13
13
# You should have received a copy of the GNU General Public License
14
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
from __future__ import absolute_import
20
21
from bzrlib import (
24
from bzrlib.xml_serializer import SubElement, Element, Serializer
25
from bzrlib.inventory import ROOT_ID, Inventory, InventoryEntry
24
revision as _mod_revision,
27
from bzrlib.xml_serializer import (
34
serialize_inventory_flat,
35
unpack_inventory_entry,
36
unpack_inventory_flat,
26
38
from bzrlib.revision import Revision
27
39
from bzrlib.errors import BzrError
34
"'":"'", # FIXME: overkill
41
def _ensure_utf8_re():
42
"""Make sure the _utf8_re and _unicode_re regexes have been compiled."""
43
global _utf8_re, _unicode_re
45
_utf8_re = re.compile('[&<>\'\"]|[\x80-\xff]+')
46
if _unicode_re is None:
47
_unicode_re = re.compile(u'[&<>\'\"\u0080-\uffff]')
50
def _unicode_escape_replace(match, _map=_xml_escape_map):
51
"""Replace a string of non-ascii, non XML safe characters with their escape
53
This will escape both Standard XML escapes, like <>"', etc.
54
As well as escaping non ascii characters, because ElementTree did.
55
This helps us remain compatible to older versions of bzr. We may change
56
our policy in the future, though.
58
# jam 20060816 Benchmarks show that try/KeyError is faster if you
59
# expect the entity to rarely miss. There is about a 10% difference
60
# in overall time. But if you miss frequently, then if None is much
61
# faster. For our use case, we *rarely* have a revision id, file id
62
# or path name that is unicode. So use try/KeyError.
64
return _map[match.group()]
66
return "&#%d;" % ord(match.group())
69
def _utf8_escape_replace(match, _map=_xml_escape_map):
70
"""Escape utf8 characters into XML safe ones.
72
This uses 2 tricks. It is either escaping "standard" characters, like "&<>,
73
or it is handling characters with the high-bit set. For ascii characters,
74
we just lookup the replacement in the dictionary. For everything else, we
75
decode back into Unicode, and then use the XML escape code.
78
return _map[match.group()]
80
return ''.join('&#%d;' % ord(uni_chr)
81
for uni_chr in match.group().decode('utf8'))
86
def _encode_and_escape(unicode_or_utf8_str, _map=_to_escaped_map):
87
"""Encode the string into utf8, and escape invalid XML characters"""
88
# We frequently get entities we have not seen before, so it is better
89
# to check if None, rather than try/KeyError
90
text = _map.get(unicode_or_utf8_str)
92
if unicode_or_utf8_str.__class__ == unicode:
93
# The alternative policy is to do a regular UTF8 encoding
94
# and then escape only XML meta characters.
95
# Performance is equivalent once you use cache_utf8. *However*
96
# this makes the serialized texts incompatible with old versions
97
# of bzr. So no net gain. (Perhaps the read code would handle utf8
98
# better than entity escapes, but cElementTree seems to do just fine
100
text = str(_unicode_re.sub(_unicode_escape_replace,
101
unicode_or_utf8_str)) + '"'
103
# Plain strings are considered to already be in utf-8 so we do a
104
# slightly different method for escaping.
105
text = _utf8_re.sub(_utf8_escape_replace,
106
unicode_or_utf8_str) + '"'
107
_map[unicode_or_utf8_str] = text
111
def _get_utf8_or_ascii(a_str,
112
_encode_utf8=cache_utf8.encode,
113
_get_cached_ascii=cache_utf8.get_cached_ascii):
114
"""Return a cached version of the string.
116
cElementTree will return a plain string if the XML is plain ascii. It only
117
returns Unicode when it needs to. We want to work in utf-8 strings. So if
118
cElementTree returns a plain string, we can just return the cached version.
119
If it is Unicode, then we need to encode it.
121
:param a_str: An 8-bit string or Unicode as returned by
122
cElementTree.Element.get()
123
:return: A utf-8 encoded 8-bit string.
125
# This is fairly optimized because we know what cElementTree does, this is
126
# not meant as a generic function for all cases. Because it is possible for
127
# an 8-bit string to not be ascii or valid utf8.
128
if a_str.__class__ == unicode:
129
return _encode_utf8(a_str)
131
return _get_cached_ascii(a_str)
135
"""Clean out the unicode => escaped map"""
136
_to_escaped_map.clear()
139
class Serializer_v5(Serializer):
140
"""Version 5 serializer
142
Packs objects into XML and vice versa.
51
def _unescaper(match, _map=_xml_unescape_map):
56
if not code.startswith('#'):
58
return unichr(int(code[1:])).encode('utf8')
61
_unescape_re = lazy_regex.lazy_compile('\&([^;]*);')
63
def _unescape_xml(data):
64
"""Unescape predefined XML entities in a string of data."""
65
return _unescape_re.sub(_unescaper, data)
68
class Serializer_v8(XMLSerializer):
69
"""This serialiser adds rich roots.
71
Its revision format number matches its inventory number.
147
77
support_altered_by_hack = True
148
78
# This format supports the altered-by hack that reads file ids directly out
149
79
# of the versionedfile, without doing XML parsing.
151
def write_inventory_to_string(self, inv):
152
"""Just call write_inventory with a StringIO and return the value"""
81
supported_kinds = set(['file', 'directory', 'symlink'])
83
revision_format_num = None
85
# The search regex used by xml based repositories to determine what things
86
# where changed in a single commit.
87
_file_ids_altered_regex = lazy_regex.lazy_compile(
88
r'file_id="(?P<file_id>[^"]+)"'
89
r'.* revision="(?P<revision_id>[^"]+)"'
92
def _check_revisions(self, inv):
93
"""Extension point for subclasses to check during serialisation.
95
:param inv: An inventory about to be serialised, to be checked.
96
:raises: AssertionError if an error has occurred.
98
if inv.revision_id is None:
99
raise AssertionError("inv.revision_id is None")
100
if inv.root.revision is None:
101
raise AssertionError("inv.root.revision is None")
103
def _check_cache_size(self, inv_size, entry_cache):
104
"""Check that the entry_cache is large enough.
106
We want the cache to be ~2x the size of an inventory. The reason is
107
because we use a FIFO cache, and how Inventory records are likely to
108
change. In general, you have a small number of records which change
109
often, and a lot of records which do not change at all. So when the
110
cache gets full, you actually flush out a lot of the records you are
111
interested in, which means you need to recreate all of those records.
112
An LRU Cache would be better, but the overhead negates the cache
115
One way to look at it, only the size of the cache > len(inv) is your
116
'working' set. And in general, it shouldn't be a problem to hold 2
117
inventories in memory anyway.
119
:param inv_size: The number of entries in an inventory.
121
if entry_cache is None:
123
# 1.5 times might also be reasonable.
124
recommended_min_cache_size = inv_size * 1.5
125
if entry_cache.cache_size() < recommended_min_cache_size:
126
recommended_cache_size = inv_size * 2
127
trace.mutter('Resizing the inventory entry cache from %d to %d',
128
entry_cache.cache_size(), recommended_cache_size)
129
entry_cache.resize(recommended_cache_size)
131
def write_inventory_to_lines(self, inv):
132
"""Return a list of lines with the encoded inventory."""
133
return self.write_inventory(inv, None)
135
def write_inventory_to_string(self, inv, working=False):
136
"""Just call write_inventory with a StringIO and return the value.
138
:param working: If True skip history data - text_sha1, text_size,
139
reference_revision, symlink_target.
153
141
sio = cStringIO.StringIO()
154
self.write_inventory(inv, sio)
142
self.write_inventory(inv, sio, working)
155
143
return sio.getvalue()
157
def write_inventory(self, inv, f):
145
def write_inventory(self, inv, f, working=False):
158
146
"""Write inventory to a file.
160
148
:param inv: the inventory to write.
161
:param f: the file to write.
149
:param f: the file to write. (May be None if the lines are the desired
151
:param working: If True skip history data - text_sha1, text_size,
152
reference_revision, symlink_target.
153
:return: The inventory as a list of lines.
165
156
append = output.append
166
157
self._append_inventory_root(append, inv)
167
entries = inv.iter_entries()
169
root_path, root_ie = entries.next()
170
for path, ie in entries:
171
self._append_entry(append, ie)
172
append('</inventory>\n')
158
serialize_inventory_flat(inv, append,
159
self.root_id, self.supported_kinds, working)
174
162
# Just to keep the cache from growing without bounds
175
163
# but we may actually not want to do clear the cache
178
167
def _append_inventory_root(self, append, inv):
179
168
"""Append the inventory root to output."""
181
if inv.root.file_id not in (None, ROOT_ID):
183
append(_encode_and_escape(inv.root.file_id))
184
append(' format="5"')
185
169
if inv.revision_id is not None:
186
append(' revision_id="')
187
append(_encode_and_escape(inv.revision_id))
190
def _append_entry(self, append, ie):
191
"""Convert InventoryEntry to XML element and append to output."""
192
# TODO: should just be a plain assertion
193
assert InventoryEntry.versionable_kind(ie.kind), \
194
'unsupported entry kind %s' % ie.kind
199
append(' executable="yes"')
201
append(_encode_and_escape(ie.file_id))
203
append(_encode_and_escape(ie.name))
204
if self._parent_condition(ie):
205
assert isinstance(ie.parent_id, basestring)
206
append(' parent_id="')
207
append(_encode_and_escape(ie.parent_id))
208
if ie.revision is not None:
209
append(' revision="')
210
append(_encode_and_escape(ie.revision))
211
if ie.symlink_target is not None:
212
append(' symlink_target="')
213
append(_encode_and_escape(ie.symlink_target))
214
if ie.text_sha1 is not None:
215
append(' text_sha1="')
218
if ie.text_size is not None:
219
append(' text_size="%d"' % ie.text_size)
223
def _parent_condition(self, ie):
224
return ie.parent_id != ROOT_ID
170
revid1 = ' revision_id="'
171
revid2 = encode_and_escape(inv.revision_id)
175
append('<inventory format="%s"%s%s>\n' % (
176
self.format_num, revid1, revid2))
177
append('<directory file_id="%s name="%s revision="%s />\n' % (
178
encode_and_escape(inv.root.file_id),
179
encode_and_escape(inv.root.name),
180
encode_and_escape(inv.root.revision)))
226
182
def _pack_revision(self, rev):
227
183
"""Revision object -> xml tree"""
262
221
def _pack_revision_properties(self, rev, under_element):
263
222
top_elt = SubElement(under_element, 'properties')
264
223
for prop_name, prop_value in sorted(rev.properties.items()):
265
assert isinstance(prop_name, basestring)
266
assert isinstance(prop_value, basestring)
267
224
prop_elt = SubElement(top_elt, 'property')
268
225
prop_elt.set('name', prop_name)
269
226
prop_elt.text = prop_value
270
227
prop_elt.tail = '\n'
271
228
top_elt.tail = '\n'
273
def _unpack_inventory(self, elt):
274
"""Construct from XML Element
276
assert elt.tag == 'inventory'
277
root_id = elt.get('file_id') or ROOT_ID
278
root_id = _get_utf8_or_ascii(root_id)
230
def _unpack_entry(self, elt, entry_cache=None, return_from_cache=False):
231
# This is here because it's overridden by xml7
232
return unpack_inventory_entry(elt, entry_cache,
280
format = elt.get('format')
281
if format is not None:
283
raise BzrError("invalid format version %r on inventory"
285
revision_id = elt.get('revision_id')
286
if revision_id is not None:
287
revision_id = cache_utf8.encode(revision_id)
288
inv = Inventory(root_id, revision_id=revision_id)
290
ie = self._unpack_entry(e)
291
if ie.parent_id is None:
292
ie.parent_id = root_id
235
def _unpack_inventory(self, elt, revision_id=None, entry_cache=None,
236
return_from_cache=False):
237
"""Construct from XML Element"""
238
inv = unpack_inventory_flat(elt, self.format_num, self._unpack_entry,
239
entry_cache, return_from_cache)
240
self._check_cache_size(len(inv), entry_cache)
296
def _unpack_entry(self, elt):
298
if not InventoryEntry.versionable_kind(kind):
299
raise AssertionError('unsupported entry kind %s' % kind)
301
get_cached = _get_utf8_or_ascii
303
parent_id = elt.get('parent_id')
304
if parent_id is not None:
305
parent_id = get_cached(parent_id)
306
file_id = get_cached(elt.get('file_id'))
308
if kind == 'directory':
309
ie = inventory.InventoryDirectory(file_id,
313
ie = inventory.InventoryFile(file_id,
316
ie.text_sha1 = elt.get('text_sha1')
317
if elt.get('executable') == 'yes':
319
v = elt.get('text_size')
320
ie.text_size = v and int(v)
321
elif kind == 'symlink':
322
ie = inventory.InventoryLink(file_id,
325
ie.symlink_target = elt.get('symlink_target')
327
raise BzrError("unknown kind %r" % kind)
328
revision = elt.get('revision')
329
if revision is not None:
330
revision = get_cached(revision)
331
ie.revision = revision
335
243
def _unpack_revision(self, elt):
336
244
"""XML Element -> Revision object"""
337
assert elt.tag == 'revision'
338
245
format = elt.get('format')
246
format_num = self.format_num
247
if self.revision_format_num is not None:
248
format_num = self.revision_format_num
339
249
if format is not None:
341
raise BzrError("invalid format version %r on inventory"
250
if format != format_num:
251
raise BzrError("invalid format version %r on revision"
343
get_cached = _get_utf8_or_ascii
253
get_cached = get_utf8_or_ascii
344
254
rev = Revision(committer = elt.get('committer'),
345
255
timestamp = float(elt.get('timestamp')),
346
256
revision_id = get_cached(elt.get('revision_id')),
376
284
# properties have string values
377
285
if value is None:
379
assert name not in rev.properties, \
380
"repeated property %r" % name
287
if name in rev.properties:
288
raise AssertionError("repeated property %r" % name)
381
289
rev.properties[name] = value
384
serializer_v5 = Serializer_v5()
291
def _find_text_key_references(self, line_iterator):
292
"""Core routine for extracting references to texts from inventories.
294
This performs the translation of xml lines to revision ids.
296
:param line_iterator: An iterator of lines, origin_version_id
297
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
298
to whether they were referred to by the inventory of the
299
revision_id that they contain. Note that if that revision_id was
300
not part of the line_iterator's output then False will be given -
301
even though it may actually refer to that key.
303
if not self.support_altered_by_hack:
304
raise AssertionError(
305
"_find_text_key_references only "
306
"supported for branches which store inventory as unnested xml"
307
", not on %r" % self)
310
# this code needs to read every new line in every inventory for the
311
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
312
# not present in one of those inventories is unnecessary but not
313
# harmful because we are filtering by the revision id marker in the
314
# inventory lines : we only select file ids altered in one of those
315
# revisions. We don't need to see all lines in the inventory because
316
# only those added in an inventory in rev X can contain a revision=X
318
unescape_revid_cache = {}
319
unescape_fileid_cache = {}
321
# jam 20061218 In a big fetch, this handles hundreds of thousands
322
# of lines, so it has had a lot of inlining and optimizing done.
323
# Sorry that it is a little bit messy.
324
# Move several functions to be local variables, since this is a long
326
search = self._file_ids_altered_regex.search
327
unescape = _unescape_xml
328
setdefault = result.setdefault
329
for line, line_key in line_iterator:
333
# One call to match.group() returning multiple items is quite a
334
# bit faster than 2 calls to match.group() each returning 1
335
file_id, revision_id = match.group('file_id', 'revision_id')
337
# Inlining the cache lookups helps a lot when you make 170,000
338
# lines and 350k ids, versus 8.4 unique ids.
339
# Using a cache helps in 2 ways:
340
# 1) Avoids unnecessary decoding calls
341
# 2) Re-uses cached strings, which helps in future set and
343
# (2) is enough that removing encoding entirely along with
344
# the cache (so we are using plain strings) results in no
345
# performance improvement.
347
revision_id = unescape_revid_cache[revision_id]
349
unescaped = unescape(revision_id)
350
unescape_revid_cache[revision_id] = unescaped
351
revision_id = unescaped
353
# Note that unconditionally unescaping means that we deserialise
354
# every fileid, which for general 'pull' is not great, but we don't
355
# really want to have some many fulltexts that this matters anyway.
358
file_id = unescape_fileid_cache[file_id]
360
unescaped = unescape(file_id)
361
unescape_fileid_cache[file_id] = unescaped
364
key = (file_id, revision_id)
365
setdefault(key, False)
366
if revision_id == line_key[-1]:
371
serializer_v8 = Serializer_v8()
added
removed
bzrlib/xml8.py
