Branches of Bazaar

Author: John A Meinel
Revision Date: 2011-09-30 11:55:30 UTC

more debugging bits.

Author: John A Meinel
Revision Date: 2011-09-02 22:26:49 UTC

Merge 2.3 into 2.4 including the get_parent_map tweak.

Author: John A Meinel
Revision Date: 2011-08-31 07:18:59 UTC

Switch sync to default True, and try another failed run.

Author: John A Meinel
Revision Date: 2011-08-30 10:49:48 UTC

Add a failing test, so that I don't submit stuff I don't want.

Author: John A Meinel
Revision Date: 2011-08-29 17:32:36 UTC

Don't actually call fdatasync, make osutils.fdatasync a no-op

Author: John A Meinel
Revision Date: 2011-05-30 10:32:30 UTC

Prototype a format that removes redundancy, should be faster to parse.

Author: John A Meinel
Revision Date: 2011-05-03 12:40:03 UTC

Second commit, more debugging.

Author: Canonical.com Patch Queue Manager
Revision Date: 2011-05-02 21:13:54 UTC

(jelmer) Remove unused RepositoryFormat._get_versioned_file_store. (Jelmer
 Vernooij)

Author: John A Meinel
Revision Date: 2011-05-02 12:04:29 UTC

First commit, with debugging

Author: Canonical.com Patch Queue Manager
Revision Date: 2011-04-21 13:22:08 UTC

(jameinel) Fix bug #765881,
 we don't need to save the dirstate file for trivial changes. (John A Meinel)

Author: John A Meinel
Revision Date: 2011-04-22 10:30:42 UTC

Update setup.py and release-notes to say python 2.6

Author: John A Meinel
Revision Date: 2011-03-31 14:01:09 UTC

no change 3

Author: John A Meinel
Revision Date: 2011-03-08 11:57:31 UTC

Merge the fix from 2.0 all the way up to the 2.4 release.

Author: John A Meinel
Revision Date: 2011-02-22 22:06:28 UTC

Add flags for enabling --delta computation.

Something isn't 100% correct, because post-filter we end up with different counts.
The performance is pretty darn good, though.
[--filter] Took 4.184s for 100 revisions (610 changes, 88005 file_ids) and path: bzrlib
vs
Took 0.702s for 100 revisions (599 changes, 88005 file_ids) and path: bzrlib
vs
[--filter] Took 3.268s for 100 revisions (0 changes, 88005 file_ids) and path: bzrlib
and
Took 0.288s for 100 revisions (0 changes, 88005 file_ids) and path: bzrlib
Note also that iter_changes is faster in the chk form, than the pre-filtered form.
Not huge, but 0.7-0.28 = 0.414, vs 4.184-3.268=0.916.

Author: John A Meinel
Revision Date: 2010-12-02 21:42:46 UTC

A TODO entry.

Overall this experiment hasn't been particularly beneficial. The final speed
is still slower than the existing code, and the primary knob that reduced
peak memory size is to change the stride for large content. Which can be
trivially added to the existing match code.

I like the code quite a bit, but I wish I had more to show for the amount
of effort put into it.

Author: John A Meinel
Revision Date: 2010-07-20 14:08:15 UTC

Revive my really-old split-pack code.

Author: John A Meinel
Revision Date: 2010-06-23 20:56:41 UTC

calling repo.reconcile(thorough=True) seems to fail on all formats.

This is unexpected, as I would have thought 1.9 formats would have been fine.

Author: John A Meinel
Revision Date: 2010-06-17 13:56:44 UTC

Wrap socket objects so we can get tracebacks.

Author: John A Meinel
Revision Date: 2010-06-16 18:59:40 UTC

we can now round-trip content via data_access.

Author: John A Meinel
Revision Date: 2010-06-07 21:01:46 UTC

Add the simplest setup I could think of to test that we can handle a ghost.

Author: John A Meinel
Revision Date: 2010-05-01 04:52:08 UTC

Possible thread-safe LRUCache implementation.

You could still break stuff by using .add() by multiple threads, but
at least the double-linked list structure should always be correct.
Which is the greater source of threading failures.

Author: John A Meinel
Revision Date: 2010-03-29 23:47:20 UTC

Some comments on the reverse mapping.

Author: John A Meinel
Revision Date: 2010-03-14 11:08:31 UTC

Merge and revert the removal, but keep some of the changes.

Author: philyoon
Revision Date: 2010-02-11 17:43:26 UTC

Escape the to_location for 'bzr switch'
so that it can handle non-ascii sibling branch names.

Needs tests. Bug #317778

Author: John A Meinel
Revision Date: 2010-01-15 14:37:12 UTC

Change the match order.
We lose a tiny bit of precision, but it should increase our match speed.
Specifically, if a pattern is not ignored, we don't need to evaluate
the exclusion pattern. Further we don't need to evaluate the always-ignore
pattern unless the exclusion pattern matches.
The loss of precision is that if a pattern is always-ignore but not
excluded, then we don't add the '\!\!' back to the pattern.
The gain is that in the 'initial add' case, we only need to check one
match, and not 3. And in the 'ignored' case, we only need to check
2 matches, and not 3. This should help decrease the performance hit.

Author: John A Meinel
Revision Date: 2010-01-06 01:12:07 UTC

Merge bzr.dev 4934, resolve a small conflict, come up to date.

Author: John A Meinel
Revision Date: 2009-12-15 17:39:31 UTC

Move the patch for dir exporter into a helper.

This makes it easier for other exporters to be fast. Note that because
the content is defined as being returned in arbitrary order, this may
not actually be appropriate for tar and zip exporters. People complained
that 'bzr export foo.tar.gz' wasn't stable enough because of gz timestamps,
this would be far worse.

Author: John A Meinel
Revision Date: 2009-12-09 17:29:36 UTC

Handle when pull --overwrite no longer has the tip in the ancestry.

Author: John A Meinel
Revision Date: 2009-10-23 15:01:54 UTC

Remove some of the build dependencies from the bzr source tree.

Note that the build process still uses *some* files from the bzr source.
Which is bad (IMO), but the way it is until we get a real overhaul and cleanup.

Author: John A Meinel
Revision Date: 2009-10-13 22:01:39 UTC

Handle the case where the recursive call ends up returning NotImplemented.

This had been raising an exception, which wasn't the nicest way to handle it.
I don't know how we have the mixed tuple I'm seeing now, I'm guessing something else
is wrong. But at least this avoids one failure point.

Author: John A Meinel
Revision Date: 2009-10-08 22:01:25 UTC

And a fix for has_id()

Author: John A Meinel
Revision Date: 2009-09-24 22:15:25 UTC

We can check if old matches first, which gives us a couple more hits.
Down to 49s. Not the <36s I'd like to be at, though...

Author: John A Meinel
Revision Date: 2009-09-17 12:19:47 UTC

Start trying to put together a 'windows' front end
that won't create the dos box.
I think it would work, but py2exe is confused about my current version
of PyQt, which has some py3k code that can't be imported, but if
I list it in 'excludes' then it complains that the code isn't there...

Author: John A Meinel
Revision Date: 2009-08-24 18:33:43 UTC

small tweaks

For some reason, it doesn't seem to be faster than the pure python version...
very strange.

Author: John A Meinel
Revision Date: 2009-07-01 22:17:39 UTC

Implement LeafNode.items() rather than going directly to .keys.

At this point 'timeit' shows that a partial parse gives 28ms down from 49ms.
So an approx savings of 2x.

Author: John A Meinel
Revision Date: 2009-06-19 16:40:50 UTC

Add some more 'number of bytes' debugging to '-Dhpss'

Author: John A Meinel
Revision Date: 2009-06-16 22:07:32 UTC

Merge bzr.dev 4449 to get the _known_graph code.

Author: John A Meinel
Revision Date: 2009-06-15 17:17:57 UTC

Add a fairly broken test for StreamSource working to send data to RemoteRepo.

It turns out we have 3 bugs.
1) (already known) Knit formats don't work, because they try to lock the target again.
2) RemoteRepositoryFormat has a bogus value for 'supports_chks'. This causes the code
that is determining how to transmit inventories to get confused. (In these tests the
two formats are identical, we just don't realize that...)
3) The StreamSink code uses 'from_serializer == to_serializer' to determine if it
needs to cast the fulltexts back up to Inventory objects or not. However, because
of (2) we would have sent a full text inventory, but Sink would think we did not...
4) Latent bug... supports_chk isn't quite enough detail given serializer
issues. However, it is sufficient for all current formats, so we'll live with
it.

Author: John A Meinel
Revision Date: 2009-06-04 20:19:29 UTC

Use the chk_serializer in actual work.

Author: John A Meinel
Revision Date: 2009-04-27 22:14:38 UTC

Merge in the change to cause 'bzr pack' to pack 'recent' things together.
Will probably revert the part that moves recent texts to the front,
but it should be evaluated.

Author: John A Meinel
Revision Date: 2009-04-20 22:24:57 UTC

Filter keys before checking a given backing index.

Author: Vincent Ladeuil
Revision Date: 2009-02-03 20:22:14 UTC

Ensure compatibility up to bzr-1.6.1 (at least).

* reconcile.py:
(InventoryAncestryReconcilePacker._copy_fixed_nodes): Ensure
old (as far as 1.6.1) bzr versions compatibility by using either
'chunked' or 'fulltext' storage kind. Fix another edge case in
batch calculation

Author: John A Meinel
Revision Date: 2009-01-20 20:51:29 UTC

Expose get_lazy_revno_mapper() as part of Branch.

Author: John A Meinel
Revision Date: 2009-01-16 19:07:49 UTC

Properly clean up the progress stack when a nested progress bar is finished.

Author: John A Meinel
Revision Date: 2008-11-06 17:27:21 UTC

Fix bug #293746, we just need to handle when bisect_left gives us the end of the list.

Author: John A Meinel
Revision Date: 2008-06-30 18:19:14 UTC

Update the lca_multi_way logic so that it can handle when bases supersede eachother for a given entry.

Author: John A Meinel
Revision Date: 2008-06-20 16:13:41 UTC

Change the default annotation policy to 'right-head'

Author: John A Meinel
Revision Date: 2008-03-13 16:29:38 UTC

(broken) start exposing 'annotator' as a Tree and Repository api.