diff options
author | mathieui <mathieui@mathieui.net> | 2014-04-05 17:57:45 +0200 |
---|---|---|
committer | mathieui <mathieui@mathieui.net> | 2014-04-05 17:57:45 +0200 |
commit | 8def5609d925c89004c906d67a7e1d3028706f32 (patch) | |
tree | 8fea6007ca73b82e0be992b90dc472634a050701 /src | |
parent | 587450939b5f2ab3b4cb46f44c6ab08e15722caf (diff) | |
download | poezio-8def5609d925c89004c906d67a7e1d3028706f32.tar.gz poezio-8def5609d925c89004c906d67a7e1d3028706f32.tar.bz2 poezio-8def5609d925c89004c906d67a7e1d3028706f32.tar.xz poezio-8def5609d925c89004c906d67a7e1d3028706f32.zip |
Remove the pubsub-related stuff from the main branch
as it is unused code anyway.
Diffstat (limited to 'src')
-rw-r--r-- | src/atom_parser.py | 39 | ||||
-rw-r--r-- | src/core/commands.py | 17 | ||||
-rw-r--r-- | src/core/core.py | 1 | ||||
-rw-r--r-- | src/pubsub.py | 472 |
4 files changed, 0 insertions, 529 deletions
diff --git a/src/atom_parser.py b/src/atom_parser.py deleted file mode 100644 index bd1de8d0..00000000 --- a/src/atom_parser.py +++ /dev/null @@ -1,39 +0,0 @@ -# Copyright 2010-2011 Florent Le Coz <louiz@louiz.org> -# -# This file is part of Poezio. -# -# Poezio is free software: you can redistribute it and/or modify -# it under the terms of the zlib license. See the COPYING file. - -""" -Defines a function returning a dict containing the values from an -atom entry contained in a pubsub entry -""" - -ATOM_XMLNS = "http://www.w3.org/2005/Atom" - -def parse_atom_entry(pubsub_item): - """ - Takes a pubsub ET.Element item and returns a dict containing - all needed values from the atom entry element. - Returns None if the item does not contain an atom entry. - """ - entry_elem = pubsub_item.find('{%s}entry' % (ATOM_XMLNS,)) - if entry_elem is None: - return None - res = {'author':{}} - author_elem = entry_elem.find('{%s}author' % (ATOM_XMLNS,)) - if author_elem is not None: - for sub in ('name', 'uri'): - sub_elem = author_elem.find('{%s}%s' % (ATOM_XMLNS, sub,)) - if sub_elem is not None: - res['author'][sub] = sub_elem.text - for elem_name in {'title':'text', 'updated':'date', 'published': 'date', - 'summary':'text'}: - elem = entry_elem.find('{%s}%s' % (ATOM_XMLNS, elem_name,)) - if elem is not None: - res[elem_name] = elem.text - link_elem = entry_elem.find('{%s}link' % (ATOM_XMLNS,)) - if link_elem is not None: - res['link_href'] = link_elem.attrib.get('href') or '' - return res diff --git a/src/core/commands.py b/src/core/commands.py index fb8dc2eb..5be95207 100644 --- a/src/core/commands.py +++ b/src/core/commands.py @@ -20,7 +20,6 @@ import bookmark import common import fixes import pep -import pubsub import tabs import theming from common import safeJID @@ -764,22 +763,6 @@ def command_bind(self, arg): else: self.information('%s is now unbound' % args[0], 'Info') -def command_pubsub(self, args): - """ - Opens a pubsub browser on the given domain - """ - args = common.shell_split(args) - if len(args) != 1: - return self.command_help('pubsub') - domain = args[0] - tab = self.get_tab_by_name('%s@@pubsubbrowser' % (domain,), pubsub.PubsubBrowserTab) - if tab: - self.command_win('%s' % tab.nb) - else: - new_tab = pubsub.PubsubBrowserTab(domain) - self.add_tab(new_tab, True) - self.refresh_window() - def command_rawxml(self, arg): """ /rawxml <xml stanza> diff --git a/src/core/core.py b/src/core/core.py index 2ef6642d..9ae68c55 100644 --- a/src/core/core.py +++ b/src/core/core.py @@ -1681,7 +1681,6 @@ class Core(object): command_invitations = commands.command_invitations command_quit = commands.command_quit command_bind = commands.command_bind - command_pubsub = commands.command_pubsub command_rawxml = commands.command_rawxml command_load = commands.command_load command_unload = commands.command_unload diff --git a/src/pubsub.py b/src/pubsub.py deleted file mode 100644 index 55e9bc38..00000000 --- a/src/pubsub.py +++ /dev/null @@ -1,472 +0,0 @@ -# Copyright 2010-2011 Florent Le Coz <louiz@louiz.org> -# -# This file is part of Poezio. -# -# Poezio is free software: you can redistribute it and/or modify -# it under the terms of the zlib license. See the COPYING file. - -import logging -log = logging.getLogger(__name__) - -import os -import curses -import tempfile -import subprocess - -import tabs -import windows -import atom_parser - -from config import config - -from datetime import datetime -from sleekxmpp.xmlstream import ET - -ATOM_XMLNS = "http://www.w3.org/2005/Atom" - -item_template = \ -""" --- About you --- Your name -Author: %(author)s - --- Your jid (e.g. xmpp:romeo@example.com) or website’s url (http://example.com) -URI: xmpp:%(jid)s - --- Your email address (e.g. romeo@mail.com) -email: -Title: - --- Please use the form dd/mm/yyy hh:mm:ss, or leave empty --- If you leave the date empty it will automatically be filled with the current time and date (recommended) -Date: - ---- body --- Until the end of file, this will be your item's body - -""" - -def parse_template(lines): - """ - takes a template string (splitted by lines) and returns a dict with the correstponding values - """ - lines = [line.lower() for line in lines] - res = {} - reading_body = False - body = [] - for line in lines: - if line.startswith('--- body ---'): - reading_body = True - continue - if reading_body: - body.append(line) - continue - if line.startswith('-- '): - continue - for value in ('author', 'uri', 'email', 'title', 'date'): - if line.startswith('%s:' % (value,)): - res[value] = line.split(':', 1)[1].strip() - res['body'] = ''.join(body) - return res - -def create_entry_from_dict(dic): - """ - Takes a dict with the correct values and returns an ET.Element - representing an Atom entry - """ - entry_elem = ET.Element('entry', xmlns=ATOM_XMLNS) - - author_elem = ET.Element('author', xmlns=ATOM_XMLNS) - if dic.get('author'): - name_elem = ET.Element('name', xmlns=ATOM_XMLNS) - name_elem.text = dic.get('author') - author_elem.append(name_elem) - if dic.get('uri'): - uri_elem = ET.Element('uri', xmlns=ATOM_XMLNS) - uri_elem.text = dic.get('uri') - author_elem.append(uri_elem) - if dic.get('email'): - email_elem = ET.Element('email', xmlns=ATOM_XMLNS) - email_elem.text = dic.get('email') - author_elem.append(email_elem) - entry_elem.append(author_elem) - if dic.get('title'): - title_elem = ET.Element('title', xmlns=ATOM_XMLNS) - title_elem.text = dic.get('title') - entry_elem.append(title_elem) - # TODO - # if dic.get('date'): - date_elem = ET.Element('published', xmlns=ATOM_XMLNS) - date_elem.text = '%s' % datetime.now() - entry_elem.append(date_elem) - summary_elem = ET.Element('summary', xmlns=ATOM_XMLNS) - summary_elem.text = dic.get('body') or '' - entry_elem.append(summary_elem) - return entry_elem - -class PubsubNode(object): - node_type = None # unknown yet - def __init__(self, name, parent=None): - self.items = [] - self.name = name - self.parent = parent - - -class LeafNode(PubsubNode): - node_type = "leaf" - def __init__(self, name, parent=None): - PubsubNode.__init__(self, name, parent) - - -class CollectionNode(PubsubNode): - node_type = "collection" - def __init__(self, name, parent=None): - PubsubNode.__init__(self, name, parent) - self.subnodes = [] - - -class PubsubItem(object): - def __init__(self, idd, content): - self.id = idd - self.content = content - self.parsed_content = atom_parser.parse_atom_entry(content) - - def to_dict(self, columns): - """ - returns a dict of the values listed in columns - """ - ret = {} - for col in columns: - ret[col] = self.__dict__.get(col) or '' - if self.parsed_content: - ret['title'] = self.parsed_content.get('title') - ret['author'] = self.parsed_content['author'].get('name') - return ret - -class PubsubBrowserTab(tabs.Tab): - """ - A tab containing a pubsub browser letting the user - list nodes and items, view, add and delete items, etc - """ - def __init__(self, server): - """ - Server is the name of the pubsub server, for example: - pubsub.example.com - All action done in this tab will be made on that server. - """ - tabs.Tab.__init__(self) - self.current_node = None # the subnode we are listing. None means the root - self.server = server - self.nodes = [] # the lower level of nodes - - self.upper_message = windows.Topic() - self.set_info_message('Loading') - - # Node List View - node_columns = ('node', 'name',) - self.node_list_header = windows.ColumnHeaderWin(node_columns) - self.node_listview = windows.ListWin(node_columns) - - # Item List View - item_columns = ('title', 'author', 'id') - self.item_list_header = windows.ColumnHeaderWin(item_columns) - self.item_listview = windows.ListWin(item_columns) - - # Vertical Separator - self.vertical_separator = windows.VerticalSeparator() - - # Item viewer - self.item_viewer = windows.SimpleTextWin('') - self.default_help_message = windows.HelpText("“c”: create a node.") - self.input = self.default_help_message - - self.key_func['c'] = self.command_create_node - self.key_func['p'] = self.command_publish_item - self.key_func["M-KEY_DOWN"] = self.scroll_node_down - self.key_func["M-KEY_UP"] = self.scroll_node_up - self.key_func["KEY_DOWN"] = self.item_listview.move_cursor_down - self.key_func["KEY_UP"] = self.item_listview.move_cursor_up - self.key_func["^M"] = self.open_selected_item - self.resize() - - self.get_nodes() - - def command_publish_item(self): - self.core.background = True - editor = config.get('editor', '') or os.getenv('EDITOR') or 'vi' - log.debug('Starting item edition with command %s' % editor) - tmpfile = tempfile.NamedTemporaryFile(mode='r+') - tmpfile.write(item_template % {'author': self.core.xmpp.boundjid.user, 'jid': self.core.xmpp.boundjid.bare}) - tmpfile.flush() - process = subprocess.call(editor.split() + [tmpfile.name]) - tmpfile.flush() - tmpfile.seek(0) - item_dict = parse_template(tmpfile.readlines()) - tmpfile.close() - log.debug('[%s]' % item_dict) - - self.core.background = False - self.core.full_screen_redraw() - entry = create_entry_from_dict(item_dict) - self.publish_item(entry, self.get_selected_node_name()) - - def publish_item(self, content, node_name): - """ - publish the given item on the given node - """ - def callback(res): - if res: - self.set_info_message('Item published') - else: - self.set_info_message('Item not published') - self.force_refresh() - - self.core.xmpp.plugin['xep_0060'].setItem(self.server, node_name, content, callback=callback) - - def set_info_message(self, message): - """ - Set an informative message in the upper line, near the server name - """ - self.upper_message.set_message('Pubsub server: %s/%s [%s]' % (self.server, self.current_node or '', message)) - - def resize(self): - self.upper_message.resize(1, self.width, 0, 0) - self.tab_win.resize(1, self.width, self.height-2, 0) - - column_size = {'node': self.width//4, - 'name': self.width//4,} - self.node_list_header.resize_columns(column_size) - self.node_list_header.resize(1, self.width//2, 1, 0) - self.node_listview.resize_columns(column_size) - self.node_listview.resize(self.height//2-2, self.width//2, 2, 0) - - w = self.width//2 - column_size = {'id': w//8, - 'title':w//8*5, - 'author':w//8*2} - self.item_list_header.resize_columns(column_size) - self.item_list_header.resize(self.height//2+1, self.width//2, self.height//2, 0) - self.item_listview.resize_columns(column_size) - self.item_listview.resize(self.height//2-3, self.width//2, self.height//2+1, 0) - - self.vertical_separator.resize(self.height-3, 1, 1, self.width//2) - - self.item_viewer.resize(self.height-3, self.width//2+1, 1, self.width//2+1) - self.input.resize(1, self.width, self.height-1, 0) - - def refresh(self): - if self.need_resize: - self.resize() - log.debug(' TAB Refresh: %s'%self.__class__.__name__) - self.upper_message.refresh() - self.node_list_header.refresh() - self.node_listview.refresh() - self.item_list_header.refresh() - self.item_listview.refresh() - self.item_viewer.refresh() - self.tab_win.refresh() - self.vertical_separator.refresh() - self.input.refresh() - - def force_refresh(self): - if self.core.current_tab() is self: - self.core.refresh_window() - - def get_name(self): - return '%s@@pubsubbrowser' % (self.server,) - - def on_input(self, key): - res = self.input.do_command(key) - if res: - return True - if key in self.key_func: - return self.key_func[key]() - - def get_selected_node_name(self): - """ - From the node_view_list, returns the node name of the selected - one. None can be returned - """ - line = self.node_listview.get_selected_row() - if not line: - return None - return line['node'] - - def get_node_by_name(self, name): - """ - in the currently browsed node (or on the root), return the node with that name - """ - nodes = self.current_node and self.current_node.subnodes or self.nodes - for node in nodes: - if node.name == name: - return node - return None - - def get_item_by_id(self, idd): - """ - in the currently selected node, return the item with that id - """ - selected_node_name = self.get_selected_node_name() - if not selected_node_name: - return None - selected_node = self.get_node_by_name(selected_node_name) - if not selected_node: - return None - for item in selected_node.items: - if item.id == idd: - return item - return None - - def get_selected_item_id(self): - """ - returns the id of the currently selected item - """ - line = self.item_listview.get_selected_row() - if not line: - return None - return line['id'] - - def get_items(self, node): - """ - Get all items in the given node - """ - def callback(items): - item_list = [] - if items: - for it in items: - item_list.append(PubsubItem(it.attrib['id'], it)) - node.items = item_list - log.debug('get_selected_node_name: %s' % self.get_selected_node_name()) - if self.get_selected_node_name() == node.name: - self.display_items_from_node(node) - log.debug('Item on node %s: %s' % (node.name, item_list)) - self.set_info_message('Items received') - self.force_refresh() - - self.core.xmpp.plugin['xep_0060'].get_items(self.server, node.name, callback=callback) - - def display_items_from_node(self, node): - """ - takes a node, and set fill the item_listview with that - node’s items - """ - columns = self.item_list_header.get_columns() - self.item_listview.empty() - log.debug('display_items_from_node: %s' % node.items) - for item in node.items: - line = item.to_dict(columns) - self.item_listview.lines.append(item.to_dict(columns)) - - def add_nodes(self, node_list, parent=None): - """ - Add Node objects to the list of the parent. - If parent is None, they are added to the root list. - If the current selected node is parent, we add - them directly to the node_listview - """ - log.debug('Adding nodes to %s: %s' % (node_list, parent,)) - if not parent: - list_to_append = self.nodes - else: - list_to_append = parent.nodes - self.node_listview.add_lines(node_list) - for node in node_list: - new_node = LeafNode(node['node']) - list_to_append.append(new_node) - self.get_items(new_node) - - def get_nodes(self, node=None): - """ - Get all subnodes of the given node. If no node is given, get - the root nodes - """ - def callback(nodes): - lines = [{'name': nodes[node] or '', - 'node': node} for node in nodes.keys()] - self.add_nodes(lines) - self.set_info_message('Nodes received') - self.force_refresh() - self.core.xmpp.plugin['xep_0060'].get_nodes(self.server, callback=callback) - - def create_node(self, node_name): - def callback(res): - if res: - self.node_listview.add_lines([{'name': '', 'node': node_name}]) - self.set_info_message('Node created') - else: - self.set_info_message('Node not created') - self.reset_help_message() - self.force_refresh() - if node_name: - self.core.xmpp.plugin['xep_0060'].create_node(self.server, node_name, callback=callback) - return True - - def reset_help_message(self, txt=None): - """ - Just reset the help message when a command ends - """ - curses.curs_set(0) - self.input = self.default_help_message - return True - - def command_create_node(self): - """ - Prompt for a node name and create it on Enter key - """ - curses.curs_set(1) - self.input = windows.CommandInput("[Create node]", self.reset_help_message, self.create_node, None) - self.input.resize(1, self.width, self.height-1, 0) - return True - - def scroll_node_up(self): - """ - scroll the node up, and update the item list if needed - """ - selected_node_before = self.get_selected_node_name() - self.node_listview.move_cursor_up() - selected_node_after = self.get_selected_node_name() - if selected_node_after is not selected_node_before: - self.display_items_from_node(self.get_node_by_name(selected_node_after)) - return True - return False - - def scroll_node_down(self): - """ - scroll the node down, and update the item list if needed - """ - selected_node_before = self.get_selected_node_name() - self.node_listview.move_cursor_down() - selected_node_after = self.get_selected_node_name() - if selected_node_after is not selected_node_before: - self.display_items_from_node(self.get_node_by_name(selected_node_after)) - return True - return False - - def open_selected_item(self): - """ - displays the currently selected item in the item view window - """ - selected_item = self.get_item_by_id(self.get_selected_item_id()) - if not selected_item: - return - log.debug('Content: %s'%ET.tostring(selected_item.content)) - entry = atom_parser.parse_atom_entry(selected_item.content) - if not entry: - self.item_viewer._text = str(ET.tostring(selected_item.content)) - else: - self.item_viewer._text = \ -"""\x193Title:\x19o %(title)s -\x193Author:\x19o %(author_name)s (%(author_uri)s) -%(dates)s\x193Link:\x19o %(link)s - -\x193Summary:\x19o -%(summary)s -""" % {'title': entry.get('title') or '', - 'author_name': entry['author'].get('name') or '', - 'author_uri': entry['author'].get('uri') or '', - 'link': entry.get('link_href') or '', - 'summary': entry.get('summary') or '', - 'dates': '\x193Published:\x19o %(published)s\n%(updated)s' % {'published':entry.get('published') or '', - 'updated': '' if (entry.get('updated') is None) or (entry.get('published') == entry.get('updated')) else '\x193Published:\x19o %s\n' % entry.get('updated')} - } - self.item_viewer.rebuild_text() - return True |