238
246
rev_id).get_file_text('this-file-id'), text)
241
class TestKnitToPackFetch(TestCaseWithTransport):
243
def find_get_record_stream(self, calls, expected_count=1):
244
"""In a list of calls, find the last 'get_record_stream'.
246
:param expected_count: The number of calls we should exepect to find.
247
If a different number is found, an assertion is raised.
249
get_record_call = None
252
if call[0] == 'get_record_stream':
254
get_record_call = call
255
self.assertEqual(expected_count, call_count)
256
return get_record_call
258
def test_fetch_with_deltas_no_delta_closure(self):
259
tree = self.make_branch_and_tree('source', format='dirstate')
260
target = self.make_repository('target', format='pack-0.92')
261
self.build_tree(['source/file'])
262
tree.set_root_id('root-id')
263
tree.add('file', 'file-id')
264
tree.commit('one', rev_id='rev-one')
265
source = tree.branch.repository
266
source.texts = versionedfile.RecordingVersionedFilesDecorator(
268
source.signatures = versionedfile.RecordingVersionedFilesDecorator(
270
source.revisions = versionedfile.RecordingVersionedFilesDecorator(
272
source.inventories = versionedfile.RecordingVersionedFilesDecorator(
275
self.assertTrue(target._format._fetch_uses_deltas)
276
target.fetch(source, revision_id='rev-one')
277
self.assertEqual(('get_record_stream', [('file-id', 'rev-one')],
278
target._format._fetch_order, False),
279
self.find_get_record_stream(source.texts.calls))
280
self.assertEqual(('get_record_stream', [('rev-one',)],
281
target._format._fetch_order, False),
282
self.find_get_record_stream(source.inventories.calls, 2))
283
self.assertEqual(('get_record_stream', [('rev-one',)],
284
target._format._fetch_order, False),
285
self.find_get_record_stream(source.revisions.calls))
286
# XXX: Signatures is special, and slightly broken. The
287
# standard item_keys_introduced_by actually does a lookup for every
288
# signature to see if it exists, rather than waiting to do them all at
289
# once at the end. The fetch code then does an all-at-once and just
290
# allows for some of them to be missing.
291
# So we know there will be extra calls, but the *last* one is the one
293
signature_calls = source.signatures.calls[-1:]
294
self.assertEqual(('get_record_stream', [('rev-one',)],
295
target._format._fetch_order, False),
296
self.find_get_record_stream(signature_calls))
298
def test_fetch_no_deltas_with_delta_closure(self):
299
tree = self.make_branch_and_tree('source', format='dirstate')
300
target = self.make_repository('target', format='pack-0.92')
301
self.build_tree(['source/file'])
302
tree.set_root_id('root-id')
303
tree.add('file', 'file-id')
304
tree.commit('one', rev_id='rev-one')
305
source = tree.branch.repository
306
source.texts = versionedfile.RecordingVersionedFilesDecorator(
308
source.signatures = versionedfile.RecordingVersionedFilesDecorator(
310
source.revisions = versionedfile.RecordingVersionedFilesDecorator(
312
source.inventories = versionedfile.RecordingVersionedFilesDecorator(
314
# XXX: This won't work in general, but for the dirstate format it does.
315
self.overrideAttr(target._format, '_fetch_uses_deltas', False)
316
target.fetch(source, revision_id='rev-one')
317
self.assertEqual(('get_record_stream', [('file-id', 'rev-one')],
318
target._format._fetch_order, True),
319
self.find_get_record_stream(source.texts.calls))
320
self.assertEqual(('get_record_stream', [('rev-one',)],
321
target._format._fetch_order, True),
322
self.find_get_record_stream(source.inventories.calls, 2))
323
self.assertEqual(('get_record_stream', [('rev-one',)],
324
target._format._fetch_order, True),
325
self.find_get_record_stream(source.revisions.calls))
326
# XXX: Signatures is special, and slightly broken. The
327
# standard item_keys_introduced_by actually does a lookup for every
328
# signature to see if it exists, rather than waiting to do them all at
329
# once at the end. The fetch code then does an all-at-once and just
330
# allows for some of them to be missing.
331
# So we know there will be extra calls, but the *last* one is the one
333
signature_calls = source.signatures.calls[-1:]
334
self.assertEqual(('get_record_stream', [('rev-one',)],
335
target._format._fetch_order, True),
336
self.find_get_record_stream(signature_calls))
338
def test_fetch_revisions_with_deltas_into_pack(self):
339
# See BUG #261339, dev versions of bzr could accidentally create deltas
340
# in revision texts in knit branches (when fetching from packs). So we
341
# ensure that *if* a knit repository has a delta in revisions, that it
342
# gets properly expanded back into a fulltext when stored in the pack
344
tree = self.make_branch_and_tree('source', format='dirstate')
345
target = self.make_repository('target', format='pack-0.92')
346
self.build_tree(['source/file'])
347
tree.set_root_id('root-id')
348
tree.add('file', 'file-id')
349
tree.commit('one', rev_id='rev-one')
350
# Hack the KVF for revisions so that it "accidentally" allows a delta
351
tree.branch.repository.revisions._max_delta_chain = 200
352
tree.commit('two', rev_id='rev-two')
353
source = tree.branch.repository
354
# Ensure that we stored a delta
356
self.addCleanup(source.unlock)
357
record = source.revisions.get_record_stream([('rev-two',)],
358
'unordered', False).next()
359
self.assertEqual('knit-delta-gz', record.storage_kind)
360
target.fetch(tree.branch.repository, revision_id='rev-two')
361
# The record should get expanded back to a fulltext
363
self.addCleanup(target.unlock)
364
record = target.revisions.get_record_stream([('rev-two',)],
365
'unordered', False).next()
366
self.assertEqual('knit-ft-gz', record.storage_kind)
368
def test_fetch_with_fallback_and_merge(self):
369
builder = self.make_branch_builder('source', format='pack-0.92')
370
builder.start_series()
381
# A & B are present in the base (stacked-on) repository, A-E are
382
# present in the source.
383
# This reproduces bug #304841
384
# We need a large enough inventory that total size of compressed deltas
385
# is shorter than the size of a compressed fulltext. We have to use
386
# random ids because otherwise the inventory fulltext compresses too
387
# well and the deltas get bigger.
389
('add', ('', 'TREE_ROOT', 'directory', None))]
391
fname = 'file%03d' % (i,)
392
fileid = '%s-%s' % (fname, osutils.rand_chars(64))
393
to_add.append(('add', (fname, fileid, 'file', 'content\n')))
394
builder.build_snapshot('A', None, to_add)
395
builder.build_snapshot('B', ['A'], [])
396
builder.build_snapshot('C', ['A'], [])
397
builder.build_snapshot('D', ['C'], [])
398
builder.build_snapshot('E', ['D'], [])
399
builder.build_snapshot('F', ['E', 'B'], [])
400
builder.finish_series()
401
source_branch = builder.get_branch()
402
source_branch.bzrdir.sprout('base', revision_id='B')
403
target_branch = self.make_branch('target', format='1.6')
404
target_branch.set_stacked_on_url('../base')
405
source = source_branch.repository
407
self.addCleanup(source.unlock)
408
source.inventories = versionedfile.OrderingVersionedFilesDecorator(
410
key_priority={('E',): 1, ('D',): 2, ('C',): 4,
412
# Ensure that the content is yielded in the proper order, and given as
414
records = [(record.key, record.storage_kind)
415
for record in source.inventories.get_record_stream(
416
[('D',), ('C',), ('E',), ('F',)], 'unordered', False)]
417
self.assertEqual([(('E',), 'knit-delta-gz'), (('D',), 'knit-delta-gz'),
418
(('F',), 'knit-delta-gz'), (('C',), 'knit-delta-gz')],
421
target_branch.lock_write()
422
self.addCleanup(target_branch.unlock)
423
target = target_branch.repository
424
target.fetch(source, revision_id='F')
425
# 'C' should be expanded to a fulltext, but D and E should still be
427
stream = target.inventories.get_record_stream(
428
[('C',), ('D',), ('E',), ('F',)],
430
kinds = dict((record.key, record.storage_kind) for record in stream)
431
self.assertEqual({('C',): 'knit-ft-gz', ('D',): 'knit-delta-gz',
432
('E',): 'knit-delta-gz', ('F',): 'knit-delta-gz'},
249
class TestHttpFetch(TestCaseWithWebserver):
250
# FIXME RBC 20060124 this really isn't web specific, perhaps an
251
# instrumented readonly transport? Can we do an instrumented
252
# adapter and use self.get_readonly_url ?
254
def test_fetch(self):
255
#highest indices a: 5, b: 7
256
br_a, br_b = make_branches(self)
257
br_rem_a = Branch.open(self.get_readonly_url('branch1'))
258
fetch_steps(self, br_rem_a, br_b, br_a)
260
def _count_log_matches(self, target, logs):
261
"""Count the number of times the target file pattern was fetched in an http log"""
262
get_succeeds_re = re.compile(
263
'.*"GET .*%s HTTP/1.1" 20[06] - "-" "bzr/%s' %
264
( target, bzrlib.__version__))
267
if get_succeeds_re.match(line):
271
def test_weaves_are_retrieved_once(self):
272
self.build_tree(("source/", "source/file", "target/"))
273
# This test depends on knit dasta storage.
274
wt = self.make_branch_and_tree('source', format='dirstate-tags')
276
wt.add(["file"], ["id"])
277
wt.commit("added file")
278
open("source/file", 'w').write("blah\n")
279
wt.commit("changed file")
280
target = BzrDir.create_branch_and_repo("target/")
281
source = Branch.open(self.get_readonly_url("source/"))
282
self.assertEqual(target.fetch(source), (2, []))
283
# this is the path to the literal file. As format changes
284
# occur it needs to be updated. FIXME: ask the store for the
286
self.log("web server logs are:")
287
http_logs = self.get_readonly_server().logs
288
self.log('\n'.join(http_logs))
289
# unfortunately this log entry is branch format specific. We could
290
# factor out the 'what files does this format use' to a method on the
291
# repository, which would let us to this generically. RBC 20060419
292
# RBC 20080408: Or perhaps we can assert that no files are fully read
294
self.assertEqual(1, self._count_log_matches('/ce/id.kndx', http_logs))
295
self.assertEqual(1, self._count_log_matches('/ce/id.knit', http_logs))
296
self.assertEqual(1, self._count_log_matches('inventory.kndx', http_logs))
297
# this r-h check test will prevent regressions, but it currently already
298
# passes, before the patch to cache-rh is applied :[
299
self.assertTrue(1 >= self._count_log_matches('revision-history',
301
self.assertTrue(1 >= self._count_log_matches('last-revision',
303
# FIXME naughty poking in there.
304
self.get_readonly_server().logs = []
305
# check there is nothing more to fetch. We take care to re-use the
306
# existing transport so that the request logs we're about to examine
307
# aren't cluttered with redundant probes for a smart server.
308
# XXX: Perhaps this further parameterisation: test http with smart
309
# server, and test http without smart server?
310
source = Branch.open(
311
self.get_readonly_url("source/"),
312
possible_transports=[source.bzrdir.root_transport])
313
self.assertEqual(target.fetch(source), (0, []))
314
# should make just two requests
315
http_logs = self.get_readonly_server().logs
316
self.log("web server logs are:")
317
self.log('\n'.join(http_logs))
318
self.assertEqual(1, self._count_log_matches('branch-format', http_logs))
319
self.assertEqual(1, self._count_log_matches('branch/format', http_logs))
320
self.assertEqual(1, self._count_log_matches('repository/format',
322
self.assertTrue(1 >= self._count_log_matches('revision-history',
324
self.assertTrue(1 >= self._count_log_matches('last-revision',
326
self.assertEqual(4, len(http_logs))
436
329
class Test1To2Fetch(TestCaseWithTransport):