« back to all changes in this revision
Viewing changes to bzrlib/repofmt/groupcompress_repo.py
- browse files at revision 3735.31.4
- compare with another revision
- compare with revision 6170
- download diff
- download tarball
- view history from revision 3735.31.4
- Committer: John Arbash Meinel
- Date: 2009-03-11 07:42:37 UTC
- mto: (3735.2.143 brisbane-core)
- mto: This revision was merged to the branch mainline in revision 4280.
- Revision ID: john@arbash-meinel.com-20090311074237-6lzsbd3r0tmbofjz
Continue trying to reduce the friction for the groupcompress repo formats.
added
removed
bzrlib/repofmt/groupcompress_repo.py
20
20
21
21
from bzrlib import (
22
22
chk_map,
23
chk_serializer,
23
24
debug,
24
25
errors,
25
26
inventory,
31
32
ui,
32
33
)
33
34
from bzrlib.index import GraphIndex, GraphIndexBuilder
34
from bzrlib.repository import InterPackRepo
35
35
from bzrlib.groupcompress import (
36
36
_GCGraphIndex,
37
37
GroupCompressVersionedFiles,
43
43
RepositoryPackCollection,
44
44
RepositoryFormatKnitPack6,
45
45
Packer,
46
)
47
try:
48
from bzrlib.repofmt.pack_repo import (
49
46
CHKInventoryRepository,
50
RepositoryFormatPackDevelopment5,
51
47
RepositoryFormatPackDevelopment5Hash16,
52
## RepositoryFormatPackDevelopment5Hash16b,
53
## RepositoryFormatPackDevelopment5Hash63,
54
## RepositoryFormatPackDevelopment5Hash127a,
55
## RepositoryFormatPackDevelopment5Hash127b,
56
48
RepositoryFormatPackDevelopment5Hash255,
57
49
)
58
from bzrlib import
59
chk_support = True
60
except ImportError:
61
chk_support = False
62
50
63
51
64
52
def open_pack(self):
89
77
# The relative locations of the packs are constrained, but all are
90
78
# passed in because the caller has them, so as to avoid object churn.
91
79
index_builder_class = pack_collection._index_builder_class
92
if chk_support:
93
# from brisbane-core
94
if pack_collection.chk_index is not None:
95
chk_index = index_builder_class(reference_lists=0)
96
else:
97
chk_index = None
98
Pack.__init__(self,
99
# Revisions: parents list, no text compression.
100
index_builder_class(reference_lists=1),
101
# Inventory: We want to map compression only, but currently the
102
# knit code hasn't been updated enough to understand that, so we
103
# have a regular 2-list index giving parents and compression
104
# source.
105
index_builder_class(reference_lists=1),
106
# Texts: compression and per file graph, for all fileids - so two
107
# reference lists and two elements in the key tuple.
108
index_builder_class(reference_lists=1, key_elements=2),
109
# Signatures: Just blobs to store, no compression, no parents
110
# listing.
111
index_builder_class(reference_lists=0),
112
# CHK based storage - just blobs, no compression or parents.
113
chk_index=chk_index
114
)
80
# from brisbane-core
81
if pack_collection.chk_index is not None:
82
chk_index = index_builder_class(reference_lists=0)
115
83
else:
116
# from bzr.dev
117
Pack.__init__(self,
118
# Revisions: parents list, no text compression.
119
index_builder_class(reference_lists=1),
120
# Inventory: compressed, with graph for compatibility with other
121
# existing bzrlib code.
122
index_builder_class(reference_lists=1),
123
# Texts: per file graph:
124
index_builder_class(reference_lists=1, key_elements=2),
125
# Signatures: Just blobs to store, no compression, no parents
126
# listing.
127
index_builder_class(reference_lists=0),
128
)
84
chk_index = None
85
Pack.__init__(self,
86
# Revisions: parents list, no text compression.
87
index_builder_class(reference_lists=1),
88
# Inventory: We want to map compression only, but currently the
89
# knit code hasn't been updated enough to understand that, so we
90
# have a regular 2-list index giving parents and compression
91
# source.
92
index_builder_class(reference_lists=1),
93
# Texts: compression and per file graph, for all fileids - so two
94
# reference lists and two elements in the key tuple.
95
index_builder_class(reference_lists=1, key_elements=2),
96
# Signatures: Just blobs to store, no compression, no parents
97
# listing.
98
index_builder_class(reference_lists=0),
99
# CHK based storage - just blobs, no compression or parents.
100
chk_index=chk_index
101
)
129
102
self._pack_collection = pack_collection
130
103
# When we make readonly indices, we need this.
131
104
self.index_class = pack_collection._index_class
208
181
self._new_pack)
209
182
self.signature_index.add_writable_index(self._new_pack.signature_index,
210
183
self._new_pack)
211
if chk_support and self.chk_index is not None:
184
if self.chk_index is not None:
212
185
self.chk_index.add_writable_index(self._new_pack.chk_index,
213
186
self._new_pack)
214
187
self.repo.chk_bytes._index._add_callback = self.chk_index.add_callback
426
399
class GCPackRepository(KnitPackRepository):
427
400
"""GC customisation of KnitPackRepository."""
428
401
429
# Note: I think the CHK support can be dropped from this class as it's
430
# implemented via the GCCHKPackRepository class defined next. IGC 20090301
431
432
402
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
433
403
_serializer):
434
404
"""Overridden to change pack collection class."""
436
406
_commit_builder_class, _serializer)
437
407
# and now replace everything it did :)
438
408
index_transport = self._transport.clone('indices')
439
if chk_support:
440
self._pack_collection = GCRepositoryPackCollection(self,
441
self._transport, index_transport,
442
self._transport.clone('upload'),
443
self._transport.clone('packs'),
444
_format.index_builder_class,
445
_format.index_class,
446
use_chk_index=self._format.supports_chks,
447
)
448
else:
449
self._pack_collection = GCRepositoryPackCollection(self,
450
self._transport, index_transport,
451
self._transport.clone('upload'),
452
self._transport.clone('packs'),
453
_format.index_builder_class,
454
_format.index_class)
409
self._pack_collection = GCRepositoryPackCollection(self,
410
self._transport, index_transport,
411
self._transport.clone('upload'),
412
self._transport.clone('packs'),
413
_format.index_builder_class,
414
_format.index_class,
415
use_chk_index=self._format.supports_chks,
416
)
455
417
self.inventories = GroupCompressVersionedFiles(
456
418
_GCGraphIndex(self._pack_collection.inventory_index.combined_index,
457
419
add_callback=self._pack_collection.inventory_index.add_callback,
474
436
add_callback=self._pack_collection.text_index.add_callback,
475
437
parents=True, is_locked=self.is_locked),
476
438
access=self._pack_collection.text_index.data_access)
477
if chk_support and _format.supports_chks:
439
if _format.supports_chks:
478
440
# No graph, no compression:- references from chks are between
479
441
# different objects not temporal versions of the same; and without
480
442
# some sort of temporal structure knit compression will just fail.
498
460
self._reconcile_backsup_inventory = False
499
461
500
462
501
if chk_support:
502
class GCCHKPackRepository(CHKInventoryRepository):
503
"""GC customisation of CHKInventoryRepository."""
463
class GCCHKPackRepository(CHKInventoryRepository):
464
"""GC customisation of CHKInventoryRepository."""
504
465
505
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
506
_serializer):
507
"""Overridden to change pack collection class."""
508
KnitPackRepository.__init__(self, _format, a_bzrdir, control_files,
509
_commit_builder_class, _serializer)
510
# and now replace everything it did :)
511
index_transport = self._transport.clone('indices')
512
self._pack_collection = GCRepositoryPackCollection(self,
513
self._transport, index_transport,
514
self._transport.clone('upload'),
515
self._transport.clone('packs'),
516
_format.index_builder_class,
517
_format.index_class,
518
use_chk_index=self._format.supports_chks,
519
)
520
self.inventories = GroupCompressVersionedFiles(
521
_GCGraphIndex(self._pack_collection.inventory_index.combined_index,
522
add_callback=self._pack_collection.inventory_index.add_callback,
523
parents=True, is_locked=self.is_locked),
524
access=self._pack_collection.inventory_index.data_access)
525
self.revisions = GroupCompressVersionedFiles(
526
_GCGraphIndex(self._pack_collection.revision_index.combined_index,
527
add_callback=self._pack_collection.revision_index.add_callback,
528
parents=True, is_locked=self.is_locked),
529
access=self._pack_collection.revision_index.data_access,
530
delta=False)
531
self.signatures = GroupCompressVersionedFiles(
532
_GCGraphIndex(self._pack_collection.signature_index.combined_index,
533
add_callback=self._pack_collection.signature_index.add_callback,
534
parents=False, is_locked=self.is_locked),
535
access=self._pack_collection.signature_index.data_access,
536
delta=False)
537
self.texts = GroupCompressVersionedFiles(
538
_GCGraphIndex(self._pack_collection.text_index.combined_index,
539
add_callback=self._pack_collection.text_index.add_callback,
540
parents=True, is_locked=self.is_locked),
541
access=self._pack_collection.text_index.data_access)
542
# No parents, individual CHK pages don't have specific ancestry
543
self.chk_bytes = GroupCompressVersionedFiles(
544
_GCGraphIndex(self._pack_collection.chk_index.combined_index,
545
add_callback=self._pack_collection.chk_index.add_callback,
546
parents=False, is_locked=self.is_locked),
547
access=self._pack_collection.chk_index.data_access)
548
# True when the repository object is 'write locked' (as opposed to the
549
# physical lock only taken out around changes to the pack-names list.)
550
# Another way to represent this would be a decorator around the control
551
# files object that presents logical locks as physical ones - if this
552
# gets ugly consider that alternative design. RBC 20071011
553
self._write_lock_count = 0
554
self._transaction = None
555
# for tests
556
self._reconcile_does_inventory_gc = True
557
self._reconcile_fixes_text_parents = True
558
self._reconcile_backsup_inventory = False
466
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
467
_serializer):
468
"""Overridden to change pack collection class."""
469
KnitPackRepository.__init__(self, _format, a_bzrdir, control_files,
470
_commit_builder_class, _serializer)
471
# and now replace everything it did :)
472
index_transport = self._transport.clone('indices')
473
self._pack_collection = GCRepositoryPackCollection(self,
474
self._transport, index_transport,
475
self._transport.clone('upload'),
476
self._transport.clone('packs'),
477
_format.index_builder_class,
478
_format.index_class,
479
use_chk_index=self._format.supports_chks,
480
)
481
self.inventories = GroupCompressVersionedFiles(
482
_GCGraphIndex(self._pack_collection.inventory_index.combined_index,
483
add_callback=self._pack_collection.inventory_index.add_callback,
484
parents=True, is_locked=self.is_locked),
485
access=self._pack_collection.inventory_index.data_access)
486
self.revisions = GroupCompressVersionedFiles(
487
_GCGraphIndex(self._pack_collection.revision_index.combined_index,
488
add_callback=self._pack_collection.revision_index.add_callback,
489
parents=True, is_locked=self.is_locked),
490
access=self._pack_collection.revision_index.data_access,
491
delta=False)
492
self.signatures = GroupCompressVersionedFiles(
493
_GCGraphIndex(self._pack_collection.signature_index.combined_index,
494
add_callback=self._pack_collection.signature_index.add_callback,
495
parents=False, is_locked=self.is_locked),
496
access=self._pack_collection.signature_index.data_access,
497
delta=False)
498
self.texts = GroupCompressVersionedFiles(
499
_GCGraphIndex(self._pack_collection.text_index.combined_index,
500
add_callback=self._pack_collection.text_index.add_callback,
501
parents=True, is_locked=self.is_locked),
502
access=self._pack_collection.text_index.data_access)
503
# No parents, individual CHK pages don't have specific ancestry
504
self.chk_bytes = GroupCompressVersionedFiles(
505
_GCGraphIndex(self._pack_collection.chk_index.combined_index,
506
add_callback=self._pack_collection.chk_index.add_callback,
507
parents=False, is_locked=self.is_locked),
508
access=self._pack_collection.chk_index.data_access)
509
# True when the repository object is 'write locked' (as opposed to the
510
# physical lock only taken out around changes to the pack-names list.)
511
# Another way to represent this would be a decorator around the control
512
# files object that presents logical locks as physical ones - if this
513
# gets ugly consider that alternative design. RBC 20071011
514
self._write_lock_count = 0
515
self._transaction = None
516
# for tests
517
self._reconcile_does_inventory_gc = True
518
self._reconcile_fixes_text_parents = True
519
self._reconcile_backsup_inventory = False
559
520
560
521
561
522
class RepositoryFormatPackGCPlain(RepositoryFormatKnitPack6):
570
531
# multiple in-a-row (and sharing strings). Topological is better
571
532
# for remote, because we access less data.
572
533
_fetch_order = 'unordered'
573
_fetch_gc_optimal = True
574
534
_fetch_uses_deltas = False
575
535
576
536
def get_format_string(self):
584
544
", interoperates with pack-0.92\n")
585
545
586
546
587
if chk_support:
588
from bzrlib import chk_serializer
589
class RepositoryFormatPackGCCHK16(RepositoryFormatPackDevelopment5Hash16):
590
"""A hashed CHK+group compress pack repository."""
591
592
repository_class = GCCHKPackRepository
593
rich_root_data = True
594
# Note: We cannot unpack a delta that references a text we haven't
595
# seen yet. There are 2 options, work in fulltexts, or require
596
# topological sorting. Using fulltexts is more optimal for local
597
# operations, because the source can be smart about extracting
598
# multiple in-a-row (and sharing strings). Topological is better
599
# for remote, because we access less data.
600
_fetch_order = 'unordered'
601
_fetch_gc_optimal = True
602
_fetch_uses_deltas = False
603
604
def get_format_string(self):
605
"""See RepositoryFormat.get_format_string()."""
606
return ('Bazaar development format - hash16chk+gc rich-root'
607
' (needs bzr.dev from 1.13)\n')
608
609
def get_format_description(self):
610
"""See RepositoryFormat.get_format_description()."""
611
return ("Development repository format - hash16chk+groupcompress")
612
613
614
class RepositoryFormatPackGCCHK255(RepositoryFormatPackDevelopment5Hash255):
615
"""A hashed CHK+group compress pack repository."""
616
617
repository_class = GCCHKPackRepository
618
# Setting this to True causes us to use InterModel1And2, so for now set
619
# it to False which uses InterDifferingSerializer. When IM1&2 is
620
# removed (as it is in bzr.dev) we can set this back to True.
621
rich_root_data = True
622
623
def get_format_string(self):
624
"""See RepositoryFormat.get_format_string()."""
625
return ('Bazaar development format - hash255chk+gc rich-root'
626
' (needs bzr.dev from 1.13)\n')
627
628
def get_format_description(self):
629
"""See RepositoryFormat.get_format_description()."""
630
return ("Development repository format - hash255chk+groupcompress")
631
632
633
chk_serializer_255_bigpage = chk_serializer.CHKSerializer(65536, 'hash-255-way')
634
class RepositoryFormatPackGCCHK255Big(RepositoryFormatPackGCCHK255):
635
"""A hashed CHK+group compress pack repository."""
636
637
repository_class = GCCHKPackRepository
638
# For right now, setting this to True gives us InterModel1And2 rather
639
# than InterDifferingSerializer
640
rich_root_data = True
641
_serializer = chk_serializer_255_bigpage
642
# Note: We cannot unpack a delta that references a text we haven't
643
# seen yet. There are 2 options, work in fulltexts, or require
644
# topological sorting. Using fulltexts is more optimal for local
645
# operations, because the source can be smart about extracting
646
# multiple in-a-row (and sharing strings). Topological is better
647
# for remote, because we access less data.
648
_fetch_order = 'unordered'
649
_fetch_gc_optimal = True
650
_fetch_uses_deltas = False
651
652
def get_format_string(self):
653
"""See RepositoryFormat.get_format_string()."""
654
return ('Bazaar development format - hash255chk+gc rich-root bigpage'
655
' (needs bzr.dev from 1.13)\n')
656
657
def get_format_description(self):
658
"""See RepositoryFormat.get_format_description()."""
659
return ("Development repository format - hash255chk+groupcompress + bigpage")
660
661
662
def pack_incompatible(source, target, orig_method=InterPackRepo.is_compatible):
663
"""Be incompatible with the regular fetch code."""
664
formats = (RepositoryFormatPackGCPlain,)
665
if chk_support:
666
formats = formats + (RepositoryFormatPackGCCHK16,
667
RepositoryFormatPackGCCHK255,
668
RepositoryFormatPackGCCHK255Big)
669
if isinstance(source._format, formats) or isinstance(target._format, formats):
670
return False
671
else:
672
return orig_method(source, target)
673
674
675
InterPackRepo.is_compatible = staticmethod(pack_incompatible)
547
class RepositoryFormatPackGCCHK16(RepositoryFormatPackDevelopment5Hash16):
548
"""A hashed CHK+group compress pack repository."""
549
550
repository_class = GCCHKPackRepository
551
rich_root_data = True
552
supports_external_lookups = True
553
supports_tree_reference = True
554
supports_chks = True
555
# Note: We cannot unpack a delta that references a text we haven't
556
# seen yet. There are 2 options, work in fulltexts, or require
557
# topological sorting. Using fulltexts is more optimal for local
558
# operations, because the source can be smart about extracting
559
# multiple in-a-row (and sharing strings). Topological is better
560
# for remote, because we access less data.
561
_fetch_order = 'unordered'
562
_fetch_uses_deltas = False
563
564
def get_format_string(self):
565
"""See RepositoryFormat.get_format_string()."""
566
return ('Bazaar development format - hash16chk+gc rich-root'
567
' (needs bzr.dev from 1.13)\n')
568
569
def get_format_description(self):
570
"""See RepositoryFormat.get_format_description()."""
571
return ("Development repository format - hash16chk+groupcompress")
572
573
574
class RepositoryFormatPackGCCHK255(RepositoryFormatPackDevelopment5Hash255):
575
"""A hashed CHK+group compress pack repository."""
576
577
repository_class = GCCHKPackRepository
578
supports_chks = True
579
# Setting this to True causes us to use InterModel1And2, so for now set
580
# it to False which uses InterDifferingSerializer. When IM1&2 is
581
# removed (as it is in bzr.dev) we can set this back to True.
582
rich_root_data = True
583
584
def get_format_string(self):
585
"""See RepositoryFormat.get_format_string()."""
586
return ('Bazaar development format - hash255chk+gc rich-root'
587
' (needs bzr.dev from 1.13)\n')
588
589
def get_format_description(self):
590
"""See RepositoryFormat.get_format_description()."""
591
return ("Development repository format - hash255chk+groupcompress")
592
593
594
class RepositoryFormatPackGCCHK255Big(RepositoryFormatPackGCCHK255):
595
"""A hashed CHK+group compress pack repository."""
596
597
repository_class = GCCHKPackRepository
598
supports_chks = True
599
# For right now, setting this to True gives us InterModel1And2 rather
600
# than InterDifferingSerializer
601
rich_root_data = True
602
_serializer = chk_serializer.chk_serializer_255_bigpage
603
# Note: We cannot unpack a delta that references a text we haven't
604
# seen yet. There are 2 options, work in fulltexts, or require
605
# topological sorting. Using fulltexts is more optimal for local
606
# operations, because the source can be smart about extracting
607
# multiple in-a-row (and sharing strings). Topological is better
608
# for remote, because we access less data.
609
_fetch_order = 'unordered'
610
_fetch_uses_deltas = False
611
612
def get_format_string(self):
613
"""See RepositoryFormat.get_format_string()."""
614
return ('Bazaar development format - hash255chk+gc rich-root bigpage'
615
' (needs bzr.dev from 1.13)\n')
616
617
def get_format_description(self):
618
"""See RepositoryFormat.get_format_description()."""
619
return ("Development repository format - hash255chk+groupcompress + bigpage")
Loggerhead is a web-based interface for Breezy
Version: 2.0.1