~bzr-pqm/bzr/bzr.dev

--------------------

Bazaar Release Notes

--------------------

.. contents::

IN DEVELOPMENT

--------------

  CHANGES:

    * ``bzr export`` can now export a subdirectory of a project.

      (Robert Collins)

    * ``bzr rm`` will now scan for files that are missing and remove just

      them automatically, much as ``bzr add`` scans for new files that

      are not ignored and adds them automatically. (Robert Collins)

  FEATURES

    * Support for GSSAPI authentication when using FTP as documented in 

      RFC2228. (Jelmer Vernooij, #49623)

  IMPROVEMENTS:

    * A url like ``log+file:///tmp`` will log all access to that Transport 

      to ``.bzr.log``, which may help in debugging or profiling.

      (Martin Pool)

    * ``bzr branch`` and ``bzr push`` use the default stacking policy if the

      branch format supports it. (Aaron Bentley)

    * ``bzr init`` and ``bzr init-repo`` will now print out the same as

      ``bzr info`` if it completed successfully.

      (Marius Kruger)

    * ``bzr uncommit`` logs the old tip revision id, and displays how to

      restore the branch to that tip using ``bzr pull``.  This allows you

      to recover if you realize you uncommitted the wrong thing.

      (John Arbash Meinel)

    * When reading index files, if we happen to read the whole file in a

      single request treat it as a ``_buffer_all`` request. This happens

      most often on small indexes over remote transports, where we default

      to reading 64kB. It saves a round trip for each small index during

      fetch operations. Also, if we have read more than 50% of an index

      file, trigger a ``_buffer_all`` on the next request. This works

      around some inefficiencies because reads don't fall neatly on page

      boundaries, so we would ignore those bytes, but request them again

      later. This could trigger a total read size of more than the whole

      file. (John Arbash Meinel)

  BUG FIXES:

    * ``bzr rm`` is now aliased to ``bzr del`` for the convenience of svn

      users. (Robert Collins, #205416)

    * Catch the infamous "select/poll returned error" which occurs when

      pycurl try to send a body request to an HTTP/1.0 server which has

      already refused to handle the request. (Vincent Ladeuil, #225020)

    * ``FTPTransport.stat()`` would return ``0000`` as the permission bits

      for the containing ``.bzr/`` directory (it does not implement

      permissions). This would cause us to set all subdirectories to

      ``0700`` and files to ``0600`` rather than leaving them unmodified.

      Now we ignore ``0000`` as the permissions and assume they are

      invalid. (John Arbash Meinel, #259855)

    * Merging from a previously joined branch will no longer cause 

      a traceback. (Jelmer Vernooij, #203376)

    * Pack operations on windows network shares will work even with large

      files. (Robert Collins, #255656)

    * Running ``bzr st PATH_TO_TREE`` will no longer suppress merge

      status. Status is also about 7% faster on mozilla sized trees

      when the path to the root of the tree has been given. Users of

      the internal ``show_tree_status`` function should be aware that

      the show_pending flag is now authoritative for showing pending

      merges, as it was originally. (Robert Collins, #225204)

    * Set valid default _param_name for Option so that ListOption can embed

      '-' in names. (Vincent Ladeuil, #263249)

    * Show proper error rather than traceback when an unknown revision 

      id is specified to ``bzr cat-revision``. (Jelmer Vernooij, #175569)

    * Trailing text in the dirstate file could cause the C dirstate parser

      to try to allocate an invalid amount of memory. We now properly

      check and test for parsing a dirstate with invalid trailing data.

      (John Arbash Meinel, #186014)

    * Use a Windows api function to get a Unicode host name, rather than

      assuming the host name is ascii.

      (Mark Hammond, John Arbash Meinel, #256550)

    * ``WorkingTree4`` trees will now correctly report missing-and-new

      paths in the output of ``iter_changes``. (Robert Collins)

  API CHANGES:

    * Exporters now take 4 parameters. (Robert Collins)

    * ``Tree.iter_changes`` will now return False for the content change

      field when a file is missing in the basis tree and not present in

      the target tree. Previously it returned True unconditionally.

      (Robert Collins)

    * The deprecated ``Branch.abspath`` and unimplemented 

      ``Branch.rename_one`` and ``Branch.move`` were removed. (Jelmer Vernooij)

    * BzrDir.clone_on_transport implementations must now accept a stacked_on

      parameter.  (Aaron Bentley)

    * BzrDir.cloning_metadir implementations must now take a require_stacking

      parameter.  (Aaron Bentley)

  TESTING:

    * ``addCleanup`` now takes ``*arguments`` and ``**keyword_arguments``

      which are then passed to the cleanup callable as it is run. In

      addition, addCleanup no longer requires that the callables passed to

      it be unique. (Jonathan Lange)

    * ``selftest``'s ``--starting-with`` option can now use predefined

      prefixes so that one can say ``bzr selftest -s bp.loom`` instead of

      ``bzr selftest -s bzrlib.plugins.loom``. (Vincent Ladeuil)

    * ``selftest``'s ``--starting-with`` option now accepts multiple values.

      (Vincent Ladeuil)

  INTERNALS:

    * A new plugin interface, ``bzrlib.log.log_adapters``, has been added.

      This allows dynamic log output filtering by plugins.

      (Robert Collins)

    * ``bzrlib.btree_index`` is now available, providing a b-tree index

      layer. The design is memory conservative (limited memory cache),

      faster to seek (approx 100 nodes per page, gives 100-way fan out),

      and stores compressed pages allowing more keys per page.

      (Robert Collins, John Arbash Meinel)

    * ``bzrlib.diff.DiffTree.show_diff`` now skips changes where the kind

      is unknown in both source and target.

      (Robert Collins, Aaron Bentley)

    * ``GraphIndexBuilder.add_node`` and ``BTreeBuilder`` have been

      streamlined a bit. This should make creating large indexes faster.

      (In benchmarking, it now takes less time to create a BTree index than

      it takes to read the GraphIndex one.) (John Arbash Meinel)

    * Mail clients for `bzr send` are now listed in a registry.  This

      allows plugins to add new clients by registering them with

      ``bzrlib.mail_client.mail_client_registry``.  All of the built-in

      clients now use this mechanism.  (Neil Martinsen-Burrell)

bzr 1.6.1rc2 2008-09-03

-----------------------

  BUG FIXES:

    * Copying between ``rich-root`` and ``rich-root-pack`` (and vice

      versa) was accidentally using the inter-model fetcher, instead of

      recognizing that both were 'rich root' formats.

      (John Arbash Meinel, #264321)

bzr 1.6.1rc1 2008-08-29

-----------------------

This release fixes a few regressions found in the 1.6 client. Fetching

changes was using an O(N^2) buffering algorithm, so for large projects it

would cause memory thrashing. There is also a specific problem with the

``--1.6-rich-root`` format, which prevented stacking on top of

``--rich-root-pack`` repositories, and could allow users to accidentally

fetch experimental data (``-subtree``) without representing it properly.

The ``--1.6-rich-root`` format has been deprecated and users are

recommended to upgrade to ``--1.6.1-rich-root`` immediately.  Also we

re-introduced a workaround for users who have repositories with incorrect

nodes (not possible if you only used official releases).

I should also clarify that none of this is data loss level issues, but

still sufficient enough to warrant an updated release.

  BUG FIXES:

    * ``RemoteTransport.readv()`` was being inefficient about how it

      buffered the readv data and processed it. It would keep appending to

      the same string (causing many copies) and then pop bytes out of the

      start of the string (causing more copies).

      With this patch "bzr+ssh://local" can improve dramatically,

      especially for projects with large files.

      (John Arbash Meinel)

    * Revision texts were always meant to be stored as fulltexts. There

      was a bug in a bzr.dev version that would accidentally create deltas

      when copying from a Pack repo to a Knit repo. This has been fixed,

      but to support those repositories, we know always request full texts

      for Revision texts. (John Arbash Meinel, #261339)

    * The previous ``--1.6-rich-root`` format used an incorrect xml

      serializer, which would accidentally support fetching from a

      repository that supported subtrees, even though the local one would

      not. We deprecated that format, and introduced a new one that uses

      the correct serializer ``--1.6.1-rich-root``.

      (John Arbash Meinel, #262333)

bzr 1.6 2008-08-25

------------------

Finally, the long awaited bzr 1.6 has been released. This release includes

new features like Stacked Branches, improved weave merge, and an updated

server protocol (now on v3) which will allow for better cross version

compatibility. With this release we have deprecated Knit format

repositories, and recommend that users upgrade them, we will continue to

support reading and writing them for the forseeable future, but we will

not be tuning them for performance as pack repositories have proven to be

better at scaling. This will also be the first release to bundle

TortoiseBzr in the standalone Windows installer.

bzr 1.6rc5 2008-08-19

---------------------

  BUG FIXES: 

    * Disable automatic detection of stacking based on a containing

      directory of the target. It interacted badly with push, and needs a

      bit more work to get the edges polished before it should happen

      automatically. (John Arbash Meinel, #259275)

      (This change was reverted when merged to bzr.dev)

bzr 1.6rc4 2008-08-18

---------------------

  BUG FIXES: 

    * Fix a regression in knit => pack fetching.  We had a logic

      inversion, causing the fetch to insert fulltexts in random order,

      rather than preserving deltas.  (John Arbash Meinel, #256757)

bzr 1.6rc3 2008-08-14

---------------------

  CHANGES:

    * Disable reading ``.bzrrules`` as a per-branch rule preferences

      file. The feature was not quite ready for a full release.

      (Robert Collins)

  IMPROVEMENTS:

    * Update the windows installer to bundle TortoiseBzr and ``qbzr``

      into the standalone installer. This will be the first official

      windows release that installs Tortoise by default.

      (Mark Hammond)

  BUG FIXES: 

    * Fix a regression in ``bzr+http`` support. There was a missing

      function (``_read_line``) that needed to be carried over from

      ``bzr+ssh`` support. (Andrew Bennetts)

    * ``GraphIndex`` objects will internally read an entire index if more

      than 1/20th of their keyspace is requested in a single operation.

      This largely mitigates a performance regression in ``bzr log FILE``

      and completely corrects the performance regression in ``bzr log``.

      The regression was caused by removing an accomodation which had been

      supporting the index format in use. A newer index format is in

      development which is substantially faster. (Robert Collins)

bzr 1.6rc2 2008-08-13

---------------------

This release candidate has a few minor bug fixes, and some regression

fixes for Windows.

  BUG FIXES:

    * ``bzr upgrade`` on remote branches accessed via bzr:// and

      bzr+ssh:// now works.  (Andrew Bennetts)

    * Change the ``get_format_description()`` strings for

      ``RepositoryFormatKnitPack5`` et al to be single line messages.

      (Aaron Bentley)

    * Fix for a regression on Win32 where we would try to call

      ``os.listdir()`` on a file and not catch the exception properly.

      (Windows raises a different exception.) This would manifest in

      places like ``bzr rm file`` or ``bzr switch``.

      (Mark Hammond, John Arbash Meinel)

    * ``Inventory.copy()`` was failing to set the revision property for

      the root entry. (Jelmer Vernooij)

    * sftp transport: added missing ``FileExists`` case to

      ``_translate_io_exception`` (Christophe Troestler, #123475)

    * The help for ``bzr ignored`` now suggests ``bzr ls --ignored`` for

      scripting use. (Robert Collins, #3834)

    * The default ``annotate`` logic will now always assign the

      last-modified value of a line to one of the revisions that modified

      it, rather than a merge revision. This would happen when both sides

      claimed to have modified the line resulting in the same text. The

      choice is arbitrary but stable, so merges in different directions

      will get the same results.  (John Arbash Meinel, #232188)

bzr 1.6rc1 2008-08-06

---------------------

This release candidate for bzr 1.6 solidifies the new branch stacking

feature.  Bazaar now recommends that users upgrade all knit repositories,

because later formats are much faster.  However, we plan to continue read/write and

upgrade support for knit repostories for the forseeable future.  Several

other bugs and performance issues were fixed.

  CHANGES:

    * Knit format repositories are deprecated and bzr will now emit

      warnings whenever it encounters one.  Use ``bzr upgrade`` to upgrade

      knit repositories to pack format.  (Andrew Bennetts)

  IMPROVEMENTS:

    * ``bzr check`` can now be told which elements at a location it should

      check.  (Daniel Watkins)

    * Commit now supports ``--exclude`` (or ``-x``) to exclude some files

      from the commit. (Robert Collins, #3117)

    * Fetching data between repositories that have the same model but no 

      optimised fetcher will not reserialise all the revisions, increasing

      performance. (Robert Collins, John Arbash Meinel)

    * Give a more specific error when target branch is not reachable.

      (James Westby)

    * Implemented a custom ``walkdirs_utf8`` implementation for win32.

      This uses a pyrex extension to get direct access to the

      ``FindFirstFileW`` style apis, rather than using ``listdir`` +

      ``lstat``. Shows a very strong improvement in commands like

      ``status`` and ``diff`` which have to iterate the working tree.

      Anywhere from 2x-6x faster depending on the size of the tree (bigger

      trees, bigger benefit.) (John Arbash Meinel)

    * New registry for log properties handles  and the method in 

      LongLogFormatter to display the custom properties returned by the 

      registered handlers. (Guillermo Gonzalez, #162469)

  BUG FIXES:

    * Add more tests that stacking does not create deltas spanning

      physical repository boundaries.

      (Martin Pool, #252428)

    * Better message about incompatible repositories.

      (Martin Pool, #206258)

    * ``bzr branch --stacked`` ensures the destination branch format can

      support stacking, even if the origin does not.

      (Martin Pool)

    * ``bzr export`` no longer exports ``.bzrrules``.

      (Ian Clatworthy)

    * ``bzr serve --directory=/`` now correctly allows the whole

      filesystem to be accessed on Windows, not just the root of the drive

      that Python is running from.

      (Adrian Wilkins, #240910)

    * Deleting directories by hand before running ``bzr rm`` will not

      cause subsequent errors in ``bzr st`` and ``bzr commit``.

      (Robert Collins, #150438)

    * Fix a test case that was failing if encoding wasn't UTF-8.

      (John Arbash Meinel, #247585)

    * Fix "no buffer space available" error when branching with the new

      smart server protocol to or from Windows.

      (Andrew Bennetts, #246180)

    * Fixed problem in branching from smart server.

      (#249256, Michael Hudson, Martin Pool) 

    * Handle a file turning in to a directory in TreeTransform.

      (James Westby, #248448)

  API CHANGES:

    * ``MutableTree.commit`` has an extra optional keywork parameter

      ``exclude`` that will be unconditionally supplied by the command

      line UI - plugins that add tree formats may need an update.

      (Robert Collins)

    * The API minimum version for plugin compatibility has been raised to

      1.6 - there are significant changes throughout the code base.

      (Robert Collins)

    * The generic fetch code now uses three attributes on Repository objects

      to control fetch. The streams requested are controlled via :

      ``_fetch_order`` and ``_fetch_uses_deltas``. Setting these

      appropriately allows different repository implementations to recieve

      data in their optimial form. If the ``_fetch_reconcile`` is set then

      a reconcile operation is triggered at the end of the fetch.

      (Robert Collins)

    * The ``put_on_disk`` and ``get_tar_item`` methods in

      ``InventoryEntry`` were deprecated. (Ian Clatworthy)

    * ``Repository.is_shared`` doesn't take a read lock. It didn't

      need one in the first place (nobody cached the value, and

      ``RemoteRepository`` wasn't taking one either). This saves a round

      trip when probing Pack repositories, as they read the ``pack-names``

      file when locked. And during probe, locking the repo isn't very

      useful. (John Arbash Meinel)

  INTERNALS:

    * ``bzrlib.branchbuilder.BranchBuilder`` is now much more capable of

      putting together a real history without having to create a full

      WorkingTree. It is recommended that tests that are not directly

      testing the WorkingTree use BranchBuilder instead.  See

      ``BranchBuilder.build_snapshot`` or

      ``TestCaseWithMemoryTree.make_branch_builder``.  (John Arbash Meinel)

    * ``bzrlib.builtins.internal_tree_files`` broken into two giving a new

      helper ``safe_relpath_files`` - used by the new ``exclude``

      parameter to commit. (Robert Collins)

    * Make it easier to introduce new WorkingTree formats.

      (Ian Clatworthy)

    * The code for exporting trees was refactored not to use the

      deprecated ``InventoryEntry`` methods. (Ian Clatworthy)

    * RuleSearchers return () instead of [] now when there are no matches.

      (Ian Clatworthy)

bzr 1.6beta3 2008-07-17

-----------------------

This release adds a new 'stacked branches' feature allowing branches to

share storage without being in the same repository or on the same machine.

(See the user guide for more details.)  It also adds a new hook, improved

weaves, aliases for related locations, faster bzr+ssh push, and several

bug fixes.

  FEATURES:

    * New ``pre_change_branch_tip`` hook that is called before the

      branch tip is moved, while the branch is write-locked.  See the User

      Reference for signature details.  (Andrew Bennetts)

    * Rule-based preferences can now be defined for selected files in

      selected branches, allowing commands and plugins to provide

      custom behaviour for files matching defined patterns.

      See ``Rule-based preferences`` (part of ``Configuring Bazaar``)

      in the User Guide and ``bzr help rules`` for more information.

      (Ian Clatworthy)

    * Sites may suggest a branch to stack new branches on.  (Aaron Bentley)

    * Stacked branches are now supported. See ``bzr help branch`` and 

      ``bzr help push``.  Branches must be in the ``development1`` format

      to stack, though the stacked-on branch can be of any format. 

      (Robert Collins)

  IMPROVEMENTS:

    * ``bzr export --format=tgz --root=NAME -`` to export a gzipped tarball 

      to stdout; also ``tar`` and ``tbz2``.

      (Martin Pool)

    * ``bzr (re)merge --weave`` will now use a standard Weave algorithm,

      rather than the annotation-based merge it was using. It does so by

      building up a Weave of the important texts, without needing to build

      the full ancestry. (John Arbash Meinel, #238895)

    * ``bzr send`` documents and better supports ``emacsclient`` (proper

      escaping of mail headers and handling of the MUA Mew).

      (Christophe Troestler)

    * Remembered locations can be specified by aliases, e.g. :parent, :public,

      :submit.  (Aaron Bentley)

    * The smart protocol now has improved support for setting branches'

      revision info directly.  This makes operations like push

      faster.  The new request method name is

      ``Branch.set_last_revision_ex``.  (Andrew Bennetts)

  BUG FIXES:

    * Bazaar is now able to be a client to the web server of IIS 6 and 7.

      The broken implementations of RFC822 in Python and RFC2046 in IIS

      combined with boundary-line checking in Bazaar previously made this

      impossible. (NB, IIS 5 does not suffer from this problem).

      (Adrian Wilkins, #247585)

    * ``bzr log --long`` with a ghost in your mainline now handles that

      ghost properly. (John Arbash Meinel, #243536)

    * ``check`` handles the split-up .bzr layout correctly, so no longer

      requires a branch to be present.

      (Daniel Watkins, #64783)

    * Clearer message about how to set the PYTHONPATH if bzrlib can't be

      loaded. 

      (Martin Pool, #205230)

    * Errors about missing libraries are now shown without a traceback,

      and with a suggestion to install the library.  The full traceback is 

      still in ``.bzr.log`` and can be shown with ``-Derror``.

      (Martin Pool, #240161)

    * Fetch from a stacked branch copies all required data.

      (Aaron Bentley, #248506)

    * Handle urls such as ftp://user@host.com@www.host.com where the user

      name contains an @.

      (Neil Martinsen-Burrell, #228058)

    * ``needs_read_lock`` and ``needs_write_lock`` now suppress an error during

      ``unlock`` if there was an error in the original function. This helps

      most when there is a failure with a smart server action, since often the

      connection closes and we cannot unlock.