1
# Copyright (C) 2005-2010 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25
revision as _mod_revision,
28
from bzrlib.xml_serializer import (
34
from bzrlib.inventory import InventoryEntry
35
from bzrlib.revision import Revision
36
from bzrlib.errors import BzrError
43
"'":"'", # FIXME: overkill
58
def _unescaper(match, _map=_xml_unescape_map):
63
if not code.startswith('#'):
65
return unichr(int(code[1:])).encode('utf8')
71
def _unescape_xml(data):
72
"""Unescape predefined XML entities in a string of data."""
74
if _unescape_re is None:
75
_unescape_re = re.compile('\&([^;]*);')
76
return _unescape_re.sub(_unescaper, data)
79
def _ensure_utf8_re():
80
"""Make sure the _utf8_re and _unicode_re regexes have been compiled."""
81
global _utf8_re, _unicode_re
83
_utf8_re = re.compile('[&<>\'\"]|[\x80-\xff]+')
84
if _unicode_re is None:
85
_unicode_re = re.compile(u'[&<>\'\"\u0080-\uffff]')
88
def _unicode_escape_replace(match, _map=_xml_escape_map):
89
"""Replace a string of non-ascii, non XML safe characters with their escape
91
This will escape both Standard XML escapes, like <>"', etc.
92
As well as escaping non ascii characters, because ElementTree did.
93
This helps us remain compatible to older versions of bzr. We may change
94
our policy in the future, though.
96
# jam 20060816 Benchmarks show that try/KeyError is faster if you
97
# expect the entity to rarely miss. There is about a 10% difference
98
# in overall time. But if you miss frequently, then if None is much
99
# faster. For our use case, we *rarely* have a revision id, file id
100
# or path name that is unicode. So use try/KeyError.
102
return _map[match.group()]
104
return "&#%d;" % ord(match.group())
107
def _utf8_escape_replace(match, _map=_xml_escape_map):
108
"""Escape utf8 characters into XML safe ones.
110
This uses 2 tricks. It is either escaping "standard" characters, like "&<>,
111
or it is handling characters with the high-bit set. For ascii characters,
112
we just lookup the replacement in the dictionary. For everything else, we
113
decode back into Unicode, and then use the XML escape code.
116
return _map[match.group()]
118
return ''.join('&#%d;' % ord(uni_chr)
119
for uni_chr in match.group().decode('utf8'))
124
def _encode_and_escape(unicode_or_utf8_str, _map=_to_escaped_map):
125
"""Encode the string into utf8, and escape invalid XML characters"""
126
# We frequently get entities we have not seen before, so it is better
127
# to check if None, rather than try/KeyError
128
text = _map.get(unicode_or_utf8_str)
130
if unicode_or_utf8_str.__class__ is unicode:
131
# The alternative policy is to do a regular UTF8 encoding
132
# and then escape only XML meta characters.
133
# Performance is equivalent once you use cache_utf8. *However*
134
# this makes the serialized texts incompatible with old versions
135
# of bzr. So no net gain. (Perhaps the read code would handle utf8
136
# better than entity escapes, but cElementTree seems to do just fine
138
text = str(_unicode_re.sub(_unicode_escape_replace,
139
unicode_or_utf8_str)) + '"'
141
# Plain strings are considered to already be in utf-8 so we do a
142
# slightly different method for escaping.
143
text = _utf8_re.sub(_utf8_escape_replace,
144
unicode_or_utf8_str) + '"'
145
_map[unicode_or_utf8_str] = text
149
def _get_utf8_or_ascii(a_str,
150
_encode_utf8=cache_utf8.encode,
151
_get_cached_ascii=cache_utf8.get_cached_ascii):
152
"""Return a cached version of the string.
154
cElementTree will return a plain string if the XML is plain ascii. It only
155
returns Unicode when it needs to. We want to work in utf-8 strings. So if
156
cElementTree returns a plain string, we can just return the cached version.
157
If it is Unicode, then we need to encode it.
159
:param a_str: An 8-bit string or Unicode as returned by
160
cElementTree.Element.get()
161
:return: A utf-8 encoded 8-bit string.
163
# This is fairly optimized because we know what cElementTree does, this is
164
# not meant as a generic function for all cases. Because it is possible for
165
# an 8-bit string to not be ascii or valid utf8.
166
if a_str.__class__ is unicode:
167
return _encode_utf8(a_str)
173
"""Clean out the unicode => escaped map"""
174
_to_escaped_map.clear()
177
class Serializer_v8(XMLSerializer):
178
"""This serialiser adds rich roots.
180
Its revision format number matches its inventory number.
186
support_altered_by_hack = True
187
# This format supports the altered-by hack that reads file ids directly out
188
# of the versionedfile, without doing XML parsing.
190
supported_kinds = set(['file', 'directory', 'symlink'])
192
revision_format_num = None
194
# The search regex used by xml based repositories to determine what things
195
# where changed in a single commit.
196
_file_ids_altered_regex = lazy_regex.lazy_compile(
197
r'file_id="(?P<file_id>[^"]+)"'
198
r'.* revision="(?P<revision_id>[^"]+)"'
201
def _check_revisions(self, inv):
202
"""Extension point for subclasses to check during serialisation.
204
:param inv: An inventory about to be serialised, to be checked.
205
:raises: AssertionError if an error has occurred.
207
if inv.revision_id is None:
208
raise AssertionError("inv.revision_id is None")
209
if inv.root.revision is None:
210
raise AssertionError("inv.root.revision is None")
212
def _check_cache_size(self, inv_size, entry_cache):
213
"""Check that the entry_cache is large enough.
215
We want the cache to be ~2x the size of an inventory. The reason is
216
because we use a FIFO cache, and how Inventory records are likely to
217
change. In general, you have a small number of records which change
218
often, and a lot of records which do not change at all. So when the
219
cache gets full, you actually flush out a lot of the records you are
220
interested in, which means you need to recreate all of those records.
221
An LRU Cache would be better, but the overhead negates the cache
224
One way to look at it, only the size of the cache > len(inv) is your
225
'working' set. And in general, it shouldn't be a problem to hold 2
226
inventories in memory anyway.
228
:param inv_size: The number of entries in an inventory.
230
if entry_cache is None:
232
# 1.5 times might also be reasonable.
233
recommended_min_cache_size = inv_size * 1.5
234
if entry_cache.cache_size() < recommended_min_cache_size:
235
recommended_cache_size = inv_size * 2
236
trace.mutter('Resizing the inventory entry cache from %d to %d',
237
entry_cache.cache_size(), recommended_cache_size)
238
entry_cache.resize(recommended_cache_size)
240
def write_inventory_to_lines(self, inv):
241
"""Return a list of lines with the encoded inventory."""
242
return self.write_inventory(inv, None)
244
def write_inventory_to_string(self, inv, working=False):
245
"""Just call write_inventory with a StringIO and return the value.
247
:param working: If True skip history data - text_sha1, text_size,
248
reference_revision, symlink_target.
250
sio = cStringIO.StringIO()
251
self.write_inventory(inv, sio, working)
252
return sio.getvalue()
254
def write_inventory(self, inv, f, working=False):
255
"""Write inventory to a file.
257
:param inv: the inventory to write.
258
:param f: the file to write. (May be None if the lines are the desired
260
:param working: If True skip history data - text_sha1, text_size,
261
reference_revision, symlink_target.
262
:return: The inventory as a list of lines.
265
self._check_revisions(inv)
267
append = output.append
268
self._append_inventory_root(append, inv)
269
entries = inv.iter_entries()
271
root_path, root_ie = entries.next()
272
for path, ie in entries:
273
if ie.parent_id != self.root_id:
274
parent_str = ' parent_id="'
275
parent_id = _encode_and_escape(ie.parent_id)
279
if ie.kind == 'file':
281
executable = ' executable="yes"'
285
append('<file%s file_id="%s name="%s%s%s revision="%s '
286
'text_sha1="%s" text_size="%d" />\n' % (
287
executable, _encode_and_escape(ie.file_id),
288
_encode_and_escape(ie.name), parent_str, parent_id,
289
_encode_and_escape(ie.revision), ie.text_sha1,
292
append('<file%s file_id="%s name="%s%s%s />\n' % (
293
executable, _encode_and_escape(ie.file_id),
294
_encode_and_escape(ie.name), parent_str, parent_id))
295
elif ie.kind == 'directory':
297
append('<directory file_id="%s name="%s%s%s revision="%s '
299
_encode_and_escape(ie.file_id),
300
_encode_and_escape(ie.name),
301
parent_str, parent_id,
302
_encode_and_escape(ie.revision)))
304
append('<directory file_id="%s name="%s%s%s />\n' % (
305
_encode_and_escape(ie.file_id),
306
_encode_and_escape(ie.name),
307
parent_str, parent_id))
308
elif ie.kind == 'symlink':
310
append('<symlink file_id="%s name="%s%s%s revision="%s '
311
'symlink_target="%s />\n' % (
312
_encode_and_escape(ie.file_id),
313
_encode_and_escape(ie.name),
314
parent_str, parent_id,
315
_encode_and_escape(ie.revision),
316
_encode_and_escape(ie.symlink_target)))
318
append('<symlink file_id="%s name="%s%s%s />\n' % (
319
_encode_and_escape(ie.file_id),
320
_encode_and_escape(ie.name),
321
parent_str, parent_id))
322
elif ie.kind == 'tree-reference':
323
if ie.kind not in self.supported_kinds:
324
raise errors.UnsupportedInventoryKind(ie.kind)
326
append('<tree-reference file_id="%s name="%s%s%s '
327
'revision="%s reference_revision="%s />\n' % (
328
_encode_and_escape(ie.file_id),
329
_encode_and_escape(ie.name),
330
parent_str, parent_id,
331
_encode_and_escape(ie.revision),
332
_encode_and_escape(ie.reference_revision)))
334
append('<tree-reference file_id="%s name="%s%s%s />\n' % (
335
_encode_and_escape(ie.file_id),
336
_encode_and_escape(ie.name),
337
parent_str, parent_id))
339
raise errors.UnsupportedInventoryKind(ie.kind)
340
append('</inventory>\n')
343
# Just to keep the cache from growing without bounds
344
# but we may actually not want to do clear the cache
348
def _append_inventory_root(self, append, inv):
349
"""Append the inventory root to output."""
350
if inv.revision_id is not None:
351
revid1 = ' revision_id="'
352
revid2 = _encode_and_escape(inv.revision_id)
356
append('<inventory format="%s"%s%s>\n' % (
357
self.format_num, revid1, revid2))
358
append('<directory file_id="%s name="%s revision="%s />\n' % (
359
_encode_and_escape(inv.root.file_id),
360
_encode_and_escape(inv.root.name),
361
_encode_and_escape(inv.root.revision)))
363
def _pack_revision(self, rev):
364
"""Revision object -> xml tree"""
365
# For the XML format, we need to write them as Unicode rather than as
366
# utf-8 strings. So that cElementTree can handle properly escaping
368
decode_utf8 = cache_utf8.decode
369
revision_id = rev.revision_id
370
if isinstance(revision_id, str):
371
revision_id = decode_utf8(revision_id)
372
format_num = self.format_num
373
if self.revision_format_num is not None:
374
format_num = self.revision_format_num
375
root = Element('revision',
376
committer = rev.committer,
377
timestamp = '%.3f' % rev.timestamp,
378
revision_id = revision_id,
379
inventory_sha1 = rev.inventory_sha1,
382
if rev.timezone is not None:
383
root.set('timezone', str(rev.timezone))
385
msg = SubElement(root, 'message')
386
msg.text = escape_invalid_chars(rev.message)[0]
389
pelts = SubElement(root, 'parents')
390
pelts.tail = pelts.text = '\n'
391
for parent_id in rev.parent_ids:
392
_mod_revision.check_not_reserved_id(parent_id)
393
p = SubElement(pelts, 'revision_ref')
395
if isinstance(parent_id, str):
396
parent_id = decode_utf8(parent_id)
397
p.set('revision_id', parent_id)
399
self._pack_revision_properties(rev, root)
402
def _pack_revision_properties(self, rev, under_element):
403
top_elt = SubElement(under_element, 'properties')
404
for prop_name, prop_value in sorted(rev.properties.items()):
405
prop_elt = SubElement(top_elt, 'property')
406
prop_elt.set('name', prop_name)
407
prop_elt.text = prop_value
411
def _unpack_inventory(self, elt, revision_id=None, entry_cache=None,
412
return_from_cache=False):
413
"""Construct from XML Element"""
414
if elt.tag != 'inventory':
415
raise errors.UnexpectedInventoryFormat('Root tag is %r' % elt.tag)
416
format = elt.get('format')
417
if format != self.format_num:
418
raise errors.UnexpectedInventoryFormat('Invalid format version %r'
420
revision_id = elt.get('revision_id')
421
if revision_id is not None:
422
revision_id = cache_utf8.encode(revision_id)
423
inv = inventory.Inventory(root_id=None, revision_id=revision_id)
425
ie = self._unpack_entry(e, entry_cache=entry_cache,
426
return_from_cache=return_from_cache)
428
self._check_cache_size(len(inv), entry_cache)
431
def _unpack_entry(self, elt, entry_cache=None, return_from_cache=False):
433
file_id = elt_get('file_id')
434
revision = elt_get('revision')
435
# Check and see if we have already unpacked this exact entry
436
# Some timings for "repo.revision_trees(last_100_revs)"
438
# unmodified 4.1s 40.8s
440
# using fifo 2.83s 29.1s
444
# no_copy 2.00s 20.5s
445
# no_c,dict 1.95s 18.0s
446
# Note that a cache of 10k nodes is more than sufficient to hold all of
447
# the inventory for the last 100 revs for bzr, but not for mysql (20k
448
# is enough for mysql, which saves the same 2s as using a dict)
450
# Breakdown of mysql using time.clock()
451
# 4.1s 2 calls to element.get for file_id, revision_id
452
# 4.5s cache_hit lookup
453
# 7.1s InventoryFile.copy()
454
# 2.4s InventoryDirectory.copy()
455
# 0.4s decoding unique entries
456
# 1.6s decoding entries after FIFO fills up
457
# 0.8s Adding nodes to FIFO (including flushes)
458
# 0.1s cache miss lookups
460
# 4.1s 2 calls to element.get for file_id, revision_id
461
# 9.9s cache_hit lookup
462
# 10.8s InventoryEntry.copy()
463
# 0.3s cache miss lookus
464
# 1.2s decoding entries
465
# 1.0s adding nodes to LRU
466
if entry_cache is not None and revision is not None:
467
key = (file_id, revision)
469
# We copy it, because some operations may mutate it
470
cached_ie = entry_cache[key]
474
# Only copying directory entries drops us 2.85s => 2.35s
475
if return_from_cache:
476
if cached_ie.kind == 'directory':
477
return cached_ie.copy()
479
return cached_ie.copy()
482
if not InventoryEntry.versionable_kind(kind):
483
raise AssertionError('unsupported entry kind %s' % kind)
485
get_cached = _get_utf8_or_ascii
487
file_id = get_cached(file_id)
488
if revision is not None:
489
revision = get_cached(revision)
490
parent_id = elt_get('parent_id')
491
if parent_id is not None:
492
parent_id = get_cached(parent_id)
494
if kind == 'directory':
495
ie = inventory.InventoryDirectory(file_id,
499
ie = inventory.InventoryFile(file_id,
502
ie.text_sha1 = elt_get('text_sha1')
503
if elt_get('executable') == 'yes':
505
v = elt_get('text_size')
506
ie.text_size = v and int(v)
507
elif kind == 'symlink':
508
ie = inventory.InventoryLink(file_id,
511
ie.symlink_target = elt_get('symlink_target')
513
raise errors.UnsupportedInventoryKind(kind)
514
ie.revision = revision
515
if revision is not None and entry_cache is not None:
516
# We cache a copy() because callers like to mutate objects, and
517
# that would cause the item in cache to mutate as well.
518
# This has a small effect on many-inventory performance, because
519
# the majority fraction is spent in cache hits, not misses.
520
entry_cache[key] = ie.copy()
524
def _unpack_revision(self, elt):
525
"""XML Element -> Revision object"""
526
format = elt.get('format')
527
format_num = self.format_num
528
if self.revision_format_num is not None:
529
format_num = self.revision_format_num
530
if format is not None:
531
if format != format_num:
532
raise BzrError("invalid format version %r on revision"
534
get_cached = _get_utf8_or_ascii
535
rev = Revision(committer = elt.get('committer'),
536
timestamp = float(elt.get('timestamp')),
537
revision_id = get_cached(elt.get('revision_id')),
538
inventory_sha1 = elt.get('inventory_sha1')
540
parents = elt.find('parents') or []
542
rev.parent_ids.append(get_cached(p.get('revision_id')))
543
self._unpack_revision_properties(elt, rev)
544
v = elt.get('timezone')
548
rev.timezone = int(v)
549
rev.message = elt.findtext('message') # text of <message>
552
def _unpack_revision_properties(self, elt, rev):
553
"""Unpack properties onto a revision."""
554
props_elt = elt.find('properties')
557
for prop_elt in props_elt:
558
if prop_elt.tag != 'property':
559
raise AssertionError(
560
"bad tag under properties list: %r" % prop_elt.tag)
561
name = prop_elt.get('name')
562
value = prop_elt.text
563
# If a property had an empty value ('') cElementTree reads
564
# that back as None, convert it back to '', so that all
565
# properties have string values
568
if name in rev.properties:
569
raise AssertionError("repeated property %r" % name)
570
rev.properties[name] = value
572
def _find_text_key_references(self, line_iterator):
573
"""Core routine for extracting references to texts from inventories.
575
This performs the translation of xml lines to revision ids.
577
:param line_iterator: An iterator of lines, origin_version_id
578
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
579
to whether they were referred to by the inventory of the
580
revision_id that they contain. Note that if that revision_id was
581
not part of the line_iterator's output then False will be given -
582
even though it may actually refer to that key.
584
if not self.support_altered_by_hack:
585
raise AssertionError(
586
"_find_text_key_references only "
587
"supported for branches which store inventory as unnested xml"
588
", not on %r" % self)
591
# this code needs to read every new line in every inventory for the
592
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
593
# not present in one of those inventories is unnecessary but not
594
# harmful because we are filtering by the revision id marker in the
595
# inventory lines : we only select file ids altered in one of those
596
# revisions. We don't need to see all lines in the inventory because
597
# only those added in an inventory in rev X can contain a revision=X
599
unescape_revid_cache = {}
600
unescape_fileid_cache = {}
602
# jam 20061218 In a big fetch, this handles hundreds of thousands
603
# of lines, so it has had a lot of inlining and optimizing done.
604
# Sorry that it is a little bit messy.
605
# Move several functions to be local variables, since this is a long
607
search = self._file_ids_altered_regex.search
608
unescape = _unescape_xml
609
setdefault = result.setdefault
610
for line, line_key in line_iterator:
614
# One call to match.group() returning multiple items is quite a
615
# bit faster than 2 calls to match.group() each returning 1
616
file_id, revision_id = match.group('file_id', 'revision_id')
618
# Inlining the cache lookups helps a lot when you make 170,000
619
# lines and 350k ids, versus 8.4 unique ids.
620
# Using a cache helps in 2 ways:
621
# 1) Avoids unnecessary decoding calls
622
# 2) Re-uses cached strings, which helps in future set and
624
# (2) is enough that removing encoding entirely along with
625
# the cache (so we are using plain strings) results in no
626
# performance improvement.
628
revision_id = unescape_revid_cache[revision_id]
630
unescaped = unescape(revision_id)
631
unescape_revid_cache[revision_id] = unescaped
632
revision_id = unescaped
634
# Note that unconditionally unescaping means that we deserialise
635
# every fileid, which for general 'pull' is not great, but we don't
636
# really want to have some many fulltexts that this matters anyway.
639
file_id = unescape_fileid_cache[file_id]
641
unescaped = unescape(file_id)
642
unescape_fileid_cache[file_id] = unescaped
645
key = (file_id, revision_id)
646
setdefault(key, False)
647
if revision_id == line_key[-1]:
652
serializer_v8 = Serializer_v8()
added
removed
bzrlib/xml8.py
