3338
3339
annotator = _KnitAnnotator(knit)
3339
return iter(annotator.annotate(revision_id))
3342
class _KnitAnnotator(object):
3340
return iter(annotator.annotate_flat(revision_id))
3343
class _KnitAnnotator(annotate.Annotator):
3343
3344
"""Build up the annotations for a text."""
3345
def __init__(self, knit):
3348
# Content objects, differs from fulltexts because of how final newlines
3349
# are treated by knits. the content objects here will always have a
3351
self._fulltext_contents = {}
3353
# Annotated lines of specific revisions
3354
self._annotated_lines = {}
3356
# Track the raw data for nodes that we could not process yet.
3357
# This maps the revision_id of the base to a list of children that will
3358
# annotated from it.
3359
self._pending_children = {}
3361
# Nodes which cannot be extracted
3362
self._ghosts = set()
3364
# Track how many children this node has, so we know if we need to keep
3366
self._annotate_children = {}
3367
self._compression_children = {}
3346
def __init__(self, vf):
3347
annotate.Annotator.__init__(self, vf)
3349
# TODO: handle Nodes which cannot be extracted
3350
# self._ghosts = set()
3352
# Map from (key, parent_key) => matching_blocks, should be 'use once'
3353
self._matching_blocks = {}
3355
# KnitContent objects
3356
self._content_objects = {}
3357
# The number of children that depend on this fulltext content object
3358
self._num_compression_children = {}
3359
# Delta records that need their compression parent before they can be
3361
self._pending_deltas = {}
3362
# Fulltext records that are waiting for their parents fulltexts before
3363
# they can be yielded for annotation
3364
self._pending_annotation = {}
3369
3366
self._all_build_details = {}
3370
# The children => parent revision_id graph
3371
self._revision_id_graph = {}
3373
self._heads_provider = None
3375
self._nodes_to_keep_annotations = set()
3376
self._generations_until_keep = 100
3378
def set_generations_until_keep(self, value):
3379
"""Set the number of generations before caching a node.
3381
Setting this to -1 will cache every merge node, setting this higher
3382
will cache fewer nodes.
3384
self._generations_until_keep = value
3386
def _add_fulltext_content(self, revision_id, content_obj):
3387
self._fulltext_contents[revision_id] = content_obj
3388
# TODO: jam 20080305 It might be good to check the sha1digest here
3389
return content_obj.text()
3391
def _check_parents(self, child, nodes_to_annotate):
3392
"""Check if all parents have been processed.
3394
:param child: A tuple of (rev_id, parents, raw_content)
3395
:param nodes_to_annotate: If child is ready, add it to
3396
nodes_to_annotate, otherwise put it back in self._pending_children
3398
for parent_id in child[1]:
3399
if (parent_id not in self._annotated_lines):
3400
# This parent is present, but another parent is missing
3401
self._pending_children.setdefault(parent_id,
3405
# This one is ready to be processed
3406
nodes_to_annotate.append(child)
3408
def _add_annotation(self, revision_id, fulltext, parent_ids,
3409
left_matching_blocks=None):
3410
"""Add an annotation entry.
3412
All parents should already have been annotated.
3413
:return: A list of children that now have their parents satisfied.
3415
a = self._annotated_lines
3416
annotated_parent_lines = [a[p] for p in parent_ids]
3417
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
3418
fulltext, revision_id, left_matching_blocks,
3419
heads_provider=self._get_heads_provider()))
3420
self._annotated_lines[revision_id] = annotated_lines
3421
for p in parent_ids:
3422
ann_children = self._annotate_children[p]
3423
ann_children.remove(revision_id)
3424
if (not ann_children
3425
and p not in self._nodes_to_keep_annotations):
3426
del self._annotated_lines[p]
3427
del self._all_build_details[p]
3428
if p in self._fulltext_contents:
3429
del self._fulltext_contents[p]
3430
# Now that we've added this one, see if there are any pending
3431
# deltas to be done, certainly this parent is finished
3432
nodes_to_annotate = []
3433
for child in self._pending_children.pop(revision_id, []):
3434
self._check_parents(child, nodes_to_annotate)
3435
return nodes_to_annotate
3437
3368
def _get_build_graph(self, key):
3438
3369
"""Get the graphs for building texts and annotations.
3446
3377
passing to read_records_iter to start reading in the raw data from
3449
if key in self._annotated_lines:
3452
3380
pending = set([key])
3383
self._num_needed_children[key] = 1
3457
3385
# get all pending nodes
3459
3386
this_iteration = pending
3460
build_details = self._knit._index.get_build_details(this_iteration)
3387
build_details = self._vf._index.get_build_details(this_iteration)
3461
3388
self._all_build_details.update(build_details)
3462
# new_nodes = self._knit._index._get_entries(this_iteration)
3389
# new_nodes = self._vf._index._get_entries(this_iteration)
3463
3390
pending = set()
3464
3391
for key, details in build_details.iteritems():
3465
(index_memo, compression_parent, parents,
3392
(index_memo, compression_parent, parent_keys,
3466
3393
record_details) = details
3467
self._revision_id_graph[key] = parents
3394
self._parent_map[key] = parent_keys
3395
self._heads_provider = None
3468
3396
records.append((key, index_memo))
3469
3397
# Do we actually need to check _annotated_lines?
3470
pending.update(p for p in parents
3471
if p not in self._all_build_details)
3398
pending.update([p for p in parent_keys
3399
if p not in self._all_build_details])
3401
for parent_key in parent_keys:
3402
if parent_key in self._num_needed_children:
3403
self._num_needed_children[parent_key] += 1
3405
self._num_needed_children[parent_key] = 1
3472
3406
if compression_parent:
3473
self._compression_children.setdefault(compression_parent,
3476
for parent in parents:
3477
self._annotate_children.setdefault(parent,
3479
num_gens = generation - kept_generation
3480
if ((num_gens >= self._generations_until_keep)
3481
and len(parents) > 1):
3482
kept_generation = generation
3483
self._nodes_to_keep_annotations.add(key)
3407
if compression_parent in self._num_compression_children:
3408
self._num_compression_children[compression_parent] += 1
3410
self._num_compression_children[compression_parent] = 1
3485
3412
missing_versions = this_iteration.difference(build_details.keys())
3486
self._ghosts.update(missing_versions)
3487
for missing_version in missing_versions:
3488
# add a key, no parents
3489
self._revision_id_graph[missing_version] = ()
3490
pending.discard(missing_version) # don't look for it
3491
if self._ghosts.intersection(self._compression_children):
3493
"We cannot have nodes which have a ghost compression parent:\n"
3495
"compression children: %r"
3496
% (self._ghosts, self._compression_children))
3497
# Cleanout anything that depends on a ghost so that we don't wait for
3498
# the ghost to show up
3499
for node in self._ghosts:
3500
if node in self._annotate_children:
3501
# We won't be building this node
3502
del self._annotate_children[node]
3413
if missing_versions:
3414
for key in missing_versions:
3415
if key in self._parent_map and key in self._text_cache:
3416
# We already have this text ready, we just need to
3417
# yield it later so we get it annotated
3419
parent_keys = self._parent_map[key]
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
3425
pending.update([p for p in parent_keys
3426
if p not in self._all_build_details])
3428
raise errors.RevisionNotPresent(key, self._vf)
3503
3429
# Generally we will want to read the records in reverse order, because
3504
3430
# we find the parent nodes after the children
3505
3431
records.reverse()
3508
def _annotate_records(self, records):
3509
"""Build the annotations for the listed records."""
3432
return records, ann_keys
3434
def _get_needed_texts(self, key, pb=None):
3435
# if True or len(self._vf._fallback_vfs) > 0:
3436
if len(self._vf._fallback_vfs) > 0:
3437
# If we have fallbacks, go to the generic path
3438
for v in annotate.Annotator._get_needed_texts(self, key, pb=pb):
3443
records, ann_keys = self._get_build_graph(key)
3444
for idx, (sub_key, text, num_lines) in enumerate(
3445
self._extract_texts(records)):
3447
pb.update('annotating', idx, len(records))
3448
yield sub_key, text, num_lines
3449
for sub_key in ann_keys:
3450
text = self._text_cache[sub_key]
3451
num_lines = len(text) # bad assumption
3452
yield sub_key, text, num_lines
3454
except errors.RetryWithNewPacks, e:
3455
self._vf._access.reload_or_raise(e)
3456
# The cached build_details are no longer valid
3457
self._all_build_details.clear()
3459
def _cache_delta_blocks(self, key, compression_parent, delta, lines):
3460
parent_lines = self._text_cache[compression_parent]
3461
blocks = list(KnitContent.get_line_delta_blocks(delta, parent_lines, lines))
3462
self._matching_blocks[(key, compression_parent)] = blocks
3464
def _expand_record(self, key, parent_keys, compression_parent, record,
3467
if compression_parent:
3468
if compression_parent not in self._content_objects:
3469
# Waiting for the parent
3470
self._pending_deltas.setdefault(compression_parent, []).append(
3471
(key, parent_keys, record, record_details))
3473
# We have the basis parent, so expand the delta
3474
num = self._num_compression_children[compression_parent]
3477
base_content = self._content_objects.pop(compression_parent)
3478
self._num_compression_children.pop(compression_parent)
3480
self._num_compression_children[compression_parent] = num
3481
base_content = self._content_objects[compression_parent]
3482
# It is tempting to want to copy_base_content=False for the last
3483
# child object. However, whenever noeol=False,
3484
# self._text_cache[parent_key] is content._lines. So mutating it
3485
# gives very bad results.
3486
# The alternative is to copy the lines into text cache, but then we
3487
# are copying anyway, so just do it here.
3488
content, delta = self._vf._factory.parse_record(
3489
key, record, record_details, base_content,
3490
copy_base_content=True)
3493
content, _ = self._vf._factory.parse_record(
3494
key, record, record_details, None)
3495
if self._num_compression_children.get(key, 0) > 0:
3496
self._content_objects[key] = content
3497
lines = content.text()
3498
self._text_cache[key] = lines
3499
if delta is not None:
3500
self._cache_delta_blocks(key, compression_parent, delta, lines)
3503
def _get_parent_annotations_and_matches(self, key, text, parent_key):
3504
"""Get the list of annotations for the parent, and the matching lines.
3506
:param text: The opaque value given by _get_needed_texts
3507
:param parent_key: The key for the parent text
3508
:return: (parent_annotations, matching_blocks)
3509
parent_annotations is a list as long as the number of lines in
3511
matching_blocks is a list of (parent_idx, text_idx, len) tuples
3512
indicating which lines match between the two texts
3514
block_key = (key, parent_key)
3515
if block_key in self._matching_blocks:
3516
blocks = self._matching_blocks.pop(block_key)
3517
parent_annotations = self._annotations_cache[parent_key]
3518
return parent_annotations, blocks
3519
return annotate.Annotator._get_parent_annotations_and_matches(self,
3520
key, text, parent_key)
3522
def _process_pending(self, key):
3523
"""The content for 'key' was just processed.
3525
Determine if there is any more pending work to be processed.
3528
if key in self._pending_deltas:
3529
compression_parent = key
3530
children = self._pending_deltas.pop(key)
3531
for child_key, parent_keys, record, record_details in children:
3532
lines = self._expand_record(child_key, parent_keys,
3534
record, record_details)
3535
if self._check_ready_for_annotations(child_key, parent_keys):
3536
to_return.append(child_key)
3537
# Also check any children that are waiting for this parent to be
3539
if key in self._pending_annotation:
3540
children = self._pending_annotation.pop(key)
3541
to_return.extend([c for c, p_keys in children
3542
if self._check_ready_for_annotations(c, p_keys)])
3545
def _check_ready_for_annotations(self, key, parent_keys):
3546
"""return true if this text is ready to be yielded.
3548
Otherwise, this will return False, and queue the text into
3549
self._pending_annotation
3551
for parent_key in parent_keys:
3552
if parent_key not in self._annotations_cache:
3553
# still waiting on at least one parent text, so queue it up
3554
# Note that if there are multiple parents, we need to wait
3556
self._pending_annotation.setdefault(parent_key,
3557
[]).append((key, parent_keys))
3561
def _extract_texts(self, records):
3562
"""Extract the various texts needed based on records"""
3510
3563
# We iterate in the order read, rather than a strict order requested
3511
3564
# However, process what we can, and put off to the side things that
3512
3565
# still need parents, cleaning them up when those parents are
3514
for (rev_id, record,
3515
digest) in self._knit._read_records_iter(records):
3516
if rev_id in self._annotated_lines:
3568
# 1) As 'records' are read, see if we can expand these records into
3569
# Content objects (and thus lines)
3570
# 2) If a given line-delta is waiting on its compression parent, it
3571
# gets queued up into self._pending_deltas, otherwise we expand
3572
# it, and put it into self._text_cache and self._content_objects
3573
# 3) If we expanded the text, we will then check to see if all
3574
# parents have also been processed. If so, this text gets yielded,
3575
# else this record gets set aside into pending_annotation
3576
# 4) Further, if we expanded the text in (2), we will then check to
3577
# see if there are any children in self._pending_deltas waiting to
3578
# also be processed. If so, we go back to (2) for those
3579
# 5) Further again, if we yielded the text, we can then check if that
3580
# 'unlocks' any of the texts in pending_annotations, which should
3581
# then get yielded as well
3582
# Note that both steps 4 and 5 are 'recursive' in that unlocking one
3583
# compression child could unlock yet another, and yielding a fulltext
3584
# will also 'unlock' the children that are waiting on that annotation.
3585
# (Though also, unlocking 1 parent's fulltext, does not unlock a child
3586
# if other parents are also waiting.)
3587
# We want to yield content before expanding child content objects, so
3588
# that we know when we can re-use the content lines, and the annotation
3589
# code can know when it can stop caching fulltexts, as well.
3591
# Children that are missing their compression parent
3593
for (key, record, digest) in self._vf._read_records_iter(records):
3595
details = self._all_build_details[key]
3596
(_, compression_parent, parent_keys, record_details) = details
3597
lines = self._expand_record(key, parent_keys, compression_parent,
3598
record, record_details)
3600
# Pending delta should be queued up
3518
parent_ids = self._revision_id_graph[rev_id]
3519
parent_ids = [p for p in parent_ids if p not in self._ghosts]
3520
details = self._all_build_details[rev_id]
3521
(index_memo, compression_parent, parents,
3522
record_details) = details
3523
nodes_to_annotate = []
3524
# TODO: Remove the punning between compression parents, and
3525
# parent_ids, we should be able to do this without assuming
3527
if len(parent_ids) == 0:
3528
# There are no parents for this node, so just add it
3529
# TODO: This probably needs to be decoupled
3530
fulltext_content, delta = self._knit._factory.parse_record(
3531
rev_id, record, record_details, None)
3532
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
3533
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
3534
parent_ids, left_matching_blocks=None))
3536
child = (rev_id, parent_ids, record)
3537
# Check if all the parents are present
3538
self._check_parents(child, nodes_to_annotate)
3539
while nodes_to_annotate:
3540
# Should we use a queue here instead of a stack?
3541
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
3542
(index_memo, compression_parent, parents,
3543
record_details) = self._all_build_details[rev_id]
3545
if compression_parent is not None:
3546
comp_children = self._compression_children[compression_parent]
3547
if rev_id not in comp_children:
3548
raise AssertionError("%r not in compression children %r"
3549
% (rev_id, comp_children))
3550
# If there is only 1 child, it is safe to reuse this
3552
reuse_content = (len(comp_children) == 1
3553
and compression_parent not in
3554
self._nodes_to_keep_annotations)
3556
# Remove it from the cache since it will be changing
3557
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
3558
# Make sure to copy the fulltext since it might be
3560
parent_fulltext = list(parent_fulltext_content.text())
3562
parent_fulltext_content = self._fulltext_contents[compression_parent]
3563
parent_fulltext = parent_fulltext_content.text()
3564
comp_children.remove(rev_id)
3565
fulltext_content, delta = self._knit._factory.parse_record(
3566
rev_id, record, record_details,
3567
parent_fulltext_content,
3568
copy_base_content=(not reuse_content))
3569
fulltext = self._add_fulltext_content(rev_id,
3571
if compression_parent == parent_ids[0]:
3572
# the compression_parent is the left parent, so we can
3574
blocks = KnitContent.get_line_delta_blocks(delta,
3575
parent_fulltext, fulltext)
3577
fulltext_content = self._knit._factory.parse_fulltext(
3579
fulltext = self._add_fulltext_content(rev_id,
3581
nodes_to_annotate.extend(
3582
self._add_annotation(rev_id, fulltext, parent_ids,
3583
left_matching_blocks=blocks))
3585
def _get_heads_provider(self):
3586
"""Create a heads provider for resolving ancestry issues."""
3587
if self._heads_provider is not None:
3588
return self._heads_provider
3589
self._heads_provider = _mod_graph.KnownGraph(self._revision_id_graph)
3590
return self._heads_provider
3592
def annotate(self, key):
3593
"""Return the annotated fulltext at the given key.
3595
:param key: The key to annotate.
3597
if len(self._knit._fallback_vfs) > 0:
3598
# stacked knits can't use the fast path at present.
3599
return self._simple_annotate(key)
3602
records = self._get_build_graph(key)
3603
if key in self._ghosts:
3604
raise errors.RevisionNotPresent(key, self._knit)
3605
self._annotate_records(records)
3606
return self._annotated_lines[key]
3607
except errors.RetryWithNewPacks, e:
3608
self._knit._access.reload_or_raise(e)
3609
# The cached build_details are no longer valid
3610
self._all_build_details.clear()
3612
def _simple_annotate(self, key):
3613
"""Return annotated fulltext, rediffing from the full texts.
3615
This is slow but makes no assumptions about the repository
3616
being able to produce line deltas.
3618
# TODO: this code generates a parent maps of present ancestors; it
3619
# could be split out into a separate method
3620
# -- mbp and robertc 20080704
3621
graph = _mod_graph.Graph(self._knit)
3622
parent_map = dict((k, v) for k, v in graph.iter_ancestry([key])
3625
raise errors.RevisionNotPresent(key, self)
3626
keys = parent_map.keys()
3627
heads_provider = _mod_graph.KnownGraph(parent_map)
3629
reannotate = annotate.reannotate
3630
for record in self._knit.get_record_stream(keys, 'topological', True):
3632
fulltext = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
3633
parents = parent_map[key]
3634
if parents is not None:
3635
parent_lines = [parent_cache[parent] for parent in parent_map[key]]
3638
parent_cache[key] = list(
3639
reannotate(parent_lines, fulltext, key, None, heads_provider))
3641
return parent_cache[key]
3643
raise errors.RevisionNotPresent(key, self._knit)
3602
# At this point, we may be able to yield this content, if all
3603
# parents are also finished
3604
yield_this_text = self._check_ready_for_annotations(key,
3607
# All parents present
3608
yield key, lines, len(lines)
3609
to_process = self._process_pending(key)
3611
this_process = to_process
3613
for key in this_process:
3614
lines = self._text_cache[key]
3615
yield key, lines, len(lines)
3616
to_process.extend(self._process_pending(key))
3647
3619
from bzrlib._knit_load_data_c import _load_data_c as _load_data
added
removed
bzrlib/knit.py
