164
158
def _check_revisions(self, inv):
165
159
"""Extension point for subclasses to check during serialisation.
161
By default no checking is done.
167
163
:param inv: An inventory about to be serialised, to be checked.
168
:raises: AssertionError if an error has occurred.
170
if inv.revision_id is None:
171
raise AssertionError("inv.revision_id is None")
172
if inv.root.revision is None:
173
raise AssertionError("inv.root.revision is None")
175
def _check_cache_size(self, inv_size, entry_cache):
176
"""Check that the entry_cache is large enough.
178
We want the cache to be ~2x the size of an inventory. The reason is
179
because we use a FIFO cache, and how Inventory records are likely to
180
change. In general, you have a small number of records which change
181
often, and a lot of records which do not change at all. So when the
182
cache gets full, you actually flush out a lot of the records you are
183
interested in, which means you need to recreate all of those records.
184
An LRU Cache would be better, but the overhead negates the cache
187
One way to look at it, only the size of the cache > len(inv) is your
188
'working' set. And in general, it shouldn't be a problem to hold 2
189
inventories in memory anyway.
191
:param inv_size: The number of entries in an inventory.
193
if entry_cache is None:
195
# 1.5 times might also be reasonable.
196
recommended_min_cache_size = inv_size * 1.5
197
if entry_cache.cache_size() < recommended_min_cache_size:
198
recommended_cache_size = inv_size * 2
199
trace.mutter('Resizing the inventory entry cache from %d to %d',
200
entry_cache.cache_size(), recommended_cache_size)
201
entry_cache.resize(recommended_cache_size)
164
:raises: AssertionError if an error has occured.
166
assert inv.revision_id is not None
167
assert inv.root.revision is not None
203
169
def write_inventory_to_lines(self, inv):
204
170
"""Return a list of lines with the encoded inventory."""
365
332
def _pack_revision_properties(self, rev, under_element):
366
333
top_elt = SubElement(under_element, 'properties')
367
334
for prop_name, prop_value in sorted(rev.properties.items()):
335
assert isinstance(prop_name, basestring)
336
assert isinstance(prop_value, basestring)
368
337
prop_elt = SubElement(top_elt, 'property')
369
338
prop_elt.set('name', prop_name)
370
339
prop_elt.text = prop_value
371
340
prop_elt.tail = '\n'
372
341
top_elt.tail = '\n'
374
def _unpack_inventory(self, elt, revision_id=None, entry_cache=None,
375
return_from_cache=False):
343
def _unpack_inventory(self, elt, revision_id=None):
376
344
"""Construct from XML Element"""
377
345
if elt.tag != 'inventory':
378
346
raise errors.UnexpectedInventoryFormat('Root tag is %r' % elt.tag)
385
353
revision_id = cache_utf8.encode(revision_id)
386
354
inv = inventory.Inventory(root_id=None, revision_id=revision_id)
388
ie = self._unpack_entry(e, entry_cache=entry_cache,
389
return_from_cache=return_from_cache)
356
ie = self._unpack_entry(e)
391
self._check_cache_size(len(inv), entry_cache)
358
assert inv.root.revision is not None
394
def _unpack_entry(self, elt, entry_cache=None, return_from_cache=False):
396
file_id = elt_get('file_id')
397
revision = elt_get('revision')
398
# Check and see if we have already unpacked this exact entry
399
# Some timings for "repo.revision_trees(last_100_revs)"
401
# unmodified 4.1s 40.8s
403
# using fifo 2.83s 29.1s
407
# no_copy 2.00s 20.5s
408
# no_c,dict 1.95s 18.0s
409
# Note that a cache of 10k nodes is more than sufficient to hold all of
410
# the inventory for the last 100 revs for bzr, but not for mysql (20k
411
# is enough for mysql, which saves the same 2s as using a dict)
413
# Breakdown of mysql using time.clock()
414
# 4.1s 2 calls to element.get for file_id, revision_id
415
# 4.5s cache_hit lookup
416
# 7.1s InventoryFile.copy()
417
# 2.4s InventoryDirectory.copy()
418
# 0.4s decoding unique entries
419
# 1.6s decoding entries after FIFO fills up
420
# 0.8s Adding nodes to FIFO (including flushes)
421
# 0.1s cache miss lookups
423
# 4.1s 2 calls to element.get for file_id, revision_id
424
# 9.9s cache_hit lookup
425
# 10.8s InventoryEntry.copy()
426
# 0.3s cache miss lookus
427
# 1.2s decoding entries
428
# 1.0s adding nodes to LRU
429
if entry_cache is not None and revision is not None:
430
key = (file_id, revision)
432
# We copy it, because some operations may mutate it
433
cached_ie = entry_cache[key]
437
# Only copying directory entries drops us 2.85s => 2.35s
438
if return_from_cache:
439
if cached_ie.kind == 'directory':
440
return cached_ie.copy()
442
return cached_ie.copy()
361
def _unpack_entry(self, elt):
445
363
if not InventoryEntry.versionable_kind(kind):
446
364
raise AssertionError('unsupported entry kind %s' % kind)
448
366
get_cached = _get_utf8_or_ascii
450
file_id = get_cached(file_id)
451
if revision is not None:
452
revision = get_cached(revision)
453
parent_id = elt_get('parent_id')
368
parent_id = elt.get('parent_id')
454
369
if parent_id is not None:
455
370
parent_id = get_cached(parent_id)
371
file_id = get_cached(elt.get('file_id'))
457
373
if kind == 'directory':
458
374
ie = inventory.InventoryDirectory(file_id,
461
377
elif kind == 'file':
462
378
ie = inventory.InventoryFile(file_id,
465
ie.text_sha1 = elt_get('text_sha1')
466
if elt_get('executable') == 'yes':
381
ie.text_sha1 = elt.get('text_sha1')
382
if elt.get('executable') == 'yes':
467
383
ie.executable = True
468
v = elt_get('text_size')
384
v = elt.get('text_size')
469
385
ie.text_size = v and int(v)
470
386
elif kind == 'symlink':
471
387
ie = inventory.InventoryLink(file_id,
474
ie.symlink_target = elt_get('symlink_target')
390
ie.symlink_target = elt.get('symlink_target')
476
392
raise errors.UnsupportedInventoryKind(kind)
393
revision = elt.get('revision')
394
if revision is not None:
395
revision = get_cached(revision)
477
396
ie.revision = revision
478
if revision is not None and entry_cache is not None:
479
# We cache a copy() because callers like to mutate objects, and
480
# that would cause the item in cache to mutate as well.
481
# This has a small effect on many-inventory performance, because
482
# the majority fraction is spent in cache hits, not misses.
483
entry_cache[key] = ie.copy()
487
400
def _unpack_revision(self, elt):
488
401
"""XML Element -> Revision object"""
402
assert elt.tag == 'revision'
489
403
format = elt.get('format')
490
404
format_num = self.format_num
491
405
if self.revision_format_num is not None:
515
431
def _unpack_revision_properties(self, elt, rev):
516
432
"""Unpack properties onto a revision."""
517
433
props_elt = elt.find('properties')
434
assert len(rev.properties) == 0
518
435
if not props_elt:
520
437
for prop_elt in props_elt:
521
if prop_elt.tag != 'property':
522
raise AssertionError(
523
"bad tag under properties list: %r" % prop_elt.tag)
438
assert prop_elt.tag == 'property', \
439
"bad tag under properties list: %r" % prop_elt.tag
524
440
name = prop_elt.get('name')
525
441
value = prop_elt.text
526
442
# If a property had an empty value ('') cElementTree reads