[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Bazaar 1.10 released!

From: John Arbash Meinel
Subject: Bazaar 1.10 released!
Date: Fri, 05 Dec 2008 12:17:44 -0600
User-agent: Thunderbird (X11/20080925)

Hash: SHA1

Bazaar 1.10 has several performance improvements for copying revisions
(especially for small updates to large projects).  There has also been a
significant amount of effort in polishing stacked branches.  The commands
``shelve`` and ``unshelve`` have become core commands, with an improved

bzr 1.10 is now available for download from
Packages for various systems will be available soon.

Thanks to everyone who contributed patches, suggestions, and feedback.


NEWS since the previous release:

bzr 1.10 2008-12-05
- -------------------

Bazaar 1.10 has several performance improvements for copying revisions
(especially for small updates to large projects).  There has also been a
significant amount of effort in polishing stacked branches.  The commands
``shelve`` and ``unshelve`` have become core commands, with an improved

The only changes versus bzr-1.10rc1 are bugfixes for stacked branches.


    * Don't set a pack write cache size from RepoFetcher, because the
      cache is not coherent with reads and causes ShortReadvErrors.
      This reverses the change that fixed #294479.
      (Martin Pool, #303856)

    * Properly handle when a revision can be inserted as a delta versus
      when it needs to be expanded to a fulltext for stacked branches.
      There was a bug involving merge revisions. As a method to help
      prevent future difficulties, also make stacked fetches sort
      topologically. (John Arbash Meinel, #304841)

bzr 1.10rc1 2008-11-28
- ----------------------

This release of Bazaar focuses on performance improvements when pushing
and pulling revisions, both locally and to remote networks.  The popular
``shelve`` and ``unshelve`` commands, used to interactively revert and
restore work in progress, have been merged from bzrtools into the bzr
core.  There are also bug fixes for portability, and for stacked branches.


    * New ``commit_message_template`` hook that is called by the commit
      code to generate a template commit message. (Jelmer Vernooij)

    * New `shelve` and `unshelve` commands allow undoing and redoing changes.
      (Aaron Bentley)


    * ``(Remote)Branch.copy_content_into`` no longer generates the full
      revision history just to set the last revision info.
      (Andrew Bennetts, John Arbash Meinel)

    * Fetches between formats with different serializers (such as
      pack-0.92-subtree and 1.9-rich-root) are faster now.  This is due to
      operating on batches of 100 revisions at time rather than
      one-by-one.  (Andrew Bennetts, John Arbash Meinel)

    * Search index files corresponding to pack files we've already used
      before searching others, because they are more likely to have the
      keys we're looking for.  This reduces the number of iix and tix
      files accessed when pushing 1 new revision, for instance.
      (John Arbash Meinel)

    * Signatures to transfer are calculated more efficiently in
      ``item_keys_introduced_by``.  (Andrew Bennetts, John Arbash Meinel)

    * The generic fetch code can once again copy revisions and signatures
      without extracting them completely to fulltexts and then serializing
      them back down into byte strings. This is a significant performance
      improvement when fetching from a stacked branch.
      (John Arbash Meinel, #300289)

    * When making a large readv() request over ``bzr+ssh``, break up the
      request into more manageable chunks. Because the RPC is not yet able
      to stream, this helps keep us from buffering too much information at
      once. (John Arbash Meinel)


    * Better message when the user needs to set their Launchpad ID.
      (Martin Pool, #289148)

    * ``bzr commit --local`` doesn't access the master branch anymore.
      This fixes a regression introduced in 1.9.  (Marius Kruger, #299313)

    * Don't call the system ``chdir()`` with an empty path. Sun OS seems
      to give an error in that case.  Also, don't count on ``getcwd()``
      being able to allocate a new buffer, which is a gnu extension.
      (John Arbash Meinel, Martin Pool, Harry Hirsch, #297831)

    * Don't crash when requesting log --forward <file> for a revision range
      starting with a dotted revno.
      (Vincent Ladeuil, #300055)

    * Don't create text deltas spanning stacked repositories; this could
      cause "Revision X not present in Y" when later accessing them.
      (Martin Pool, #288751)

    * Pack repositories are now able to reload the pack listing and retry
      the current operation if another action causes the data to be
      repacked.  (John Arbash Meinel, #153786)

    * PermissionDenied errors from smart servers no longer cause
      "PermissionDenied: "None"" on the client.
      (Andrew Bennetts, #299254)

    * Pushing to a stacked pack repository now batches writes, the same
      way writes are batched to ordinary pack repository.  This makes
      pushing to a stacked branch over the network much faster.
      (Andrew Bennetts, #294479)

    * TooManyConcurrentRequests no longer occur when a fetch fails and
      tries to abort a write group.  This allows the root cause (e.g. a
      network interruption) to be reported.  (Andrew Bennetts, #297014)

    * RemoteRepository.get_parent_map now uses fallback repositories.
      (Aaron Bentley, #297991?, #293679?)


    * ``CommitBuilder`` now validates the strings it will be committing,
      to ensure that they do not have characters that will not be properly
      round-tripped. For now, it just checks for characters that are
      invalid in the XML form. (John Arbash Meinel, #295161)

    * Constructor parameters for NewPack (internal to pack repositories)
      have changed incompatibly.

    * ``Repository.abort_write_group`` now accepts an optional
      ``suppress_errors`` flag.  Repository implementations that override
      ``abort_write_group`` will need to be updated to accept the new
      argument.  Subclasses that only override ``_abort_write_group``
      don't need to change.

    * Transport implementations must provide copy_tree_to_transport.  A
      default implementation is provided for Transport subclasses.


    * ``bzr selftest`` now fails if no doctests are found in a module
      that's expected to have them.  (Martin Pool)

    * Doctests now only report the first failure.  (Martin Pool)
Version: GnuPG v1.4.6 (GNU/Linux)
Comment: Using GnuPG with Mozilla -


reply via email to

[Prev in Thread] Current Thread [Next in Thread]