Mercurial > libervia-backend
view sat_frontends/jp/common.py @ 2652:baccc27d5c5c
plugin XEP-0334: added a method to add hint directly to elements (i.e. on domish.Element instead of message data)
author | Goffi <goffi@goffi.org> |
---|---|
date | Sat, 11 Aug 2018 18:24:55 +0200 |
parents | 56f94936df1e |
children | 003b8b4b56a7 |
line wrap: on
line source
#!/usr/bin/env python2 # -*- coding: utf-8 -*- # jp: a SàT command line tool # Copyright (C) 2009-2018 Jérôme Poisson (goffi@goffi.org) # This program is free software: you can redistribute it and/or modify # it under the terms of the GNU Affero General Public License as published by # the Free Software Foundation, either version 3 of the License, or # (at your option) any later version. # This program is distributed in the hope that it will be useful, # but WITHOUT ANY WARRANTY; without even the implied warranty of # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the # GNU Affero General Public License for more details. # You should have received a copy of the GNU Affero General Public License # along with this program. If not, see <http://www.gnu.org/licenses/>. from sat_frontends.jp.constants import Const as C from sat.core.i18n import _ from sat.core import exceptions from sat.tools.common import regex from sat.tools.common.ansi import ANSI as A from sat.tools.common import uri as xmpp_uri from sat.tools import config from ConfigParser import NoSectionError, NoOptionError from collections import namedtuple from functools import partial import json import os import os.path import time import tempfile import subprocess import glob import shlex # defaut arguments used for some known editors (editing with metadata) VIM_SPLIT_ARGS = "-c 'set nospr|vsplit|wincmd w|next|wincmd w'" EMACS_SPLIT_ARGS = '--eval "(split-window-horizontally)"' EDITOR_ARGS_MAGIC = { "vim": VIM_SPLIT_ARGS + " {content_file} {metadata_file}", "gvim": VIM_SPLIT_ARGS + " --nofork {content_file} {metadata_file}", "emacs": EMACS_SPLIT_ARGS + " {content_file} {metadata_file}", "xemacs": EMACS_SPLIT_ARGS + " {content_file} {metadata_file}", "nano": " -F {content_file} {metadata_file}", } SECURE_UNLINK_MAX = 10 SECURE_UNLINK_DIR = ".backup" METADATA_SUFF = "_metadata.json" def ansi_ljust(s, width): """ljust method handling ANSI escape codes""" cleaned = regex.ansiRemove(s) return s + u" " * (width - len(cleaned)) def ansi_center(s, width): """ljust method handling ANSI escape codes""" cleaned = regex.ansiRemove(s) diff = width - len(cleaned) half = diff / 2 return half * u" " + s + (half + diff % 2) * u" " def ansi_rjust(s, width): """ljust method handling ANSI escape codes""" cleaned = regex.ansiRemove(s) return u" " * (width - len(cleaned)) + s def getTmpDir(sat_conf, cat_dir, sub_dir=None): """Return directory used to store temporary files @param sat_conf(ConfigParser.ConfigParser): instance opened on sat configuration @param cat_dir(unicode): directory of the category (e.g. "blog") @param sub_dir(str): sub directory where data need to be put profile can be used here, or special directory name sub_dir will be escaped to be usable in path (use regex.pathUnescape to find initial str) @return (str): path to the dir """ local_dir = config.getConfig(sat_conf, "", "local_dir", Exception) path = [local_dir.encode("utf-8"), cat_dir.encode("utf-8")] if sub_dir is not None: path.append(regex.pathEscape(sub_dir)) return os.path.join(*path) def parse_args(host, cmd_line, **format_kw): """Parse command arguments @param cmd_line(unicode): command line as found in sat.conf @param format_kw: keywords used for formating @return (list(unicode)): list of arguments to pass to subprocess function """ try: # we split the arguments and add the known fields # we split arguments first to avoid escaping issues in file names return [a.format(**format_kw) for a in shlex.split(cmd_line)] except ValueError as e: host.disp( u"Couldn't parse editor cmd [{cmd}]: {reason}".format(cmd=cmd_line, reason=e) ) return [] class BaseEdit(object): u"""base class for editing commands This class allows to edit file for PubSub or something else. It works with temporary files in SàT local_dir, in a "cat_dir" subdir """ def __init__(self, host, cat_dir, use_metadata=False): """ @param sat_conf(ConfigParser.ConfigParser): instance opened on sat configuration @param cat_dir(unicode): directory to use for drafts this will be a sub-directory of SàT's local_dir @param use_metadata(bool): True is edition need a second file for metadata most of signature change with use_metadata with an additional metadata argument. This is done to raise error if a command needs metadata but forget the flag, and vice versa """ self.host = host self.sat_conf = config.parseMainConf() self.cat_dir_str = cat_dir.encode("utf-8") self.use_metadata = use_metadata def secureUnlink(self, path): """Unlink given path after keeping it for a while This method is used to prevent accidental deletion of a draft If there are more file in SECURE_UNLINK_DIR than SECURE_UNLINK_MAX, older file are deleted @param path(str): file to unlink """ if not os.path.isfile(path): raise OSError(u"path must link to a regular file") if not path.startswith(getTmpDir(self.sat_conf, self.cat_dir_str)): self.disp( u"File {} is not in SàT temporary hierarchy, we do not remove it".format( path.decode("utf-8") ), 2, ) return # we have 2 files per draft with use_metadata, so we double max unlink_max = SECURE_UNLINK_MAX * 2 if self.use_metadata else SECURE_UNLINK_MAX backup_dir = getTmpDir(self.sat_conf, self.cat_dir_str, SECURE_UNLINK_DIR) if not os.path.exists(backup_dir): os.makedirs(backup_dir) filename = os.path.basename(path) backup_path = os.path.join(backup_dir, filename) # we move file to backup dir self.host.disp( u"Backuping file {src} to {dst}".format( src=path.decode("utf-8"), dst=backup_path.decode("utf-8") ), 1, ) os.rename(path, backup_path) # and if we exceeded the limit, we remove older file backup_files = [os.path.join(backup_dir, f) for f in os.listdir(backup_dir)] if len(backup_files) > unlink_max: backup_files.sort(key=lambda path: os.stat(path).st_mtime) for path in backup_files[: len(backup_files) - unlink_max]: self.host.disp(u"Purging backup file {}".format(path.decode("utf-8")), 2) os.unlink(path) def runEditor( self, editor_args_opt, content_file_path, content_file_obj, meta_file_path=None, meta_ori=None, ): """run editor to edit content and metadata @param editor_args_opt(unicode): option in [jp] section in configuration for specific args @param content_file_path(str): path to the content file @param content_file_obj(file): opened file instance @param meta_file_path(str, None): metadata file path if None metadata will not be used @param meta_ori(dict, None): original cotent of metadata can't be used if use_metadata is False """ if not self.use_metadata: assert meta_file_path is None assert meta_ori is None # we calculate hashes to check for modifications import hashlib content_file_obj.seek(0) tmp_ori_hash = hashlib.sha1(content_file_obj.read()).digest() content_file_obj.close() # we prepare arguments editor = config.getConfig(self.sat_conf, "jp", "editor") or os.getenv( "EDITOR", "vi" ) try: # is there custom arguments in sat.conf ? editor_args = config.getConfig( self.sat_conf, "jp", editor_args_opt, Exception ) except (NoOptionError, NoSectionError): # no, we check if we know the editor and have special arguments if self.use_metadata: editor_args = EDITOR_ARGS_MAGIC.get(os.path.basename(editor), "") else: editor_args = "" parse_kwargs = {"content_file": content_file_path} if self.use_metadata: parse_kwargs["metadata_file"] = meta_file_path args = parse_args(self.host, editor_args, **parse_kwargs) if not args: args = [content_file_path] # actual editing editor_exit = subprocess.call([editor] + args) # edition will now be checked, and data will be sent if it was a success if editor_exit != 0: self.disp( u"Editor exited with an error code, so temporary file has not be deleted, and item is not published.\nYou can find temporary file at {path}".format( path=content_file_path ), error=True, ) else: # main content try: with open(content_file_path, "rb") as f: content = f.read() except (OSError, IOError): self.disp( u"Can read file at {content_path}, have it been deleted?\nCancelling edition".format( content_path=content_file_path ), error=True, ) self.host.quit(C.EXIT_NOT_FOUND) # metadata if self.use_metadata: try: with open(meta_file_path, "rb") as f: metadata = json.load(f) except (OSError, IOError): self.disp( u"Can read file at {meta_file_path}, have it been deleted?\nCancelling edition".format( content_path=content_file_path, meta_path=meta_file_path ), error=True, ) self.host.quit(C.EXIT_NOT_FOUND) except ValueError: self.disp( u"Can't parse metadata, please check it is correct JSON format. Cancelling edition.\n" + "You can find tmp file at {content_path} and temporary meta file at {meta_path}.".format( content_path=content_file_path, meta_path=meta_file_path ), error=True, ) self.host.quit(C.EXIT_DATA_ERROR) if self.use_metadata and not C.bool(metadata.get("publish", "true")): self.disp( u'Publication blocked by "publish" key in metadata, cancelling edition.\n\n' + "temporary file path:\t{content_path}\nmetadata file path:\t{meta_path}".format( content_path=content_file_path, meta_path=meta_file_path ), error=True, ) self.host.quit() if len(content) == 0: self.disp(u"Content is empty, cancelling the edition") if not content_file_path.startswith( getTmpDir(self.sat_conf, self.cat_dir_str) ): self.disp( u"File are not in SàT temporary hierarchy, we do not remove them", 2, ) self.host.quit() self.disp(u"Deletion of {}".format(content_file_path.decode("utf-8")), 2) os.unlink(content_file_path) if self.use_metadata: self.disp(u"Deletion of {}".format(meta_file_path.decode("utf-8")), 2) os.unlink(meta_file_path) self.host.quit() # time to re-check the hash elif tmp_ori_hash == hashlib.sha1(content).digest() and ( not self.use_metadata or meta_ori == metadata ): self.disp(u"The content has not been modified, cancelling the edition") self.host.quit() else: # we can now send the item content = content.decode("utf-8-sig") # we use utf-8-sig to avoid BOM try: if self.use_metadata: self.publish(content, metadata) else: self.publish(content) except Exception as e: if self.use_metadata: self.disp( u"Error while sending your item, the temporary files have been kept at {content_path} and {meta_path}: {reason}".format( content_path=content_file_path, meta_path=meta_file_path, reason=e, ), error=True, ) else: self.disp( u"Error while sending your item, the temporary file has been kept at {content_path}: {reason}".format( content_path=content_file_path, reason=e ), error=True, ) self.host.quit(1) self.secureUnlink(content_file_path) if self.use_metadata: self.secureUnlink(meta_file_path) def publish(self, content): # if metadata is needed, publish will be called with it last argument raise NotImplementedError def getTmpFile(self): """Create a temporary file @param suff (str): suffix to use for the filename @return (tuple(file, str)): opened (w+b) file object and file path """ suff = "." + self.getTmpSuff() cat_dir_str = self.cat_dir_str tmp_dir = getTmpDir(self.sat_conf, self.cat_dir_str, self.profile.encode("utf-8")) if not os.path.exists(tmp_dir): try: os.makedirs(tmp_dir) except OSError as e: self.disp( u"Can't create {path} directory: {reason}".format( path=tmp_dir, reason=e ), error=True, ) self.host.quit(1) try: fd, path = tempfile.mkstemp( suffix=suff.encode("utf-8"), prefix=time.strftime(cat_dir_str + "_%Y-%m-%d_%H:%M:%S_"), dir=tmp_dir, text=True, ) return os.fdopen(fd, "w+b"), path except OSError as e: self.disp( u"Can't create temporary file: {reason}".format(reason=e), error=True ) self.host.quit(1) def getCurrentFile(self, profile): """Get most recently edited file @param profile(unicode): profile linked to the draft @return(str): full path of current file """ # we guess the item currently edited by choosing # the most recent file corresponding to temp file pattern # in tmp_dir, excluding metadata files cat_dir_str = self.cat_dir_str tmp_dir = getTmpDir(self.sat_conf, self.cat_dir_str, profile.encode("utf-8")) available = [ path for path in glob.glob(os.path.join(tmp_dir, cat_dir_str + "_*")) if not path.endswith(METADATA_SUFF) ] if not available: self.disp( u"Could not find any content draft in {path}".format(path=tmp_dir), error=True, ) self.host.quit(1) return max(available, key=lambda path: os.stat(path).st_mtime) def getItemData(self, service, node, item): """return formatted content, metadata (or not if use_metadata is false), and item id""" raise NotImplementedError def getTmpSuff(self): """return suffix used for content file""" return u"xml" def getItemPath(self): """retrieve item path (i.e. service and node) from item argument This method is obviously only useful for edition of PubSub based features """ service = self.args.service node = self.args.node item = self.args.item last_item = self.args.last_item if self.args.current: # user wants to continue current draft content_file_path = self.getCurrentFile(self.profile) self.disp(u"Continuing edition of current draft", 2) content_file_obj = open(content_file_path, "r+b") # we seek at the end of file in case of an item already exist # this will write content of the existing item at the end of the draft. # This way no data should be lost. content_file_obj.seek(0, os.SEEK_END) elif self.args.draft_path: # there is an existing draft that we use content_file_path = os.path.expanduser(self.args.item) content_file_obj = open(content_file_path, "r+b") # we seek at the end for the same reason as above content_file_obj.seek(0, os.SEEK_END) else: # we need a temporary file content_file_obj, content_file_path = self.getTmpFile() if item or last_item: self.disp(u"Editing requested published item", 2) try: if self.use_metadata: content, metadata, item = self.getItemData(service, node, item) else: content, item = self.getItemData(service, node, item) except Exception as e: # FIXME: ugly but we have not good may to check errors in bridge if u"item-not-found" in unicode(e): # item doesn't exist, we create a new one with requested id metadata = None if last_item: self.disp(_(u"no item found at all, we create a new one"), 2) else: self.disp( _( u'item "{item_id}" not found, we create a new item with this id' ).format(item_id=item), 2, ) content_file_obj.seek(0) else: self.disp(u"Error while retrieving item: {}".format(e)) self.host.quit(C.EXIT_ERROR) else: # item exists, we write content if content_file_obj.tell() != 0: # we already have a draft, # we copy item content after it and add an indicator content_file_obj.write("\n*****\n") content_file_obj.write(content.encode("utf-8")) content_file_obj.seek(0) self.disp( _(u'item "{item_id}" found, we edit it').format(item_id=item), 2 ) else: self.disp(u"Editing a new item", 2) if self.use_metadata: metadata = None if self.use_metadata: return service, node, item, content_file_path, content_file_obj, metadata else: return service, node, item, content_file_path, content_file_obj class Table(object): def __init__(self, host, data, headers=None, filters=None, use_buffer=False): """ @param data(iterable[list]): table data all lines must have the same number of columns @param headers(iterable[unicode], None): names/titles of the columns if not None, must have same number of columns as data @param filters(iterable[(callable, unicode)], None): values filters the callable will get 2 arguments: - current column value - RowData with all columns values if may also only use 1 argument, which will then be current col value. the callable must return a string if it's unicode, it will be used with .format and must countain u'{}' which will be replaced with the string if not None, must have same number of columns as data @param use_buffer(bool): if True, bufferise output instead of printing it directly """ self.host = host self._buffer = [] if use_buffer else None # headers are columns names/titles, can be None self.headers = headers # sizes fof columns without headers, # headers may be larger self.sizes = [] # rows countains one list per row with columns values self.rows = [] size = None if headers: row_cls = namedtuple("RowData", headers) else: row_cls = tuple for row_data in data: new_row = [] row_data_list = list(row_data) for idx, value in enumerate(row_data_list): if filters is not None and filters[idx] is not None: filter_ = filters[idx] if isinstance(filter_, basestring): col_value = filter_.format(value) else: try: col_value = filter_(value, row_cls(*row_data_list)) except TypeError: col_value = filter_(value) # we count size without ANSI code as they will change length of the string # when it's mostly style/color changes. col_size = len(regex.ansiRemove(col_value)) else: col_value = unicode(value) col_size = len(col_value) new_row.append(col_value) if size is None: self.sizes.append(col_size) else: self.sizes[idx] = max(self.sizes[idx], col_size) if size is None: size = len(new_row) if headers is not None and len(headers) != size: raise exceptions.DataError(u"headers size is not coherent with rows") else: if len(new_row) != size: raise exceptions.DataError(u"rows size is not coherent") self.rows.append(new_row) if not data and headers is not None: # the table is empty, we print headers at their lenght self.sizes = [len(h) for h in headers] @property def string(self): if self._buffer is None: raise exceptions.InternalError(u"buffer must be used to get a string") return u"\n".join(self._buffer) @staticmethod def readDictValues(data, keys, defaults=None): if defaults is None: defaults = {} for key in keys: try: yield data[key] except KeyError as e: default = defaults.get(key) if default is not None: yield default else: raise e @classmethod def fromDict(cls, host, data, keys=None, headers=None, filters=None, defaults=None): """Prepare a table to display it the whole data will be read and kept into memory, to be printed @param data(list[dict[unicode, unicode]]): data to create the table from @param keys(iterable[unicode], None): keys to get if None, all keys will be used @param headers(iterable[unicode], None): name of the columns names must be in same order as keys @param filters(dict[unicode, (callable,unicode)), None): filter to use on values keys correspond to keys to filter, and value is the same as for Table.__init__ @param defaults(dict[unicode, unicode]): default value to use if None, an exception will be raised if not value is found """ if keys is None and headers is not None: # FIXME: keys are not needed with OrderedDict, raise exceptions.DataError(u"You must specify keys order to used headers") if keys is None: keys = data[0].keys() if headers is None: headers = keys filters = [filters.get(k) for k in keys] return cls( host, (cls.readDictValues(d, keys, defaults) for d in data), headers, filters ) def _headers(self, head_sep, headers, sizes, alignment=u"left", style=None): """Render headers @param head_sep(unicode): sequence to use as separator @param alignment(unicode): how to align, can be left, center or right @param style(unicode, iterable[unicode], None): ANSI escape sequences to apply @param headers(list[unicode]): headers to show @param sizes(list[int]): sizes of columns """ rendered_headers = [] if isinstance(style, basestring): style = [style] for idx, header in enumerate(headers): size = sizes[idx] if alignment == u"left": rendered = header[:size].ljust(size) elif alignment == u"center": rendered = header[:size].center(size) elif alignment == u"right": rendered = header[:size].rjust(size) else: raise exceptions.InternalError(u"bad alignment argument") if style: args = style + [rendered] rendered = A.color(*args) rendered_headers.append(rendered) return head_sep.join(rendered_headers) def _disp(self, data): """output data (can be either bufferised or printed)""" if self._buffer is not None: self._buffer.append(data) else: self.host.disp(data) def display( self, head_alignment=u"left", columns_alignment=u"left", head_style=None, show_header=True, show_borders=True, hide_cols=None, col_sep=u" │ ", top_left=u"┌", top=u"─", top_sep=u"─┬─", top_right=u"┐", left=u"│", right=None, head_sep=None, head_line=u"┄", head_line_left=u"├", head_line_sep=u"┄┼┄", head_line_right=u"┤", bottom_left=u"└", bottom=None, bottom_sep=u"─┴─", bottom_right=u"┘", ): """Print the table @param show_header(bool): True if header need no be shown @param show_borders(bool): True if borders need no be shown @param hide_cols(None, iterable(unicode)): columns which should not be displayed @param head_alignment(unicode): how to align headers, can be left, center or right @param columns_alignment(unicode): how to align columns, can be left, center or right @param col_sep(unicode): separator betweens columns @param head_line(unicode): character to use to make line under head @param disp(callable, None): method to use to display the table None to use self.host.disp """ if not self.sizes: # the table is empty return col_sep_size = len(regex.ansiRemove(col_sep)) # if we have columns to hide, we remove them from headers and size if not hide_cols: headers = self.headers sizes = self.sizes else: headers = list(self.headers) sizes = self.sizes[:] ignore_idx = [headers.index(to_hide) for to_hide in hide_cols] for to_hide in hide_cols: hide_idx = headers.index(to_hide) del headers[hide_idx] del sizes[hide_idx] if right is None: right = left if top_sep is None: top_sep = col_sep_size * top if head_sep is None: head_sep = col_sep if bottom is None: bottom = top if bottom_sep is None: bottom_sep = col_sep_size * bottom if not show_borders: left = right = head_line_left = head_line_right = u"" # top border if show_borders: self._disp( top_left + top_sep.join([top * size for size in sizes]) + top_right ) # headers if show_header: self._disp( left + self._headers(head_sep, headers, sizes, head_alignment, head_style) + right ) # header line self._disp( head_line_left + head_line_sep.join([head_line * size for size in sizes]) + head_line_right ) # content if columns_alignment == u"left": alignment = lambda idx, s: ansi_ljust(s, sizes[idx]) elif columns_alignment == u"center": alignment = lambda idx, s: ansi_center(s, sizes[idx]) elif columns_alignment == u"right": alignment = lambda idx, s: ansi_rjust(s, sizes[idx]) else: raise exceptions.InternalError(u"bad columns alignment argument") for row in self.rows: if hide_cols: row = [v for idx, v in enumerate(row) if idx not in ignore_idx] self._disp( left + col_sep.join([alignment(idx, c) for idx, c in enumerate(row)]) + right ) if show_borders: # bottom border self._disp( bottom_left + bottom_sep.join([bottom * size for size in sizes]) + bottom_right ) # we return self so string can be used after display (table.display().string) return self def display_blank(self, **kwargs): """Display table without visible borders""" kwargs_ = {"col_sep": u" ", "head_line_sep": u" ", "show_borders": False} kwargs_.update(kwargs) return self.display(**kwargs_) class URIFinder(object): """Helper class to find URIs in well-known locations""" def __init__(self, command, path, key, callback, meta_map=None): """ @param command(CommandBase): command instance args of this instance will be updated with found values @param path(unicode): absolute path to use as a starting point to look for URIs @param key(unicode): key to look for @param callback(callable): method to call once URIs are found (or not) @param meta_map(dict, None): if not None, map metadata to arg name key is metadata used attribute name value is name to actually use, or None to ignore use empty dict to only retrieve URI possible keys are currently: - labels """ if not command.args.service and not command.args.node: self.host = command.host self.args = command.args self.key = key self.callback = callback self.meta_map = meta_map self.host.bridge.URIFind( path, [key], callback=self.URIFindCb, errback=partial( command.errback, msg=_(u"can't find " + key + u" URI: {}"), exit_code=C.EXIT_BRIDGE_ERRBACK, ), ) else: callback() def setMetadataList(self, uri_data, key): """Helper method to set list of values from metadata @param uri_data(dict): data of the found URI @param key(unicode): key of the value to retrieve """ new_values_json = uri_data.get(key) if uri_data is not None: if self.meta_map is None: dest = key else: dest = self.meta_map.get(key) if dest is None: return try: values = getattr(self.args, key) except AttributeError: raise exceptions.InternalError( u'there is no "{key}" arguments'.format(key=key) ) else: if values is None: values = [] values.extend(json.loads(new_values_json)) setattr(self.args, dest, values) def URIFindCb(self, uris_data): try: uri_data = uris_data[self.key] except KeyError: self.host.disp( _( u"No {key} URI specified for this project, please specify service and node" ).format(key=self.key), error=True, ) self.host.quit(C.EXIT_NOT_FOUND) else: uri = uri_data[u"uri"] self.setMetadataList(uri_data, u"labels") parsed_uri = xmpp_uri.parseXMPPUri(uri) try: self.args.service = parsed_uri[u"path"] self.args.node = parsed_uri[u"node"] except KeyError: self.host.disp(_(u"Invalid URI found: {uri}").format(uri=uri), error=True) self.host.quit(C.EXIT_DATA_ERROR) self.callback()