""
if file_path.endswith(('.md', 'notes.sqlite')) \
or file_path.startswith((folder_dir + os.sep + 'media', folder_dir + os.sep + 'attachments')) \
or os.sep + '.' in file_path[len(folder_dir):] \
or '_files' + os.sep in file_path[len(folder_dir):]:
return
if file_path[:len(inbox_dir)] == inbox_dir:
if os.path.dirname(file_path) == inbox_dir:
return File_attrs(file_path=file_path, folder_dir_path=folder_dir,
output_dir_path=inbox_dir, output_file='')
else:
return File_attrs(file_path=file_path, folder_dir_path=folder_dir,
output_dir_path=inbox_dir,
output_file=os.path.dirname(file_path)[len(inbox_dir)+1:].replace(os.sep, ' - ') + '.md')
else:
return File_attrs(file_path=file_path, folder_dir_path=folder_dir,
output_dir_path=os.path.dirname(file_path), output_file='*no mtime*')
def process_by_ext(file_attrs):
"""
This will run different functions to process specified File_attrs tuple based on file extension
:param file_attrs: File_attrs named tuple
:return: Note_attrs named tuple
"""
if file_attrs.file_path.endswith('.txt') or not os.path.splitext(file_attrs.file_path)[1]:
return text_to_md(file_attrs, args.topic_marker)
elif args.pandoc_bin and args.pandoc_ver and file_attrs.file_path.endswith(('.htm', '.html')):
return html_to_md(file_attrs, args.pandoc_bin, args.pandoc_ver)
elif file_attrs.file_path.endswith(('.jpg', '.png', '.gif')):
return file_to_md(file_attrs, 'media')
else:
return file_to_md(file_attrs, 'attachments')
def write_note_and_delete(note_attrs):
"""
Create or append existing note files based on Note_attrs tuples data, then delete the source file
:param note_attrs: Note_attrs named tuple
"""
if os.path.isfile(note_attrs.output_file_path):
if os.path.dirname(note_attrs.output_file_path) == inbox_dir:
note_file_path = note_attrs.output_file_path
with open(note_file_path, 'r') as source:
content = note_attrs.mtime + note_attrs.text + '\n\n' + source.read()
else:
i = 1
while os.path.isfile(os.path.splitext(note_attrs.output_file_path)[0] + '_' + str(i) + '.md'):
i += 1
note_file_path = os.path.splitext(note_attrs.output_file_path)[0] + '_' + str(i) + '.md'
content = note_attrs.mtime + note_attrs.text
else:
note_file_path = note_attrs.output_file_path
if note_attrs.title:
content = note_attrs.title + note_attrs.text
else:
content = note_attrs.mtime + note_attrs.text
with open(note_file_path, 'w') as output:
output.write(content)
if os.path.isfile(note_file_path):
try:
os.remove(note_attrs.input_file_path)
except OSError:
pass
if __name__ == '__main__':
script_path = os.path.dirname(sys.argv[0])
for file in os.listdir(script_path):
if file[-5:] == '.lock':
os.remove(script_path + os.sep + file)
arg_parser = argparse.ArgumentParser(description='A script to turn everything in the inbox directory to markdown notes.')
arg_parser.add_argument('-i', '--inbox', action='store', dest='inbox_dir', required=True,
help="Full absolute path to the inbox directory to organize")
arg_parser.add_argument('-f', '--folder', action='store', dest='folder_dir', required=True,
help="Full absolute path to directory where 'media' and 'attachment' directories are")
arg_parser.add_argument('-m', '--marker', action='store', dest='topic_marker', required=False, default='@',
help="Symbol(s) which start the 'topic' word (for text files)")
arg_parser.add_argument('-s', '--scan-folder', action='store_true', dest='scan_folder', required=False,
help="Process whole folder rather than only inbox")
arg_parser.add_argument('-p', '--pandoc-bin', action='store', dest='pandoc_bin', required=False,
help="Command/path to run pandoc")
arg_parser.add_argument('-pv', '--pandoc-ver', action='store', dest='pandoc_ver', required=False,
help="Installed pandoc version")
arg_parser.add_argument('-w', '--watch', action='store_true', dest='watch_fs', required=False,
help="Watch and process new files as they appear after initial scan")
args = arg_parser.parse_args()
inbox_dir = args.inbox_dir
folder_dir = args.folder_dir
os.makedirs(inbox_dir, exist_ok=True)
os.makedirs(folder_dir + os.sep + 'media', exist_ok=True)
os.makedirs(folder_dir + os.sep + 'attachments', exist_ok=True)
if args.scan_folder:
scan_path = folder_dir
else:
scan_path = inbox_dir
file_list = []
for root, subdirs, files in os.walk(scan_path):
for file_path in sorted([root + os.sep + file for file in files], key=os.path.getmtime):
file_attrs = process_by_path(file_path)
if file_attrs:
file_list.append([file_attrs])
write_list = multiprocessing.dummy.Pool(100).starmap(process_by_ext, file_list)
flat_write_list = make_flat_list(write_list, Note_attrs)
for note_attrs in flat_write_list:
write_note_and_delete(note_attrs)
if args.watch_fs:
try:
import watchdog.events
import watchdog.observers
except ImportError:
print("Can't find Watchdog module. Watching for changes won't work.")
exit(1)
class FsEventHandler(watchdog.events.FileSystemEventHandler):
def on_any_event(self, event):
if event.is_directory:
return
elif event.event_type == 'created':
file_path = event.src_path
elif event.event_type == 'moved':
file_path = event.dest_path
else:
return
file_attrs = process_by_path(file_path)
if file_attrs:
# Wait for all the web page resources saved/synced
if file_path.endswith(('.htm', '.html')):
time.sleep(2)
obj_to_write = process_by_ext(file_attrs)
else:
return
if type(obj_to_write) == list:
for note_attrs in obj_to_write:
write_note_and_delete(note_attrs)
else:
write_note_and_delete(obj_to_write)
lockfile_path = script_path + os.sep + str(int(time.time())) + '.lock'
open(lockfile_path, 'w').close()
event_handler = FsEventHandler()
observer = watchdog.observers.Observer()
observer.schedule(event_handler, scan_path, recursive=True)
observer.start()
try:
while True:
if os.path.isfile(lockfile_path):
time.sleep(5)
else:
raise Exception
except:
observer.stop()
observer.join()