# Copyright (C) 2005-2010 Canonical Ltd # # This program is free software; you can redistribute it and/or modify # it under the terms of the GNU General Public License as published by # the Free Software Foundation; either version 2 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 General Public License for more details. # # You should have received a copy of the GNU General Public License # along with this program; if not, write to the Free Software # Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA import cStringIO import re from bzrlib import ( cache_utf8, errors, inventory, lazy_regex, revision as _mod_revision, trace, ) from bzrlib.xml_serializer import ( Element, SubElement, XMLSerializer, escape_invalid_chars, ) from bzrlib.inventory import InventoryEntry from bzrlib.revision import Revision from bzrlib.errors import BzrError _utf8_re = None _unicode_re = None _xml_escape_map = { "&":'&', "'":"'", # FIXME: overkill "\"":""", "<":"<", ">":">", } _xml_unescape_map = { 'apos':"'", 'quot':'"', 'amp':'&', 'lt':'<', 'gt':'>' } def _unescaper(match, _map=_xml_unescape_map): code = match.group(1) try: return _map[code] except KeyError: if not code.startswith('#'): raise return unichr(int(code[1:])).encode('utf8') _unescape_re = None def _unescape_xml(data): """Unescape predefined XML entities in a string of data.""" global _unescape_re if _unescape_re is None: _unescape_re = re.compile('\&([^;]*);') return _unescape_re.sub(_unescaper, data) def _ensure_utf8_re(): """Make sure the _utf8_re and _unicode_re regexes have been compiled.""" global _utf8_re, _unicode_re if _utf8_re is None: _utf8_re = re.compile('[&<>\'\"]|[\x80-\xff]+') if _unicode_re is None: _unicode_re = re.compile(u'[&<>\'\"\u0080-\uffff]') def _unicode_escape_replace(match, _map=_xml_escape_map): """Replace a string of non-ascii, non XML safe characters with their escape This will escape both Standard XML escapes, like <>"', etc. As well as escaping non ascii characters, because ElementTree did. This helps us remain compatible to older versions of bzr. We may change our policy in the future, though. """ # jam 20060816 Benchmarks show that try/KeyError is faster if you # expect the entity to rarely miss. There is about a 10% difference # in overall time. But if you miss frequently, then if None is much # faster. For our use case, we *rarely* have a revision id, file id # or path name that is unicode. So use try/KeyError. try: return _map[match.group()] except KeyError: return "&#%d;" % ord(match.group()) def _utf8_escape_replace(match, _map=_xml_escape_map): """Escape utf8 characters into XML safe ones. This uses 2 tricks. It is either escaping "standard" characters, like "&<>, or it is handling characters with the high-bit set. For ascii characters, we just lookup the replacement in the dictionary. For everything else, we decode back into Unicode, and then use the XML escape code. """ try: return _map[match.group()] except KeyError: return ''.join('&#%d;' % ord(uni_chr) for uni_chr in match.group().decode('utf8')) _to_escaped_map = {} def _encode_and_escape(unicode_or_utf8_str, _map=_to_escaped_map): """Encode the string into utf8, and escape invalid XML characters""" # We frequently get entities we have not seen before, so it is better # to check if None, rather than try/KeyError text = _map.get(unicode_or_utf8_str) if text is None: if unicode_or_utf8_str.__class__ is unicode: # The alternative policy is to do a regular UTF8 encoding # and then escape only XML meta characters. # Performance is equivalent once you use cache_utf8. *However* # this makes the serialized texts incompatible with old versions # of bzr. So no net gain. (Perhaps the read code would handle utf8 # better than entity escapes, but cElementTree seems to do just fine # either way) text = str(_unicode_re.sub(_unicode_escape_replace, unicode_or_utf8_str)) + '"' else: # Plain strings are considered to already be in utf-8 so we do a # slightly different method for escaping. text = _utf8_re.sub(_utf8_escape_replace, unicode_or_utf8_str) + '"' _map[unicode_or_utf8_str] = text return text def _get_utf8_or_ascii(a_str, _encode_utf8=cache_utf8.encode, _get_cached_ascii=cache_utf8.get_cached_ascii): """Return a cached version of the string. cElementTree will return a plain string if the XML is plain ascii. It only returns Unicode when it needs to. We want to work in utf-8 strings. So if cElementTree returns a plain string, we can just return the cached version. If it is Unicode, then we need to encode it. :param a_str: An 8-bit string or Unicode as returned by cElementTree.Element.get() :return: A utf-8 encoded 8-bit string. """ # This is fairly optimized because we know what cElementTree does, this is # not meant as a generic function for all cases. Because it is possible for # an 8-bit string to not be ascii or valid utf8. if a_str.__class__ is unicode: return _encode_utf8(a_str) else: return intern(a_str) def _clear_cache(): """Clean out the unicode => escaped map""" _to_escaped_map.clear() class Serializer_v8(XMLSerializer): """This serialiser adds rich roots. Its revision format number matches its inventory number. """ __slots__ = [] root_id = None support_altered_by_hack = True # This format supports the altered-by hack that reads file ids directly out # of the versionedfile, without doing XML parsing. supported_kinds = set(['file', 'directory', 'symlink']) format_num = '8' revision_format_num = None # The search regex used by xml based repositories to determine what things # where changed in a single commit. _file_ids_altered_regex = lazy_regex.lazy_compile( r'file_id="(?P[^"]+)"' r'.* revision="(?P[^"]+)"' ) def _check_revisions(self, inv): """Extension point for subclasses to check during serialisation. :param inv: An inventory about to be serialised, to be checked. :raises: AssertionError if an error has occurred. """ if inv.revision_id is None: raise AssertionError("inv.revision_id is None") if inv.root.revision is None: raise AssertionError("inv.root.revision is None") def _check_cache_size(self, inv_size, entry_cache): """Check that the entry_cache is large enough. We want the cache to be ~2x the size of an inventory. The reason is because we use a FIFO cache, and how Inventory records are likely to change. In general, you have a small number of records which change often, and a lot of records which do not change at all. So when the cache gets full, you actually flush out a lot of the records you are interested in, which means you need to recreate all of those records. An LRU Cache would be better, but the overhead negates the cache coherency benefit. One way to look at it, only the size of the cache > len(inv) is your 'working' set. And in general, it shouldn't be a problem to hold 2 inventories in memory anyway. :param inv_size: The number of entries in an inventory. """ if entry_cache is None: return # 1.5 times might also be reasonable. recommended_min_cache_size = inv_size * 1.5 if entry_cache.cache_size() < recommended_min_cache_size: recommended_cache_size = inv_size * 2 trace.mutter('Resizing the inventory entry cache from %d to %d', entry_cache.cache_size(), recommended_cache_size) entry_cache.resize(recommended_cache_size) def write_inventory_to_lines(self, inv): """Return a list of lines with the encoded inventory.""" return self.write_inventory(inv, None) def write_inventory_to_string(self, inv, working=False): """Just call write_inventory with a StringIO and return the value. :param working: If True skip history data - text_sha1, text_size, reference_revision, symlink_target. """ sio = cStringIO.StringIO() self.write_inventory(inv, sio, working) return sio.getvalue() def write_inventory(self, inv, f, working=False): """Write inventory to a file. :param inv: the inventory to write. :param f: the file to write. (May be None if the lines are the desired output). :param working: If True skip history data - text_sha1, text_size, reference_revision, symlink_target. :return: The inventory as a list of lines. """ _ensure_utf8_re() self._check_revisions(inv) output = [] append = output.append self._append_inventory_root(append, inv) entries = inv.iter_entries() # Skip the root root_path, root_ie = entries.next() for path, ie in entries: if ie.parent_id != self.root_id: parent_str = ' parent_id="' parent_id = _encode_and_escape(ie.parent_id) else: parent_str = '' parent_id = '' if ie.kind == 'file': if ie.executable: executable = ' executable="yes"' else: executable = '' if not working: append('\n' % ( executable, _encode_and_escape(ie.file_id), _encode_and_escape(ie.name), parent_str, parent_id, _encode_and_escape(ie.revision), ie.text_sha1, ie.text_size)) else: append('\n' % ( executable, _encode_and_escape(ie.file_id), _encode_and_escape(ie.name), parent_str, parent_id)) elif ie.kind == 'directory': if not working: append('\n' % ( self.format_num, revid1, revid2)) append(' xml tree""" # For the XML format, we need to write them as Unicode rather than as # utf-8 strings. So that cElementTree can handle properly escaping # them. decode_utf8 = cache_utf8.decode revision_id = rev.revision_id if isinstance(revision_id, str): revision_id = decode_utf8(revision_id) format_num = self.format_num if self.revision_format_num is not None: format_num = self.revision_format_num root = Element('revision', committer = rev.committer, timestamp = '%.3f' % rev.timestamp, revision_id = revision_id, inventory_sha1 = rev.inventory_sha1, format=format_num, ) if rev.timezone is not None: root.set('timezone', str(rev.timezone)) root.text = '\n' msg = SubElement(root, 'message') msg.text = escape_invalid_chars(rev.message)[0] msg.tail = '\n' if rev.parent_ids: pelts = SubElement(root, 'parents') pelts.tail = pelts.text = '\n' for parent_id in rev.parent_ids: _mod_revision.check_not_reserved_id(parent_id) p = SubElement(pelts, 'revision_ref') p.tail = '\n' if isinstance(parent_id, str): parent_id = decode_utf8(parent_id) p.set('revision_id', parent_id) if rev.properties: self._pack_revision_properties(rev, root) return root def _pack_revision_properties(self, rev, under_element): top_elt = SubElement(under_element, 'properties') for prop_name, prop_value in sorted(rev.properties.items()): prop_elt = SubElement(top_elt, 'property') prop_elt.set('name', prop_name) prop_elt.text = prop_value prop_elt.tail = '\n' top_elt.tail = '\n' def _unpack_inventory(self, elt, revision_id=None, entry_cache=None, return_from_cache=False): """Construct from XML Element""" if elt.tag != 'inventory': raise errors.UnexpectedInventoryFormat('Root tag is %r' % elt.tag) format = elt.get('format') if format != self.format_num: raise errors.UnexpectedInventoryFormat('Invalid format version %r' % format) revision_id = elt.get('revision_id') if revision_id is not None: revision_id = cache_utf8.encode(revision_id) inv = inventory.Inventory(root_id=None, revision_id=revision_id) for e in elt: ie = self._unpack_entry(e, entry_cache=entry_cache, return_from_cache=return_from_cache) inv.add(ie) self._check_cache_size(len(inv), entry_cache) return inv def _unpack_entry(self, elt, entry_cache=None, return_from_cache=False): elt_get = elt.get file_id = elt_get('file_id') revision = elt_get('revision') # Check and see if we have already unpacked this exact entry # Some timings for "repo.revision_trees(last_100_revs)" # bzr mysql # unmodified 4.1s 40.8s # using lru 3.5s # using fifo 2.83s 29.1s # lru._cache 2.8s # dict 2.75s 26.8s # inv.add 2.5s 26.0s # no_copy 2.00s 20.5s # no_c,dict 1.95s 18.0s # Note that a cache of 10k nodes is more than sufficient to hold all of # the inventory for the last 100 revs for bzr, but not for mysql (20k # is enough for mysql, which saves the same 2s as using a dict) # Breakdown of mysql using time.clock() # 4.1s 2 calls to element.get for file_id, revision_id # 4.5s cache_hit lookup # 7.1s InventoryFile.copy() # 2.4s InventoryDirectory.copy() # 0.4s decoding unique entries # 1.6s decoding entries after FIFO fills up # 0.8s Adding nodes to FIFO (including flushes) # 0.1s cache miss lookups # Using an LRU cache # 4.1s 2 calls to element.get for file_id, revision_id # 9.9s cache_hit lookup # 10.8s InventoryEntry.copy() # 0.3s cache miss lookus # 1.2s decoding entries # 1.0s adding nodes to LRU if entry_cache is not None and revision is not None: key = (file_id, revision) try: # We copy it, because some operations may mutate it cached_ie = entry_cache[key] except KeyError: pass else: # Only copying directory entries drops us 2.85s => 2.35s if return_from_cache: if cached_ie.kind == 'directory': return cached_ie.copy() return cached_ie return cached_ie.copy() kind = elt.tag if not InventoryEntry.versionable_kind(kind): raise AssertionError('unsupported entry kind %s' % kind) get_cached = _get_utf8_or_ascii file_id = get_cached(file_id) if revision is not None: revision = get_cached(revision) parent_id = elt_get('parent_id') if parent_id is not None: parent_id = get_cached(parent_id) if kind == 'directory': ie = inventory.InventoryDirectory(file_id, elt_get('name'), parent_id) elif kind == 'file': ie = inventory.InventoryFile(file_id, elt_get('name'), parent_id) ie.text_sha1 = elt_get('text_sha1') if elt_get('executable') == 'yes': ie.executable = True v = elt_get('text_size') ie.text_size = v and int(v) elif kind == 'symlink': ie = inventory.InventoryLink(file_id, elt_get('name'), parent_id) ie.symlink_target = elt_get('symlink_target') else: raise errors.UnsupportedInventoryKind(kind) ie.revision = revision if revision is not None and entry_cache is not None: # We cache a copy() because callers like to mutate objects, and # that would cause the item in cache to mutate as well. # This has a small effect on many-inventory performance, because # the majority fraction is spent in cache hits, not misses. entry_cache[key] = ie.copy() return ie def _unpack_revision(self, elt): """XML Element -> Revision object""" format = elt.get('format') format_num = self.format_num if self.revision_format_num is not None: format_num = self.revision_format_num if format is not None: if format != format_num: raise BzrError("invalid format version %r on revision" % format) get_cached = _get_utf8_or_ascii rev = Revision(committer = elt.get('committer'), timestamp = float(elt.get('timestamp')), revision_id = get_cached(elt.get('revision_id')), inventory_sha1 = elt.get('inventory_sha1') ) parents = elt.find('parents') or [] for p in parents: rev.parent_ids.append(get_cached(p.get('revision_id'))) self._unpack_revision_properties(elt, rev) v = elt.get('timezone') if v is None: rev.timezone = 0 else: rev.timezone = int(v) rev.message = elt.findtext('message') # text of return rev def _unpack_revision_properties(self, elt, rev): """Unpack properties onto a revision.""" props_elt = elt.find('properties') if not props_elt: return for prop_elt in props_elt: if prop_elt.tag != 'property': raise AssertionError( "bad tag under properties list: %r" % prop_elt.tag) name = prop_elt.get('name') value = prop_elt.text # If a property had an empty value ('') cElementTree reads # that back as None, convert it back to '', so that all # properties have string values if value is None: value = '' if name in rev.properties: raise AssertionError("repeated property %r" % name) rev.properties[name] = value def _find_text_key_references(self, line_iterator): """Core routine for extracting references to texts from inventories. This performs the translation of xml lines to revision ids. :param line_iterator: An iterator of lines, origin_version_id :return: A dictionary mapping text keys ((fileid, revision_id) tuples) to whether they were referred to by the inventory of the revision_id that they contain. Note that if that revision_id was not part of the line_iterator's output then False will be given - even though it may actually refer to that key. """ if not self.support_altered_by_hack: raise AssertionError( "_find_text_key_references only " "supported for branches which store inventory as unnested xml" ", not on %r" % self) result = {} # this code needs to read every new line in every inventory for the # inventories [revision_ids]. Seeing a line twice is ok. Seeing a line # not present in one of those inventories is unnecessary but not # harmful because we are filtering by the revision id marker in the # inventory lines : we only select file ids altered in one of those # revisions. We don't need to see all lines in the inventory because # only those added in an inventory in rev X can contain a revision=X # line. unescape_revid_cache = {} unescape_fileid_cache = {} # jam 20061218 In a big fetch, this handles hundreds of thousands # of lines, so it has had a lot of inlining and optimizing done. # Sorry that it is a little bit messy. # Move several functions to be local variables, since this is a long # running loop. search = self._file_ids_altered_regex.search unescape = _unescape_xml setdefault = result.setdefault for line, line_key in line_iterator: match = search(line) if match is None: continue # One call to match.group() returning multiple items is quite a # bit faster than 2 calls to match.group() each returning 1 file_id, revision_id = match.group('file_id', 'revision_id') # Inlining the cache lookups helps a lot when you make 170,000 # lines and 350k ids, versus 8.4 unique ids. # Using a cache helps in 2 ways: # 1) Avoids unnecessary decoding calls # 2) Re-uses cached strings, which helps in future set and # equality checks. # (2) is enough that removing encoding entirely along with # the cache (so we are using plain strings) results in no # performance improvement. try: revision_id = unescape_revid_cache[revision_id] except KeyError: unescaped = unescape(revision_id) unescape_revid_cache[revision_id] = unescaped revision_id = unescaped # Note that unconditionally unescaping means that we deserialise # every fileid, which for general 'pull' is not great, but we don't # really want to have some many fulltexts that this matters anyway. # RBC 20071114. try: file_id = unescape_fileid_cache[file_id] except KeyError: unescaped = unescape(file_id) unescape_fileid_cache[file_id] = unescaped file_id = unescaped key = (file_id, revision_id) setdefault(key, False) if revision_id == line_key[-1]: result[key] = True return result serializer_v8 = Serializer_v8()