3358
3304
annotator = _KnitAnnotator(knit)
3359
return iter(annotator.annotate_flat(revision_id))
3362
class _KnitAnnotator(annotate.Annotator):
3305
return iter(annotator.annotate(revision_id))
3308
class _KnitAnnotator(object):
3363
3309
"""Build up the annotations for a text."""
3365
def __init__(self, vf):
3366
annotate.Annotator.__init__(self, vf)
3368
# TODO: handle Nodes which cannot be extracted
3369
# self._ghosts = set()
3371
# Map from (key, parent_key) => matching_blocks, should be 'use once'
3372
self._matching_blocks = {}
3374
# KnitContent objects
3375
self._content_objects = {}
3376
# The number of children that depend on this fulltext content object
3377
self._num_compression_children = {}
3378
# Delta records that need their compression parent before they can be
3380
self._pending_deltas = {}
3381
# Fulltext records that are waiting for their parents fulltexts before
3382
# they can be yielded for annotation
3383
self._pending_annotation = {}
3311
def __init__(self, knit):
3314
# Content objects, differs from fulltexts because of how final newlines
3315
# are treated by knits. the content objects here will always have a
3317
self._fulltext_contents = {}
3319
# Annotated lines of specific revisions
3320
self._annotated_lines = {}
3322
# Track the raw data for nodes that we could not process yet.
3323
# This maps the revision_id of the base to a list of children that will
3324
# annotated from it.
3325
self._pending_children = {}
3327
# Nodes which cannot be extracted
3328
self._ghosts = set()
3330
# Track how many children this node has, so we know if we need to keep
3332
self._annotate_children = {}
3333
self._compression_children = {}
3385
3335
self._all_build_details = {}
3336
# The children => parent revision_id graph
3337
self._revision_id_graph = {}
3339
self._heads_provider = None
3341
self._nodes_to_keep_annotations = set()
3342
self._generations_until_keep = 100
3344
def set_generations_until_keep(self, value):
3345
"""Set the number of generations before caching a node.
3347
Setting this to -1 will cache every merge node, setting this higher
3348
will cache fewer nodes.
3350
self._generations_until_keep = value
3352
def _add_fulltext_content(self, revision_id, content_obj):
3353
self._fulltext_contents[revision_id] = content_obj
3354
# TODO: jam 20080305 It might be good to check the sha1digest here
3355
return content_obj.text()
3357
def _check_parents(self, child, nodes_to_annotate):
3358
"""Check if all parents have been processed.
3360
:param child: A tuple of (rev_id, parents, raw_content)
3361
:param nodes_to_annotate: If child is ready, add it to
3362
nodes_to_annotate, otherwise put it back in self._pending_children
3364
for parent_id in child[1]:
3365
if (parent_id not in self._annotated_lines):
3366
# This parent is present, but another parent is missing
3367
self._pending_children.setdefault(parent_id,
3371
# This one is ready to be processed
3372
nodes_to_annotate.append(child)
3374
def _add_annotation(self, revision_id, fulltext, parent_ids,
3375
left_matching_blocks=None):
3376
"""Add an annotation entry.
3378
All parents should already have been annotated.
3379
:return: A list of children that now have their parents satisfied.
3381
a = self._annotated_lines
3382
annotated_parent_lines = [a[p] for p in parent_ids]
3383
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
3384
fulltext, revision_id, left_matching_blocks,
3385
heads_provider=self._get_heads_provider()))
3386
self._annotated_lines[revision_id] = annotated_lines
3387
for p in parent_ids:
3388
ann_children = self._annotate_children[p]
3389
ann_children.remove(revision_id)
3390
if (not ann_children
3391
and p not in self._nodes_to_keep_annotations):
3392
del self._annotated_lines[p]
3393
del self._all_build_details[p]
3394
if p in self._fulltext_contents:
3395
del self._fulltext_contents[p]
3396
# Now that we've added this one, see if there are any pending
3397
# deltas to be done, certainly this parent is finished
3398
nodes_to_annotate = []
3399
for child in self._pending_children.pop(revision_id, []):
3400
self._check_parents(child, nodes_to_annotate)
3401
return nodes_to_annotate
3387
3403
def _get_build_graph(self, key):
3388
3404
"""Get the graphs for building texts and annotations.
3396
3412
passing to read_records_iter to start reading in the raw data from
3415
if key in self._annotated_lines:
3399
3418
pending = set([key])
3402
self._num_needed_children[key] = 1
3404
3423
# get all pending nodes
3405
3425
this_iteration = pending
3406
build_details = self._vf._index.get_build_details(this_iteration)
3426
build_details = self._knit._index.get_build_details(this_iteration)
3407
3427
self._all_build_details.update(build_details)
3408
# new_nodes = self._vf._index._get_entries(this_iteration)
3428
# new_nodes = self._knit._index._get_entries(this_iteration)
3409
3429
pending = set()
3410
3430
for key, details in build_details.iteritems():
3411
(index_memo, compression_parent, parent_keys,
3431
(index_memo, compression_parent, parents,
3412
3432
record_details) = details
3413
self._parent_map[key] = parent_keys
3414
self._heads_provider = None
3433
self._revision_id_graph[key] = parents
3415
3434
records.append((key, index_memo))
3416
3435
# Do we actually need to check _annotated_lines?
3417
pending.update([p for p in parent_keys
3418
if p not in self._all_build_details])
3420
for parent_key in parent_keys:
3421
if parent_key in self._num_needed_children:
3422
self._num_needed_children[parent_key] += 1
3424
self._num_needed_children[parent_key] = 1
3436
pending.update(p for p in parents
3437
if p not in self._all_build_details)
3425
3438
if compression_parent:
3426
if compression_parent in self._num_compression_children:
3427
self._num_compression_children[compression_parent] += 1
3429
self._num_compression_children[compression_parent] = 1
3439
self._compression_children.setdefault(compression_parent,
3442
for parent in parents:
3443
self._annotate_children.setdefault(parent,
3445
num_gens = generation - kept_generation
3446
if ((num_gens >= self._generations_until_keep)
3447
and len(parents) > 1):
3448
kept_generation = generation
3449
self._nodes_to_keep_annotations.add(key)
3431
3451
missing_versions = this_iteration.difference(build_details.keys())
3432
if missing_versions:
3433
for key in missing_versions:
3434
if key in self._parent_map and key in self._text_cache:
3435
# We already have this text ready, we just need to
3436
# yield it later so we get it annotated
3438
parent_keys = self._parent_map[key]
3439
for parent_key in parent_keys:
3440
if parent_key in self._num_needed_children:
3441
self._num_needed_children[parent_key] += 1
3443
self._num_needed_children[parent_key] = 1
3444
pending.update([p for p in parent_keys
3445
if p not in self._all_build_details])
3447
raise errors.RevisionNotPresent(key, self._vf)
3452
self._ghosts.update(missing_versions)
3453
for missing_version in missing_versions:
3454
# add a key, no parents
3455
self._revision_id_graph[missing_version] = ()
3456
pending.discard(missing_version) # don't look for it
3457
if self._ghosts.intersection(self._compression_children):
3459
"We cannot have nodes which have a ghost compression parent:\n"
3461
"compression children: %r"
3462
% (self._ghosts, self._compression_children))
3463
# Cleanout anything that depends on a ghost so that we don't wait for
3464
# the ghost to show up
3465
for node in self._ghosts:
3466
if node in self._annotate_children:
3467
# We won't be building this node
3468
del self._annotate_children[node]
3448
3469
# Generally we will want to read the records in reverse order, because
3449
3470
# we find the parent nodes after the children
3450
3471
records.reverse()
3451
return records, ann_keys
3453
def _get_needed_texts(self, key, pb=None):
3454
# if True or len(self._vf._immediate_fallback_vfs) > 0:
3455
if len(self._vf._immediate_fallback_vfs) > 0:
3456
# If we have fallbacks, go to the generic path
3457
for v in annotate.Annotator._get_needed_texts(self, key, pb=pb):
3462
records, ann_keys = self._get_build_graph(key)
3463
for idx, (sub_key, text, num_lines) in enumerate(
3464
self._extract_texts(records)):
3466
pb.update('annotating', idx, len(records))
3467
yield sub_key, text, num_lines
3468
for sub_key in ann_keys:
3469
text = self._text_cache[sub_key]
3470
num_lines = len(text) # bad assumption
3471
yield sub_key, text, num_lines
3473
except errors.RetryWithNewPacks, e:
3474
self._vf._access.reload_or_raise(e)
3475
# The cached build_details are no longer valid
3476
self._all_build_details.clear()
3478
def _cache_delta_blocks(self, key, compression_parent, delta, lines):
3479
parent_lines = self._text_cache[compression_parent]
3480
blocks = list(KnitContent.get_line_delta_blocks(delta, parent_lines, lines))
3481
self._matching_blocks[(key, compression_parent)] = blocks
3483
def _expand_record(self, key, parent_keys, compression_parent, record,
3486
if compression_parent:
3487
if compression_parent not in self._content_objects:
3488
# Waiting for the parent
3489
self._pending_deltas.setdefault(compression_parent, []).append(
3490
(key, parent_keys, record, record_details))
3492
# We have the basis parent, so expand the delta
3493
num = self._num_compression_children[compression_parent]
3496
base_content = self._content_objects.pop(compression_parent)
3497
self._num_compression_children.pop(compression_parent)
3499
self._num_compression_children[compression_parent] = num
3500
base_content = self._content_objects[compression_parent]
3501
# It is tempting to want to copy_base_content=False for the last
3502
# child object. However, whenever noeol=False,
3503
# self._text_cache[parent_key] is content._lines. So mutating it
3504
# gives very bad results.
3505
# The alternative is to copy the lines into text cache, but then we
3506
# are copying anyway, so just do it here.
3507
content, delta = self._vf._factory.parse_record(
3508
key, record, record_details, base_content,
3509
copy_base_content=True)
3512
content, _ = self._vf._factory.parse_record(
3513
key, record, record_details, None)
3514
if self._num_compression_children.get(key, 0) > 0:
3515
self._content_objects[key] = content
3516
lines = content.text()
3517
self._text_cache[key] = lines
3518
if delta is not None:
3519
self._cache_delta_blocks(key, compression_parent, delta, lines)
3522
def _get_parent_annotations_and_matches(self, key, text, parent_key):
3523
"""Get the list of annotations for the parent, and the matching lines.
3525
:param text: The opaque value given by _get_needed_texts
3526
:param parent_key: The key for the parent text
3527
:return: (parent_annotations, matching_blocks)
3528
parent_annotations is a list as long as the number of lines in
3530
matching_blocks is a list of (parent_idx, text_idx, len) tuples
3531
indicating which lines match between the two texts
3533
block_key = (key, parent_key)
3534
if block_key in self._matching_blocks:
3535
blocks = self._matching_blocks.pop(block_key)
3536
parent_annotations = self._annotations_cache[parent_key]
3537
return parent_annotations, blocks
3538
return annotate.Annotator._get_parent_annotations_and_matches(self,
3539
key, text, parent_key)
3541
def _process_pending(self, key):
3542
"""The content for 'key' was just processed.
3544
Determine if there is any more pending work to be processed.
3547
if key in self._pending_deltas:
3548
compression_parent = key
3549
children = self._pending_deltas.pop(key)
3550
for child_key, parent_keys, record, record_details in children:
3551
lines = self._expand_record(child_key, parent_keys,
3553
record, record_details)
3554
if self._check_ready_for_annotations(child_key, parent_keys):
3555
to_return.append(child_key)
3556
# Also check any children that are waiting for this parent to be
3558
if key in self._pending_annotation:
3559
children = self._pending_annotation.pop(key)
3560
to_return.extend([c for c, p_keys in children
3561
if self._check_ready_for_annotations(c, p_keys)])
3564
def _check_ready_for_annotations(self, key, parent_keys):
3565
"""return true if this text is ready to be yielded.
3567
Otherwise, this will return False, and queue the text into
3568
self._pending_annotation
3570
for parent_key in parent_keys:
3571
if parent_key not in self._annotations_cache:
3572
# still waiting on at least one parent text, so queue it up
3573
# Note that if there are multiple parents, we need to wait
3575
self._pending_annotation.setdefault(parent_key,
3576
[]).append((key, parent_keys))
3580
def _extract_texts(self, records):
3581
"""Extract the various texts needed based on records"""
3474
def _annotate_records(self, records):
3475
"""Build the annotations for the listed records."""
3582
3476
# We iterate in the order read, rather than a strict order requested
3583
3477
# However, process what we can, and put off to the side things that
3584
3478
# still need parents, cleaning them up when those parents are
3587
# 1) As 'records' are read, see if we can expand these records into
3588
# Content objects (and thus lines)
3589
# 2) If a given line-delta is waiting on its compression parent, it
3590
# gets queued up into self._pending_deltas, otherwise we expand
3591
# it, and put it into self._text_cache and self._content_objects
3592
# 3) If we expanded the text, we will then check to see if all
3593
# parents have also been processed. If so, this text gets yielded,
3594
# else this record gets set aside into pending_annotation
3595
# 4) Further, if we expanded the text in (2), we will then check to
3596
# see if there are any children in self._pending_deltas waiting to
3597
# also be processed. If so, we go back to (2) for those
3598
# 5) Further again, if we yielded the text, we can then check if that
3599
# 'unlocks' any of the texts in pending_annotations, which should
3600
# then get yielded as well
3601
# Note that both steps 4 and 5 are 'recursive' in that unlocking one
3602
# compression child could unlock yet another, and yielding a fulltext
3603
# will also 'unlock' the children that are waiting on that annotation.
3604
# (Though also, unlocking 1 parent's fulltext, does not unlock a child
3605
# if other parents are also waiting.)
3606
# We want to yield content before expanding child content objects, so
3607
# that we know when we can re-use the content lines, and the annotation
3608
# code can know when it can stop caching fulltexts, as well.
3610
# Children that are missing their compression parent
3612
for (key, record, digest) in self._vf._read_records_iter(records):
3614
details = self._all_build_details[key]
3615
(_, compression_parent, parent_keys, record_details) = details
3616
lines = self._expand_record(key, parent_keys, compression_parent,
3617
record, record_details)
3619
# Pending delta should be queued up
3480
for (rev_id, record,
3481
digest) in self._knit._read_records_iter(records):
3482
if rev_id in self._annotated_lines:
3621
# At this point, we may be able to yield this content, if all
3622
# parents are also finished
3623
yield_this_text = self._check_ready_for_annotations(key,
3626
# All parents present
3627
yield key, lines, len(lines)
3628
to_process = self._process_pending(key)
3630
this_process = to_process
3632
for key in this_process:
3633
lines = self._text_cache[key]
3634
yield key, lines, len(lines)
3635
to_process.extend(self._process_pending(key))
3484
parent_ids = self._revision_id_graph[rev_id]
3485
parent_ids = [p for p in parent_ids if p not in self._ghosts]
3486
details = self._all_build_details[rev_id]
3487
(index_memo, compression_parent, parents,
3488
record_details) = details
3489
nodes_to_annotate = []
3490
# TODO: Remove the punning between compression parents, and
3491
# parent_ids, we should be able to do this without assuming
3493
if len(parent_ids) == 0:
3494
# There are no parents for this node, so just add it
3495
# TODO: This probably needs to be decoupled
3496
fulltext_content, delta = self._knit._factory.parse_record(
3497
rev_id, record, record_details, None)
3498
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
3499
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
3500
parent_ids, left_matching_blocks=None))
3502
child = (rev_id, parent_ids, record)
3503
# Check if all the parents are present
3504
self._check_parents(child, nodes_to_annotate)
3505
while nodes_to_annotate:
3506
# Should we use a queue here instead of a stack?
3507
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
3508
(index_memo, compression_parent, parents,
3509
record_details) = self._all_build_details[rev_id]
3511
if compression_parent is not None:
3512
comp_children = self._compression_children[compression_parent]
3513
if rev_id not in comp_children:
3514
raise AssertionError("%r not in compression children %r"
3515
% (rev_id, comp_children))
3516
# If there is only 1 child, it is safe to reuse this
3518
reuse_content = (len(comp_children) == 1
3519
and compression_parent not in
3520
self._nodes_to_keep_annotations)
3522
# Remove it from the cache since it will be changing
3523
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
3524
# Make sure to copy the fulltext since it might be
3526
parent_fulltext = list(parent_fulltext_content.text())
3528
parent_fulltext_content = self._fulltext_contents[compression_parent]
3529
parent_fulltext = parent_fulltext_content.text()
3530
comp_children.remove(rev_id)
3531
fulltext_content, delta = self._knit._factory.parse_record(
3532
rev_id, record, record_details,
3533
parent_fulltext_content,
3534
copy_base_content=(not reuse_content))
3535
fulltext = self._add_fulltext_content(rev_id,
3537
if compression_parent == parent_ids[0]:
3538
# the compression_parent is the left parent, so we can
3540
blocks = KnitContent.get_line_delta_blocks(delta,
3541
parent_fulltext, fulltext)
3543
fulltext_content = self._knit._factory.parse_fulltext(
3545
fulltext = self._add_fulltext_content(rev_id,
3547
nodes_to_annotate.extend(
3548
self._add_annotation(rev_id, fulltext, parent_ids,
3549
left_matching_blocks=blocks))
3551
def _get_heads_provider(self):
3552
"""Create a heads provider for resolving ancestry issues."""
3553
if self._heads_provider is not None:
3554
return self._heads_provider
3555
parent_provider = _mod_graph.DictParentsProvider(
3556
self._revision_id_graph)
3557
graph_obj = _mod_graph.Graph(parent_provider)
3558
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3559
self._heads_provider = head_cache
3562
def annotate(self, key):
3563
"""Return the annotated fulltext at the given key.
3565
:param key: The key to annotate.
3567
if len(self._knit._fallback_vfs) > 0:
3568
# stacked knits can't use the fast path at present.
3569
return self._simple_annotate(key)
3572
records = self._get_build_graph(key)
3573
if key in self._ghosts:
3574
raise errors.RevisionNotPresent(key, self._knit)
3575
self._annotate_records(records)
3576
return self._annotated_lines[key]
3577
except errors.RetryWithNewPacks, e:
3578
self._knit._access.reload_or_raise(e)
3579
# The cached build_details are no longer valid
3580
self._all_build_details.clear()
3582
def _simple_annotate(self, key):
3583
"""Return annotated fulltext, rediffing from the full texts.
3585
This is slow but makes no assumptions about the repository
3586
being able to produce line deltas.
3588
# TODO: this code generates a parent maps of present ancestors; it
3589
# could be split out into a separate method, and probably should use
3590
# iter_ancestry instead. -- mbp and robertc 20080704
3591
graph = _mod_graph.Graph(self._knit)
3592
head_cache = _mod_graph.FrozenHeadsCache(graph)
3593
search = graph._make_breadth_first_searcher([key])
3597
present, ghosts = search.next_with_ghosts()
3598
except StopIteration:
3600
keys.update(present)
3601
parent_map = self._knit.get_parent_map(keys)
3603
reannotate = annotate.reannotate
3604
for record in self._knit.get_record_stream(keys, 'topological', True):
3606
fulltext = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
3607
parents = parent_map[key]
3608
if parents is not None:
3609
parent_lines = [parent_cache[parent] for parent in parent_map[key]]
3612
parent_cache[key] = list(
3613
reannotate(parent_lines, fulltext, key, None, head_cache))
3615
return parent_cache[key]
3617
raise errors.RevisionNotPresent(key, self._knit)
3638
from bzrlib._knit_load_data_pyx import _load_data_c as _load_data
3639
except ImportError, e:
3640
osutils.failed_to_load_extension(e)
3621
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3641
3623
from bzrlib._knit_load_data_py import _load_data_py as _load_data