159
165
"""Extension point for subclasses to check during serialisation.
161
167
:param inv: An inventory about to be serialised, to be checked.
162
:raises: AssertionError if an error has occured.
168
:raises: AssertionError if an error has occurred.
164
170
if inv.revision_id is None:
165
raise AssertionError()
171
raise AssertionError("inv.revision_id is None")
166
172
if inv.root.revision is None:
167
raise AssertionError()
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)
169
203
def write_inventory_to_lines(self, inv):
170
204
"""Return a list of lines with the encoded inventory."""
350
385
revision_id = cache_utf8.encode(revision_id)
351
386
inv = inventory.Inventory(root_id=None, revision_id=revision_id)
353
ie = self._unpack_entry(e)
388
ie = self._unpack_entry(e, entry_cache=entry_cache,
389
return_from_cache=return_from_cache)
391
self._check_cache_size(len(inv), entry_cache)
357
def _unpack_entry(self, elt):
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()
359
445
if not InventoryEntry.versionable_kind(kind):
360
446
raise AssertionError('unsupported entry kind %s' % kind)
362
448
get_cached = _get_utf8_or_ascii
364
parent_id = elt.get('parent_id')
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')
365
454
if parent_id is not None:
366
455
parent_id = get_cached(parent_id)
367
file_id = get_cached(elt.get('file_id'))
369
457
if kind == 'directory':
370
458
ie = inventory.InventoryDirectory(file_id,
373
461
elif kind == 'file':
374
462
ie = inventory.InventoryFile(file_id,
377
ie.text_sha1 = elt.get('text_sha1')
378
if elt.get('executable') == 'yes':
465
ie.text_sha1 = elt_get('text_sha1')
466
if elt_get('executable') == 'yes':
379
467
ie.executable = True
380
v = elt.get('text_size')
468
v = elt_get('text_size')
381
469
ie.text_size = v and int(v)
382
470
elif kind == 'symlink':
383
471
ie = inventory.InventoryLink(file_id,
386
ie.symlink_target = elt.get('symlink_target')
474
ie.symlink_target = elt_get('symlink_target')
388
476
raise errors.UnsupportedInventoryKind(kind)
389
revision = elt.get('revision')
390
if revision is not None:
391
revision = get_cached(revision)
392
477
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()