14
14
# along with this program; if not, write to the Free Software
15
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
from io import BytesIO
22
24
revision as _mod_revision,
25
from .xml_serializer import (
27
from bzrlib.xml_serializer import (
30
31
escape_invalid_chars,
32
serialize_inventory_flat,
33
unpack_inventory_entry,
34
unpack_inventory_flat,
36
from ..revision import Revision
37
from ..errors import BzrError
49
def _unescaper(match, _map=_xml_unescape_map):
54
if not code.startswith(b'#'):
56
return chr(int(code[1:])).encode('utf8')
59
_unescape_re = lazy_regex.lazy_compile(b'\\&([^;]*);')
62
def _unescape_xml(data):
63
"""Unescape predefined XML entities in a string of data."""
64
return _unescape_re.sub(_unescaper, data)
33
from bzrlib.inventory import ROOT_ID, Inventory, InventoryEntry
34
from bzrlib.revision import Revision
35
from bzrlib.errors import BzrError
42
"'":"'", # FIXME: overkill
49
def _ensure_utf8_re():
50
"""Make sure the _utf8_re and _unicode_re regexes have been compiled."""
51
global _utf8_re, _unicode_re
53
_utf8_re = re.compile('[&<>\'\"]|[\x80-\xff]+')
54
if _unicode_re is None:
55
_unicode_re = re.compile(u'[&<>\'\"\u0080-\uffff]')
58
def _unicode_escape_replace(match, _map=_xml_escape_map):
59
"""Replace a string of non-ascii, non XML safe characters with their escape
61
This will escape both Standard XML escapes, like <>"', etc.
62
As well as escaping non ascii characters, because ElementTree did.
63
This helps us remain compatible to older versions of bzr. We may change
64
our policy in the future, though.
66
# jam 20060816 Benchmarks show that try/KeyError is faster if you
67
# expect the entity to rarely miss. There is about a 10% difference
68
# in overall time. But if you miss frequently, then if None is much
69
# faster. For our use case, we *rarely* have a revision id, file id
70
# or path name that is unicode. So use try/KeyError.
72
return _map[match.group()]
74
return "&#%d;" % ord(match.group())
77
def _utf8_escape_replace(match, _map=_xml_escape_map):
78
"""Escape utf8 characters into XML safe ones.
80
This uses 2 tricks. It is either escaping "standard" characters, like "&<>,
81
or it is handling characters with the high-bit set. For ascii characters,
82
we just lookup the replacement in the dictionary. For everything else, we
83
decode back into Unicode, and then use the XML escape code.
86
return _map[match.group()]
88
return ''.join('&#%d;' % ord(uni_chr)
89
for uni_chr in match.group().decode('utf8'))
94
def _encode_and_escape(unicode_or_utf8_str, _map=_to_escaped_map):
95
"""Encode the string into utf8, and escape invalid XML characters"""
96
# We frequently get entities we have not seen before, so it is better
97
# to check if None, rather than try/KeyError
98
text = _map.get(unicode_or_utf8_str)
100
if unicode_or_utf8_str.__class__ is unicode:
101
# The alternative policy is to do a regular UTF8 encoding
102
# and then escape only XML meta characters.
103
# Performance is equivalent once you use cache_utf8. *However*
104
# this makes the serialized texts incompatible with old versions
105
# of bzr. So no net gain. (Perhaps the read code would handle utf8
106
# better than entity escapes, but cElementTree seems to do just fine
108
text = str(_unicode_re.sub(_unicode_escape_replace,
109
unicode_or_utf8_str)) + '"'
111
# Plain strings are considered to already be in utf-8 so we do a
112
# slightly different method for escaping.
113
text = _utf8_re.sub(_utf8_escape_replace,
114
unicode_or_utf8_str) + '"'
115
_map[unicode_or_utf8_str] = text
119
def _get_utf8_or_ascii(a_str,
120
_encode_utf8=cache_utf8.encode,
121
_get_cached_ascii=cache_utf8.get_cached_ascii):
122
"""Return a cached version of the string.
124
cElementTree will return a plain string if the XML is plain ascii. It only
125
returns Unicode when it needs to. We want to work in utf-8 strings. So if
126
cElementTree returns a plain string, we can just return the cached version.
127
If it is Unicode, then we need to encode it.
129
:param a_str: An 8-bit string or Unicode as returned by
130
cElementTree.Element.get()
131
:return: A utf-8 encoded 8-bit string.
133
# This is fairly optimized because we know what cElementTree does, this is
134
# not meant as a generic function for all cases. Because it is possible for
135
# an 8-bit string to not be ascii or valid utf8.
136
if a_str.__class__ is unicode:
137
return _encode_utf8(a_str)
143
"""Clean out the unicode => escaped map"""
144
_to_escaped_map.clear()
67
147
class Serializer_v8(XMLSerializer):
144
224
reference_revision, symlink_target.
145
225
:return: The inventory as a list of lines.
228
self._check_revisions(inv)
148
230
append = output.append
149
231
self._append_inventory_root(append, inv)
150
serialize_inventory_flat(inv, append,
151
self.root_id, self.supported_kinds, working)
232
entries = inv.iter_entries()
234
root_path, root_ie = entries.next()
235
for path, ie in entries:
236
if ie.parent_id != self.root_id:
237
parent_str = ' parent_id="'
238
parent_id = _encode_and_escape(ie.parent_id)
242
if ie.kind == 'file':
244
executable = ' executable="yes"'
248
append('<file%s file_id="%s name="%s%s%s revision="%s '
249
'text_sha1="%s" text_size="%d" />\n' % (
250
executable, _encode_and_escape(ie.file_id),
251
_encode_and_escape(ie.name), parent_str, parent_id,
252
_encode_and_escape(ie.revision), ie.text_sha1,
255
append('<file%s file_id="%s name="%s%s%s />\n' % (
256
executable, _encode_and_escape(ie.file_id),
257
_encode_and_escape(ie.name), parent_str, parent_id))
258
elif ie.kind == 'directory':
260
append('<directory file_id="%s name="%s%s%s revision="%s '
262
_encode_and_escape(ie.file_id),
263
_encode_and_escape(ie.name),
264
parent_str, parent_id,
265
_encode_and_escape(ie.revision)))
267
append('<directory file_id="%s name="%s%s%s />\n' % (
268
_encode_and_escape(ie.file_id),
269
_encode_and_escape(ie.name),
270
parent_str, parent_id))
271
elif ie.kind == 'symlink':
273
append('<symlink file_id="%s name="%s%s%s revision="%s '
274
'symlink_target="%s />\n' % (
275
_encode_and_escape(ie.file_id),
276
_encode_and_escape(ie.name),
277
parent_str, parent_id,
278
_encode_and_escape(ie.revision),
279
_encode_and_escape(ie.symlink_target)))
281
append('<symlink file_id="%s name="%s%s%s />\n' % (
282
_encode_and_escape(ie.file_id),
283
_encode_and_escape(ie.name),
284
parent_str, parent_id))
285
elif ie.kind == 'tree-reference':
286
if ie.kind not in self.supported_kinds:
287
raise errors.UnsupportedInventoryKind(ie.kind)
289
append('<tree-reference file_id="%s name="%s%s%s '
290
'revision="%s reference_revision="%s />\n' % (
291
_encode_and_escape(ie.file_id),
292
_encode_and_escape(ie.name),
293
parent_str, parent_id,
294
_encode_and_escape(ie.revision),
295
_encode_and_escape(ie.reference_revision)))
297
append('<tree-reference file_id="%s name="%s%s%s />\n' % (
298
_encode_and_escape(ie.file_id),
299
_encode_and_escape(ie.name),
300
parent_str, parent_id))
302
raise errors.UnsupportedInventoryKind(ie.kind)
303
append('</inventory>\n')
152
304
if f is not None:
153
305
f.writelines(output)
154
306
# Just to keep the cache from growing without bounds
155
307
# but we may actually not want to do clear the cache
159
311
def _append_inventory_root(self, append, inv):
160
312
"""Append the inventory root to output."""
161
313
if inv.revision_id is not None:
163
[b' revision_id="', encode_and_escape(inv.revision_id), b'"'])
314
revid1 = ' revision_id="'
315
revid2 = _encode_and_escape(inv.revision_id)
166
append(b'<inventory format="%s"%s>\n' % (
167
self.format_num, revid1))
168
append(b'<directory file_id="%s" name="%s" revision="%s" />\n' % (
169
encode_and_escape(inv.root.file_id),
170
encode_and_escape(inv.root.name),
171
encode_and_escape(inv.root.revision)))
319
append('<inventory format="%s"%s%s>\n' % (
320
self.format_num, revid1, revid2))
321
append('<directory file_id="%s name="%s revision="%s />\n' % (
322
_encode_and_escape(inv.root.file_id),
323
_encode_and_escape(inv.root.name),
324
_encode_and_escape(inv.root.revision)))
173
def write_revision_to_lines(self, rev):
326
def _pack_revision(self, rev):
174
327
"""Revision object -> xml tree"""
175
328
# For the XML format, we need to write them as Unicode rather than as
176
329
# utf-8 strings. So that cElementTree can handle properly escaping
179
el = (b'<revision committer="%s" format="%s" '
180
b'inventory_sha1="%s" revision_id="%s" '
181
b'timestamp="%.3f"' % (
182
encode_and_escape(rev.committer),
183
self.revision_format_num or self.format_num,
185
encode_and_escape(cache_utf8.decode(rev.revision_id)),
331
decode_utf8 = cache_utf8.decode
332
revision_id = rev.revision_id
333
if isinstance(revision_id, str):
334
revision_id = decode_utf8(revision_id)
335
format_num = self.format_num
336
if self.revision_format_num is not None:
337
format_num = self.revision_format_num
338
root = Element('revision',
339
committer = rev.committer,
340
timestamp = '%.3f' % rev.timestamp,
341
revision_id = revision_id,
342
inventory_sha1 = rev.inventory_sha1,
187
345
if rev.timezone is not None:
188
el += b' timezone="%s"' % str(rev.timezone).encode('ascii')
189
lines.append(el + b'>\n')
190
message = encode_and_escape(escape_invalid_chars(rev.message)[0])
191
lines.extend((b'<message>' + message + b'</message>\n').splitlines(True))
346
root.set('timezone', str(rev.timezone))
348
msg = SubElement(root, 'message')
349
msg.text = escape_invalid_chars(rev.message)[0]
192
351
if rev.parent_ids:
193
lines.append(b'<parents>\n')
352
pelts = SubElement(root, 'parents')
353
pelts.tail = pelts.text = '\n'
194
354
for parent_id in rev.parent_ids:
195
355
_mod_revision.check_not_reserved_id(parent_id)
197
b'<revision_ref revision_id="%s" />\n'
198
% encode_and_escape(cache_utf8.decode(parent_id)))
199
lines.append(b'</parents>\n')
356
p = SubElement(pelts, 'revision_ref')
358
if isinstance(parent_id, str):
359
parent_id = decode_utf8(parent_id)
360
p.set('revision_id', parent_id)
200
361
if rev.properties:
201
preamble = b'<properties>'
202
for prop_name, prop_value in sorted(rev.properties.items()):
204
proplines = (preamble + b'<property name="%s">%s</property>\n' % (
205
encode_and_escape(prop_name),
206
encode_and_escape(escape_invalid_chars(prop_value)[0]))).splitlines(True)
208
proplines = [preamble + b'<property name="%s" />\n' % (encode_and_escape(prop_name), )]
210
lines.extend(proplines)
211
lines.append(b'</properties>\n')
212
lines.append(b'</revision>\n')
215
def _unpack_entry(self, elt, entry_cache=None, return_from_cache=False):
216
# This is here because it's overridden by xml7
217
return unpack_inventory_entry(elt, entry_cache,
220
def _unpack_inventory(self, elt, revision_id=None, entry_cache=None,
221
return_from_cache=False):
362
self._pack_revision_properties(rev, root)
365
def _pack_revision_properties(self, rev, under_element):
366
top_elt = SubElement(under_element, 'properties')
367
for prop_name, prop_value in sorted(rev.properties.items()):
368
prop_elt = SubElement(top_elt, 'property')
369
prop_elt.set('name', prop_name)
370
prop_elt.text = prop_value
374
def _unpack_inventory(self, elt, revision_id=None, entry_cache=None):
222
375
"""Construct from XML Element"""
223
inv = unpack_inventory_flat(elt, self.format_num, self._unpack_entry,
224
entry_cache, return_from_cache)
376
if elt.tag != 'inventory':
377
raise errors.UnexpectedInventoryFormat('Root tag is %r' % elt.tag)
378
format = elt.get('format')
379
if format != self.format_num:
380
raise errors.UnexpectedInventoryFormat('Invalid format version %r'
382
revision_id = elt.get('revision_id')
383
if revision_id is not None:
384
revision_id = cache_utf8.encode(revision_id)
385
inv = inventory.Inventory(root_id=None, revision_id=revision_id)
387
ie = self._unpack_entry(e, entry_cache=entry_cache)
225
389
self._check_cache_size(len(inv), entry_cache)
392
def _unpack_entry(self, elt, entry_cache=None):
394
file_id = elt_get('file_id')
395
revision = elt_get('revision')
396
# Check and see if we have already unpacked this exact entry
397
# Some timings for "repo.revision_trees(last_100_revs)"
399
# unmodified 4.1s 40.8s
401
# using fifo 2.83s 29.1s
405
# no_copy 2.00s 20.5s
406
# no_c,dict 1.95s 18.0s
407
# Note that a cache of 10k nodes is more than sufficient to hold all of
408
# the inventory for the last 100 revs for bzr, but not for mysql (20k
409
# is enough for mysql, which saves the same 2s as using a dict)
411
# Breakdown of mysql using time.clock()
412
# 4.1s 2 calls to element.get for file_id, revision_id
413
# 4.5s cache_hit lookup
414
# 7.1s InventoryFile.copy()
415
# 2.4s InventoryDirectory.copy()
416
# 0.4s decoding unique entries
417
# 1.6s decoding entries after FIFO fills up
418
# 0.8s Adding nodes to FIFO (including flushes)
419
# 0.1s cache miss lookups
421
# 4.1s 2 calls to element.get for file_id, revision_id
422
# 9.9s cache_hit lookup
423
# 10.8s InventoryEntry.copy()
424
# 0.3s cache miss lookus
425
# 1.2s decoding entries
426
# 1.0s adding nodes to LRU
427
if entry_cache is not None and revision is not None:
428
key = (file_id, revision)
430
# We copy it, because some operations may mutate it
431
cached_ie = entry_cache[key]
435
# Only copying directory entries drops us 2.85s => 2.35s
436
# if cached_ie.kind == 'directory':
437
# return cached_ie.copy()
439
return cached_ie.copy()
442
if not InventoryEntry.versionable_kind(kind):
443
raise AssertionError('unsupported entry kind %s' % kind)
445
get_cached = _get_utf8_or_ascii
447
file_id = get_cached(file_id)
448
if revision is not None:
449
revision = get_cached(revision)
450
parent_id = elt_get('parent_id')
451
if parent_id is not None:
452
parent_id = get_cached(parent_id)
454
if kind == 'directory':
455
ie = inventory.InventoryDirectory(file_id,
459
ie = inventory.InventoryFile(file_id,
462
ie.text_sha1 = elt_get('text_sha1')
463
if elt_get('executable') == 'yes':
465
v = elt_get('text_size')
466
ie.text_size = v and int(v)
467
elif kind == 'symlink':
468
ie = inventory.InventoryLink(file_id,
471
ie.symlink_target = elt_get('symlink_target')
473
raise errors.UnsupportedInventoryKind(kind)
474
ie.revision = revision
475
if revision is not None and entry_cache is not None:
476
# We cache a copy() because callers like to mutate objects, and
477
# that would cause the item in cache to mutate as well.
478
# This has a small effect on many-inventory performance, because
479
# the majority fraction is spent in cache hits, not misses.
480
entry_cache[key] = ie.copy()
228
484
def _unpack_revision(self, elt):
229
485
"""XML Element -> Revision object"""
230
486
format = elt.get('format')
274
529
raise AssertionError("repeated property %r" % name)
275
530
rev.properties[name] = value
277
def _find_text_key_references(self, line_iterator):
278
"""Core routine for extracting references to texts from inventories.
280
This performs the translation of xml lines to revision ids.
282
:param line_iterator: An iterator of lines, origin_version_id
283
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
284
to whether they were referred to by the inventory of the
285
revision_id that they contain. Note that if that revision_id was
286
not part of the line_iterator's output then False will be given -
287
even though it may actually refer to that key.
289
if not self.support_altered_by_hack:
290
raise AssertionError(
291
"_find_text_key_references only "
292
"supported for branches which store inventory as unnested xml"
293
", not on %r" % self)
296
# this code needs to read every new line in every inventory for the
297
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
298
# not present in one of those inventories is unnecessary but not
299
# harmful because we are filtering by the revision id marker in the
300
# inventory lines : we only select file ids altered in one of those
301
# revisions. We don't need to see all lines in the inventory because
302
# only those added in an inventory in rev X can contain a revision=X
304
unescape_revid_cache = {}
305
unescape_fileid_cache = {}
307
# jam 20061218 In a big fetch, this handles hundreds of thousands
308
# of lines, so it has had a lot of inlining and optimizing done.
309
# Sorry that it is a little bit messy.
310
# Move several functions to be local variables, since this is a long
312
search = self._file_ids_altered_regex.search
313
unescape = _unescape_xml
314
setdefault = result.setdefault
315
for line, line_key in line_iterator:
319
# One call to match.group() returning multiple items is quite a
320
# bit faster than 2 calls to match.group() each returning 1
321
file_id, revision_id = match.group('file_id', 'revision_id')
323
# Inlining the cache lookups helps a lot when you make 170,000
324
# lines and 350k ids, versus 8.4 unique ids.
325
# Using a cache helps in 2 ways:
326
# 1) Avoids unnecessary decoding calls
327
# 2) Re-uses cached strings, which helps in future set and
329
# (2) is enough that removing encoding entirely along with
330
# the cache (so we are using plain strings) results in no
331
# performance improvement.
333
revision_id = unescape_revid_cache[revision_id]
335
unescaped = unescape(revision_id)
336
unescape_revid_cache[revision_id] = unescaped
337
revision_id = unescaped
339
# Note that unconditionally unescaping means that we deserialise
340
# every fileid, which for general 'pull' is not great, but we don't
341
# really want to have some many fulltexts that this matters anyway.
344
file_id = unescape_fileid_cache[file_id]
346
unescaped = unescape(file_id)
347
unescape_fileid_cache[file_id] = unescaped
350
key = (file_id, revision_id)
351
setdefault(key, False)
352
if revision_id == line_key[-1]:
357
533
serializer_v8 = Serializer_v8()