247
242
rev_id).get_file_text('this-file-id'), text)
250
class TestKnitToPackFetch(TestCaseWithTransport):
252
def find_get_record_stream(self, calls, expected_count=1):
253
"""In a list of calls, find the last 'get_record_stream'.
255
:param expected_count: The number of calls we should exepect to find.
256
If a different number is found, an assertion is raised.
258
get_record_call = None
261
if call[0] == 'get_record_stream':
263
get_record_call = call
264
self.assertEqual(expected_count, call_count)
265
return get_record_call
267
def test_fetch_with_deltas_no_delta_closure(self):
268
tree = self.make_branch_and_tree('source', format='dirstate')
269
target = self.make_repository('target', format='pack-0.92')
270
self.build_tree(['source/file'])
271
tree.set_root_id('root-id')
272
tree.add('file', 'file-id')
273
tree.commit('one', rev_id='rev-one')
274
source = tree.branch.repository
275
source.texts = versionedfile.RecordingVersionedFilesDecorator(
277
source.signatures = versionedfile.RecordingVersionedFilesDecorator(
279
source.revisions = versionedfile.RecordingVersionedFilesDecorator(
281
source.inventories = versionedfile.RecordingVersionedFilesDecorator(
284
self.assertTrue(target._format._fetch_uses_deltas)
285
target.fetch(source, revision_id='rev-one')
286
self.assertEqual(('get_record_stream', [('file-id', 'rev-one')],
287
target._format._fetch_order, False),
288
self.find_get_record_stream(source.texts.calls))
289
self.assertEqual(('get_record_stream', [('rev-one',)],
290
target._format._fetch_order, False),
291
self.find_get_record_stream(source.inventories.calls, 2))
292
self.assertEqual(('get_record_stream', [('rev-one',)],
293
target._format._fetch_order, False),
294
self.find_get_record_stream(source.revisions.calls))
295
# XXX: Signatures is special, and slightly broken. The
296
# standard item_keys_introduced_by actually does a lookup for every
297
# signature to see if it exists, rather than waiting to do them all at
298
# once at the end. The fetch code then does an all-at-once and just
299
# allows for some of them to be missing.
300
# So we know there will be extra calls, but the *last* one is the one
302
signature_calls = source.signatures.calls[-1:]
303
self.assertEqual(('get_record_stream', [('rev-one',)],
304
target._format._fetch_order, False),
305
self.find_get_record_stream(signature_calls))
307
def test_fetch_no_deltas_with_delta_closure(self):
308
tree = self.make_branch_and_tree('source', format='dirstate')
309
target = self.make_repository('target', format='pack-0.92')
310
self.build_tree(['source/file'])
311
tree.set_root_id('root-id')
312
tree.add('file', 'file-id')
313
tree.commit('one', rev_id='rev-one')
314
source = tree.branch.repository
315
source.texts = versionedfile.RecordingVersionedFilesDecorator(
317
source.signatures = versionedfile.RecordingVersionedFilesDecorator(
319
source.revisions = versionedfile.RecordingVersionedFilesDecorator(
321
source.inventories = versionedfile.RecordingVersionedFilesDecorator(
323
# XXX: This won't work in general, but for the dirstate format it does.
324
self.overrideAttr(target._format, '_fetch_uses_deltas', False)
325
target.fetch(source, revision_id='rev-one')
326
self.assertEqual(('get_record_stream', [('file-id', 'rev-one')],
327
target._format._fetch_order, True),
328
self.find_get_record_stream(source.texts.calls))
329
self.assertEqual(('get_record_stream', [('rev-one',)],
330
target._format._fetch_order, True),
331
self.find_get_record_stream(source.inventories.calls, 2))
332
self.assertEqual(('get_record_stream', [('rev-one',)],
333
target._format._fetch_order, True),
334
self.find_get_record_stream(source.revisions.calls))
335
# XXX: Signatures is special, and slightly broken. The
336
# standard item_keys_introduced_by actually does a lookup for every
337
# signature to see if it exists, rather than waiting to do them all at
338
# once at the end. The fetch code then does an all-at-once and just
339
# allows for some of them to be missing.
340
# So we know there will be extra calls, but the *last* one is the one
342
signature_calls = source.signatures.calls[-1:]
343
self.assertEqual(('get_record_stream', [('rev-one',)],
344
target._format._fetch_order, True),
345
self.find_get_record_stream(signature_calls))
347
def test_fetch_revisions_with_deltas_into_pack(self):
348
# See BUG #261339, dev versions of bzr could accidentally create deltas
349
# in revision texts in knit branches (when fetching from packs). So we
350
# ensure that *if* a knit repository has a delta in revisions, that it
351
# gets properly expanded back into a fulltext when stored in the pack
353
tree = self.make_branch_and_tree('source', format='dirstate')
354
target = self.make_repository('target', format='pack-0.92')
355
self.build_tree(['source/file'])
356
tree.set_root_id('root-id')
357
tree.add('file', 'file-id')
358
tree.commit('one', rev_id='rev-one')
359
# Hack the KVF for revisions so that it "accidentally" allows a delta
360
tree.branch.repository.revisions._max_delta_chain = 200
361
tree.commit('two', rev_id='rev-two')
362
source = tree.branch.repository
363
# Ensure that we stored a delta
365
self.addCleanup(source.unlock)
366
record = source.revisions.get_record_stream([('rev-two',)],
367
'unordered', False).next()
368
self.assertEqual('knit-delta-gz', record.storage_kind)
369
target.fetch(tree.branch.repository, revision_id='rev-two')
370
# The record should get expanded back to a fulltext
372
self.addCleanup(target.unlock)
373
record = target.revisions.get_record_stream([('rev-two',)],
374
'unordered', False).next()
375
self.assertEqual('knit-ft-gz', record.storage_kind)
377
def test_fetch_with_fallback_and_merge(self):
378
builder = self.make_branch_builder('source', format='pack-0.92')
379
builder.start_series()
390
# A & B are present in the base (stacked-on) repository, A-E are
391
# present in the source.
392
# This reproduces bug #304841
393
# We need a large enough inventory that total size of compressed deltas
394
# is shorter than the size of a compressed fulltext. We have to use
395
# random ids because otherwise the inventory fulltext compresses too
396
# well and the deltas get bigger.
398
('add', ('', 'TREE_ROOT', 'directory', None))]
400
fname = 'file%03d' % (i,)
401
fileid = '%s-%s' % (fname, osutils.rand_chars(64))
402
to_add.append(('add', (fname, fileid, 'file', 'content\n')))
403
builder.build_snapshot('A', None, to_add)
404
builder.build_snapshot('B', ['A'], [])
405
builder.build_snapshot('C', ['A'], [])
406
builder.build_snapshot('D', ['C'], [])
407
builder.build_snapshot('E', ['D'], [])
408
builder.build_snapshot('F', ['E', 'B'], [])
409
builder.finish_series()
410
source_branch = builder.get_branch()
411
source_branch.bzrdir.sprout('base', revision_id='B')
412
target_branch = self.make_branch('target', format='1.6')
413
target_branch.set_stacked_on_url('../base')
414
source = source_branch.repository
416
self.addCleanup(source.unlock)
417
source.inventories = versionedfile.OrderingVersionedFilesDecorator(
419
key_priority={('E',): 1, ('D',): 2, ('C',): 4,
421
# Ensure that the content is yielded in the proper order, and given as
423
records = [(record.key, record.storage_kind)
424
for record in source.inventories.get_record_stream(
425
[('D',), ('C',), ('E',), ('F',)], 'unordered', False)]
426
self.assertEqual([(('E',), 'knit-delta-gz'), (('D',), 'knit-delta-gz'),
427
(('F',), 'knit-delta-gz'), (('C',), 'knit-delta-gz')],
430
target_branch.lock_write()
431
self.addCleanup(target_branch.unlock)
432
target = target_branch.repository
433
target.fetch(source, revision_id='F')
434
# 'C' should be expanded to a fulltext, but D and E should still be
436
stream = target.inventories.get_record_stream(
437
[('C',), ('D',), ('E',), ('F',)],
439
kinds = dict((record.key, record.storage_kind) for record in stream)
440
self.assertEqual({('C',): 'knit-ft-gz', ('D',): 'knit-delta-gz',
441
('E',): 'knit-delta-gz', ('F',): 'knit-delta-gz'},
245
class TestHttpFetch(TestCaseWithWebserver):
246
# FIXME RBC 20060124 this really isn't web specific, perhaps an
247
# instrumented readonly transport? Can we do an instrumented
248
# adapter and use self.get_readonly_url ?
250
def test_fetch(self):
251
#highest indices a: 5, b: 7
252
br_a, br_b = make_branches(self)
253
br_rem_a = Branch.open(self.get_readonly_url('branch1'))
254
fetch_steps(self, br_rem_a, br_b, br_a)
256
def _count_log_matches(self, target, logs):
257
"""Count the number of times the target file pattern was fetched in an http log"""
258
get_succeeds_re = re.compile(
259
'.*"GET .*%s HTTP/1.1" 20[06] - "-" "bzr/%s' %
260
( target, bzrlib.__version__))
263
if get_succeeds_re.match(line):
267
def test_weaves_are_retrieved_once(self):
268
self.build_tree(("source/", "source/file", "target/"))
269
# This test depends on knit dasta storage.
270
wt = self.make_branch_and_tree('source', format='dirstate-tags')
272
wt.add(["file"], ["id"])
273
wt.commit("added file")
274
open("source/file", 'w').write("blah\n")
275
wt.commit("changed file")
276
target = BzrDir.create_branch_and_repo("target/")
277
source = Branch.open(self.get_readonly_url("source/"))
278
self.assertEqual(target.fetch(source), (2, []))
279
# this is the path to the literal file. As format changes
280
# occur it needs to be updated. FIXME: ask the store for the
282
self.log("web server logs are:")
283
http_logs = self.get_readonly_server().logs
284
self.log('\n'.join(http_logs))
285
# unfortunately this log entry is branch format specific. We could
286
# factor out the 'what files does this format use' to a method on the
287
# repository, which would let us to this generically. RBC 20060419
288
# RBC 20080408: Or perhaps we can assert that no files are fully read
290
self.assertEqual(1, self._count_log_matches('/ce/id.kndx', http_logs))
291
self.assertEqual(1, self._count_log_matches('/ce/id.knit', http_logs))
292
# XXX: This *should* be '1', but more intrusive fetch changes are
293
# needed to drop this back to 1.
294
self.assertEqual(2, self._count_log_matches('inventory.kndx', http_logs))
295
# this r-h check test will prevent regressions, but it currently already
296
# passes, before the patch to cache-rh is applied :[
297
self.assertTrue(1 >= self._count_log_matches('revision-history',
299
self.assertTrue(1 >= self._count_log_matches('last-revision',
301
# FIXME naughty poking in there.
302
self.get_readonly_server().logs = []
303
# check there is nothing more to fetch. We take care to re-use the
304
# existing transport so that the request logs we're about to examine
305
# aren't cluttered with redundant probes for a smart server.
306
# XXX: Perhaps this further parameterisation: test http with smart
307
# server, and test http without smart server?
308
source = Branch.open(
309
self.get_readonly_url("source/"),
310
possible_transports=[source.bzrdir.root_transport])
311
self.assertEqual(target.fetch(source), (0, []))
312
# should make just two requests
313
http_logs = self.get_readonly_server().logs
314
self.log("web server logs are:")
315
self.log('\n'.join(http_logs))
316
self.assertEqual(1, self._count_log_matches('branch-format', http_logs))
317
self.assertEqual(1, self._count_log_matches('branch/format', http_logs))
318
self.assertEqual(1, self._count_log_matches('repository/format',
320
self.assertTrue(1 >= self._count_log_matches('revision-history',
322
self.assertTrue(1 >= self._count_log_matches('last-revision',
324
self.assertEqual(4, len(http_logs))
445
327
class Test1To2Fetch(TestCaseWithTransport):