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
24
revision as _mod_revision,
27
from bzrlib.xml_serializer import SubElement, Element, Serializer
28
from bzrlib.inventory import ROOT_ID, Inventory, InventoryEntry
27
from bzrlib.xml_serializer import (
34
serialize_inventory_flat,
35
unpack_inventory_entry,
36
unpack_inventory_flat,
29
38
from bzrlib.revision import Revision
30
39
from bzrlib.errors import BzrError
37
"'":"'", # FIXME: overkill
44
def _ensure_utf8_re():
45
"""Make sure the _utf8_re and _unicode_re regexes have been compiled."""
46
global _utf8_re, _unicode_re
48
_utf8_re = re.compile('[&<>\'\"]|[\x80-\xff]+')
49
if _unicode_re is None:
50
_unicode_re = re.compile(u'[&<>\'\"\u0080-\uffff]')
53
def _unicode_escape_replace(match, _map=_xml_escape_map):
54
"""Replace a string of non-ascii, non XML safe characters with their escape
56
This will escape both Standard XML escapes, like <>"', etc.
57
As well as escaping non ascii characters, because ElementTree did.
58
This helps us remain compatible to older versions of bzr. We may change
59
our policy in the future, though.
61
# jam 20060816 Benchmarks show that try/KeyError is faster if you
62
# expect the entity to rarely miss. There is about a 10% difference
63
# in overall time. But if you miss frequently, then if None is much
64
# faster. For our use case, we *rarely* have a revision id, file id
65
# or path name that is unicode. So use try/KeyError.
67
return _map[match.group()]
69
return "&#%d;" % ord(match.group())
72
def _utf8_escape_replace(match, _map=_xml_escape_map):
73
"""Escape utf8 characters into XML safe ones.
75
This uses 2 tricks. It is either escaping "standard" characters, like "&<>,
76
or it is handling characters with the high-bit set. For ascii characters,
77
we just lookup the replacement in the dictionary. For everything else, we
78
decode back into Unicode, and then use the XML escape code.
81
return _map[match.group()]
83
return ''.join('&#%d;' % ord(uni_chr)
84
for uni_chr in match.group().decode('utf8'))
89
def _encode_and_escape(unicode_or_utf8_str, _map=_to_escaped_map):
90
"""Encode the string into utf8, and escape invalid XML characters"""
91
# We frequently get entities we have not seen before, so it is better
92
# to check if None, rather than try/KeyError
93
text = _map.get(unicode_or_utf8_str)
95
if unicode_or_utf8_str.__class__ == unicode:
96
# The alternative policy is to do a regular UTF8 encoding
97
# and then escape only XML meta characters.
98
# Performance is equivalent once you use cache_utf8. *However*
99
# this makes the serialized texts incompatible with old versions
100
# of bzr. So no net gain. (Perhaps the read code would handle utf8
101
# better than entity escapes, but cElementTree seems to do just fine
103
text = str(_unicode_re.sub(_unicode_escape_replace,
104
unicode_or_utf8_str)) + '"'
106
# Plain strings are considered to already be in utf-8 so we do a
107
# slightly different method for escaping.
108
text = _utf8_re.sub(_utf8_escape_replace,
109
unicode_or_utf8_str) + '"'
110
_map[unicode_or_utf8_str] = text
114
def _get_utf8_or_ascii(a_str,
115
_encode_utf8=cache_utf8.encode,
116
_get_cached_ascii=cache_utf8.get_cached_ascii):
117
"""Return a cached version of the string.
119
cElementTree will return a plain string if the XML is plain ascii. It only
120
returns Unicode when it needs to. We want to work in utf-8 strings. So if
121
cElementTree returns a plain string, we can just return the cached version.
122
If it is Unicode, then we need to encode it.
124
:param a_str: An 8-bit string or Unicode as returned by
125
cElementTree.Element.get()
126
:return: A utf-8 encoded 8-bit string.
128
# This is fairly optimized because we know what cElementTree does, this is
129
# not meant as a generic function for all cases. Because it is possible for
130
# an 8-bit string to not be ascii or valid utf8.
131
if a_str.__class__ == unicode:
132
return _encode_utf8(a_str)
134
return _get_cached_ascii(a_str)
138
"""Clean out the unicode => escaped map"""
139
_to_escaped_map.clear()
142
class Serializer_v8(Serializer):
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):
143
69
"""This serialiser adds rich roots.
145
71
Its revision format number matches its inventory number.
157
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>[^"]+)"'
159
92
def _check_revisions(self, inv):
160
93
"""Extension point for subclasses to check during serialisation.
162
95
:param inv: An inventory about to be serialised, to be checked.
163
:raises: AssertionError if an error has occured.
96
:raises: AssertionError if an error has occurred.
165
98
if inv.revision_id is None:
166
raise AssertionError()
99
raise AssertionError("inv.revision_id is None")
167
100
if inv.root.revision is None:
168
raise AssertionError()
101
raise AssertionError("inv.root.revision is None")
170
103
def _check_cache_size(self, inv_size, entry_cache):
171
104
"""Check that the entry_cache is large enough.
219
152
reference_revision, symlink_target.
220
153
:return: The inventory as a list of lines.
223
self._check_revisions(inv)
225
156
append = output.append
226
157
self._append_inventory_root(append, inv)
227
entries = inv.iter_entries()
229
root_path, root_ie = entries.next()
230
for path, ie in entries:
231
if ie.parent_id != self.root_id:
232
parent_str = ' parent_id="'
233
parent_id = _encode_and_escape(ie.parent_id)
237
if ie.kind == 'file':
239
executable = ' executable="yes"'
243
append('<file%s file_id="%s name="%s%s%s revision="%s '
244
'text_sha1="%s" text_size="%d" />\n' % (
245
executable, _encode_and_escape(ie.file_id),
246
_encode_and_escape(ie.name), parent_str, parent_id,
247
_encode_and_escape(ie.revision), ie.text_sha1,
250
append('<file%s file_id="%s name="%s%s%s />\n' % (
251
executable, _encode_and_escape(ie.file_id),
252
_encode_and_escape(ie.name), parent_str, parent_id))
253
elif ie.kind == 'directory':
255
append('<directory file_id="%s name="%s%s%s revision="%s '
257
_encode_and_escape(ie.file_id),
258
_encode_and_escape(ie.name),
259
parent_str, parent_id,
260
_encode_and_escape(ie.revision)))
262
append('<directory file_id="%s name="%s%s%s />\n' % (
263
_encode_and_escape(ie.file_id),
264
_encode_and_escape(ie.name),
265
parent_str, parent_id))
266
elif ie.kind == 'symlink':
268
append('<symlink file_id="%s name="%s%s%s revision="%s '
269
'symlink_target="%s />\n' % (
270
_encode_and_escape(ie.file_id),
271
_encode_and_escape(ie.name),
272
parent_str, parent_id,
273
_encode_and_escape(ie.revision),
274
_encode_and_escape(ie.symlink_target)))
276
append('<symlink file_id="%s name="%s%s%s />\n' % (
277
_encode_and_escape(ie.file_id),
278
_encode_and_escape(ie.name),
279
parent_str, parent_id))
280
elif ie.kind == 'tree-reference':
281
if ie.kind not in self.supported_kinds:
282
raise errors.UnsupportedInventoryKind(ie.kind)
284
append('<tree-reference file_id="%s name="%s%s%s '
285
'revision="%s reference_revision="%s />\n' % (
286
_encode_and_escape(ie.file_id),
287
_encode_and_escape(ie.name),
288
parent_str, parent_id,
289
_encode_and_escape(ie.revision),
290
_encode_and_escape(ie.reference_revision)))
292
append('<tree-reference file_id="%s name="%s%s%s />\n' % (
293
_encode_and_escape(ie.file_id),
294
_encode_and_escape(ie.name),
295
parent_str, parent_id))
297
raise errors.UnsupportedInventoryKind(ie.kind)
298
append('</inventory>\n')
158
serialize_inventory_flat(inv, append,
159
self.root_id, self.supported_kinds, working)
299
160
if f is not None:
300
161
f.writelines(output)
301
162
# Just to keep the cache from growing without bounds
307
168
"""Append the inventory root to output."""
308
169
if inv.revision_id is not None:
309
170
revid1 = ' revision_id="'
310
revid2 = _encode_and_escape(inv.revision_id)
171
revid2 = encode_and_escape(inv.revision_id)
314
175
append('<inventory format="%s"%s%s>\n' % (
315
176
self.format_num, revid1, revid2))
316
177
append('<directory file_id="%s name="%s revision="%s />\n' % (
317
_encode_and_escape(inv.root.file_id),
318
_encode_and_escape(inv.root.name),
319
_encode_and_escape(inv.root.revision)))
178
encode_and_escape(inv.root.file_id),
179
encode_and_escape(inv.root.name),
180
encode_and_escape(inv.root.revision)))
321
182
def _pack_revision(self, rev):
322
183
"""Revision object -> xml tree"""
366
227
prop_elt.tail = '\n'
367
228
top_elt.tail = '\n'
369
def _unpack_inventory(self, elt, revision_id=None, entry_cache=None):
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,
235
def _unpack_inventory(self, elt, revision_id=None, entry_cache=None,
236
return_from_cache=False):
370
237
"""Construct from XML Element"""
371
if elt.tag != 'inventory':
372
raise errors.UnexpectedInventoryFormat('Root tag is %r' % elt.tag)
373
format = elt.get('format')
374
if format != self.format_num:
375
raise errors.UnexpectedInventoryFormat('Invalid format version %r'
377
revision_id = elt.get('revision_id')
378
if revision_id is not None:
379
revision_id = cache_utf8.encode(revision_id)
380
inv = inventory.Inventory(root_id=None, revision_id=revision_id)
382
ie = self._unpack_entry(e, entry_cache=entry_cache)
238
inv = unpack_inventory_flat(elt, self.format_num, self._unpack_entry,
239
entry_cache, return_from_cache)
384
240
self._check_cache_size(len(inv), entry_cache)
387
def _unpack_entry(self, elt, entry_cache=None):
389
file_id = elt_get('file_id')
390
revision = elt_get('revision')
391
# Check and see if we have already unpacked this exact entry
392
# Some timings for "repo.revision_trees(last_100_revs)"
394
# unmodified 4.1s 40.8s
396
# using fifo 2.83s 29.1s
400
# no_copy 2.00s 20.5s
401
# no_c,dict 1.95s 18.0s
402
# Note that a cache of 10k nodes is more than sufficient to hold all of
403
# the inventory for the last 100 revs for bzr, but not for mysql (20k
404
# is enough for mysql, which saves the same 2s as using a dict)
406
# Breakdown of mysql using time.clock()
407
# 4.1s 2 calls to element.get for file_id, revision_id
408
# 4.5s cache_hit lookup
409
# 7.1s InventoryFile.copy()
410
# 2.4s InventoryDirectory.copy()
411
# 0.4s decoding unique entries
412
# 1.6s decoding entries after FIFO fills up
413
# 0.8s Adding nodes to FIFO (including flushes)
414
# 0.1s cache miss lookups
416
# 4.1s 2 calls to element.get for file_id, revision_id
417
# 9.9s cache_hit lookup
418
# 10.8s InventoryEntry.copy()
419
# 0.3s cache miss lookus
420
# 1.2s decoding entries
421
# 1.0s adding nodes to LRU
422
if entry_cache is not None and revision is not None:
423
key = (file_id, revision)
425
# We copy it, because some operatations may mutate it
426
cached_ie = entry_cache[key]
430
# Only copying directory entries drops us 2.85s => 2.35s
431
# if cached_ie.kind == 'directory':
432
# return cached_ie.copy()
434
return cached_ie.copy()
437
if not InventoryEntry.versionable_kind(kind):
438
raise AssertionError('unsupported entry kind %s' % kind)
440
get_cached = _get_utf8_or_ascii
442
file_id = get_cached(file_id)
443
if revision is not None:
444
revision = get_cached(revision)
445
parent_id = elt_get('parent_id')
446
if parent_id is not None:
447
parent_id = get_cached(parent_id)
449
if kind == 'directory':
450
ie = inventory.InventoryDirectory(file_id,
454
ie = inventory.InventoryFile(file_id,
457
ie.text_sha1 = elt_get('text_sha1')
458
if elt_get('executable') == 'yes':
460
v = elt_get('text_size')
461
ie.text_size = v and int(v)
462
elif kind == 'symlink':
463
ie = inventory.InventoryLink(file_id,
466
ie.symlink_target = elt_get('symlink_target')
468
raise errors.UnsupportedInventoryKind(kind)
469
ie.revision = revision
470
if revision is not None and entry_cache is not None:
471
# We cache a copy() because callers like to mutate objects, and
472
# that would cause the item in cache to mutate as well.
473
# This has a small effect on many-inventory performance, because
474
# the majority fraction is spent in cache hits, not misses.
475
entry_cache[key] = ie.copy()
479
243
def _unpack_revision(self, elt):
480
244
"""XML Element -> Revision object"""
481
245
format = elt.get('format')
524
288
raise AssertionError("repeated property %r" % name)
525
289
rev.properties[name] = value
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]:
528
371
serializer_v8 = Serializer_v8()
added
removed
bzrlib/xml8.py
