~bzr-pqm/bzr/bzr.dev

IN DEVELOPMENT

  IMPROVEMENTS:

    * ``bzr mv`` enhanced to support already moved files.

      In the past the mv command would have failed if the source file doesn't

      exist. In this situation ``bzr mv`` would now detect that the file has

      already moved and update the repository accordingly, if the target file

      does exist.

      A new option ``--after`` has been added so that if two files already

      exist, you could notify Bazaar that you have moved a (versioned) file and

      replaced it with another. Thus in this case ``bzr move --after`` will

      only update the Bazaar identifier.

      (Steffen Eichenberg, Marius Kruger)

    * ``ls`` now works on treeless branches and remote branches.

      (Aaron Bentley)

    * ``bzr help global-options`` describes the global options.

      (Aaron Bentley)

  INTERNALS:

    * Reserved ids (any revision-id ending in a colon) are rejected by

      versionedfiles, repositories, branches, and working trees

      (Aaron Bentley)

  BUGFIXES:

  TESTING:

    * New ``--first`` option to ``bzr selftest`` to run specified tests

      before the rest of the suite.  (Martin Pool)

bzr 0.14  2007-01-23

  IMPROVEMENTS:

    * ``bzr help global-options`` describes the global options. (Aaron Bentley)

  BUG FIXES:

    * Skip documentation generation tests if the tools to do so are not

      available. Fixes running selftest for installled copies of bzr. 

      (John Arbash Meinel, #80330)

    * Fix the code that discovers whether bzr is being run from it's

      working tree to handle the case when it isn't but the directory

      it is in is below a repository. (James Westby, #77306)

bzr 0.14rc1  2007-01-16

  IMPROVEMENTS:

    * New connection: ``bzr+http://`` which supports tunnelling the smart

      protocol over an HTTP connection. If writing is enabled on the bzr

      server, then you can write over the http connection.

      (Andrew Bennetts, John Arbash Meinel)

    * Aliases now support quotation marks, so they can contain whitespace

      (Marius Kruger)

    * PyCurlTransport now use a single curl object. By specifying explicitly

      the 'Range' header, we avoid the need to use two different curl objects

      (and two connections to the same server). (Vincent Ladeuil)

    * ``bzr commit`` does not prompt for a message until it is very likely to

      succeed.  (Aaron Bentley)

    * ``bzr conflicts`` now takes --text to list pathnames of text conflicts

      (Aaron Bentley)

    * Fix ``iter_lines_added_or_present_in_versions`` to use a set instead

      of a list while checking if a revision id was requested. Takes 10s

      off of the ``fileids_affected_by_revision_ids`` time, which is 10s

      of the ``bzr branch`` time. Also improve ``fileids_...`` time by

      filtering lines with a regex rather than multiple ``str.find()``

      calls. (saves another 300ms) (John Arbash Meinel)

    * Policy can be set for each configuration key. This allows keys to be

      inherited properly across configuration entries. For example, this

      should enable you to do::

        [/home/user/project]

        push_location = sftp://host/srv/project/

        push_location:policy = appendpath

      And then a branch like ``/home/user/project/mybranch`` should get an

      automatic push location of ``sftp://host/srv/project/mybranch``.

      (James Henstridge)

    * Added ``bzr status --short`` to make status report svn style flags

      for each file.  For example::

        $ bzr status --short

        A  foo

        A  bar

        D  baz

        ?  wooley

    * 'bzr selftest --clean-output' allows easily clean temporary tests 

      directories without running tests. (Alexander Belchenko)

    * ``bzr help hidden-commands`` lists all hidden commands. (Aaron Bentley)

    * ``bzr merge`` now has an option ``--pull`` to fall back to pull if

      local is fully merged into remote. (Jan Hudec)

    * ``bzr help formats`` describes available directory formats. (Aaron Bentley)

  INTERNALS:

    * A few tweaks directly to ``fileids_affected_by_revision_ids`` to

      help speed up processing, as well allowing to extract unannotated

      lines. Between the two ``fileids_affected_by_revision_ids`` is

      improved by approx 10%. (John Arbash Meinel)

    * Change Revision serialization to only write out millisecond

      resolution. Rather than expecting floating point serialization to

      preserve more resolution than we need. (Henri Weichers, Martin Pool)

    * Test suite ends cleanly on Windows.  (Vincent Ladeuil)

    * When 'encoding_type' attribute of class Command is equal to 'exact', 

      force sys.stdout to be a binary stream on Windows, and therefore

      keep exact line-endings (without LF -> CRLF conversion).

      (Alexander Belchenko)

    * Single-letter short options are no longer globally declared.  (Martin

      Pool)

    * Before using detected user/terminal encoding bzr should check

      that Python has corresponding codec. (Alexander Belchenko)

    * Formats for end-user selection are provided via a FormatRegistry (Aaron Bentley)

  BUG FIXES:

    * ``bzr missing --verbose`` was showing adds/removals in the wrong

      direction. (John Arbash Meinel)

    * ``bzr annotate`` now defaults to showing dotted revnos for merged

      revisions. It cuts them off at a depth of 12 characters, but you can

      supply ``--long`` to see the full number. You can also use

      ``--show-ids`` to display the original revision ids, rather than

      revision numbers and committer names. (John Arbash Meinel, #75637)

    * bzr now supports Win32 UNC path (e.g. \\HOST\path). 

      (Alexander Belchenko, #57869)

    * Win32-specific: output of cat, bundle and diff commands don't mangle

      line-endings (Alexander Belchenko, #55276)

    * Replace broken fnmatch based ignore pattern matching with custom pattern

      matcher.

      (Kent Gibson, Jan Hudec #57637)

    * pycurl and urllib can detect short reads at different places. Update

      the test suite to test more cases. Also detect http error code 416

      which was raised for that specific bug. Also enhance the urllib

      robustness by detecting invalid ranges (and pycurl's one by detecting

      short reads during the initial GET). (Vincent Ladeuil, #73948)

    * The urllib connection sharing interacts badly with urllib2

      proxy setting (the connections didn't go thru the proxy

      anymore). Defining a proper ProxyHandler solves the

      problem.  (Vincent Ladeuil, #74759)

    * Use urlutils to generate relative URLs, not osutils 

      (Aaron Bentley, #76229)

    * ``bzr status`` in a readonly directory should work without giving

      lots of errors. (John Arbash Meinel, #76299)

    * Mention the revisionspec topic for the revision option help.

      (Wouter van Heyst, #31663)

    * Allow plugins import from zip archives.

      (Alexander Belchenko, #68124)

bzr 0.13  2006-12-05

  No changes from 0.13rc1

bzr 0.13rc1  2006-11-27

  IMPROVEMENTS:

    * New command ``bzr remove-tree`` allows the removal of the working

      tree from a branch.

      (Daniel Silverstone)

    * urllib uses shared keep-alive connections, so http 

      operations are substantially faster.

      (Vincent Ladeuil, #53654)

    * ``bzr export`` allows an optional branch parameter, to export a bzr

      tree from some other url. For example:

      ``bzr export bzr.tar.gz http://bazaar-vcs.org/bzr/bzr.dev``

      (Daniel Silverstone)

    * Added ``bzr help topics`` to the bzr help system. This gives a

      location for general information, outside of a specific command.

      This includes updates for ``bzr help revisionspec`` the first topic

      included. (Goffredo Baroncelli, John Arbash Meinel, #42714)

    * WSGI-compatible HTTP smart server.  See ``doc/http_smart_server.txt``.

      (Andrew Bennetts)

    * Knit files will now cache full texts only when the size of the

      deltas is as large as the size of the fulltext. (Or after 200

      deltas, whichever comes first). This has the most benefit on large

      files with small changes, such as the inventory for a large project.

      (eg For a project with 2500 files, and 7500 revisions, it changes

      the size of inventory.knit from 11MB to 5.4MB) (John Arbash Meinel)

  INTERNALS:

    * New -D option given before the command line turns on debugging output

      for particular areas.  -Derror shows tracebacks on all errors.

      (Martin Pool)

    * Clean up ``bzr selftest --benchmark bundle`` to correct an import,

      and remove benchmarks that take longer than 10min to run.

      (John Arbash Meinel)

    * Use ``time.time()`` instead of ``time.clock()`` to decide on

      progress throttling. Because ``time.clock()`` is actually CPU time,

      so over a high-latency connection, too many updates get throttled.

      (John Arbash Meinel)

    * ``MemoryTransport.list_dir()`` would strip the first character for

      files or directories in root directory. (John Arbash Meinel)

    * New ``ChrootTransportDecorator``, accessible via the ``chroot+`` url

      prefix.  It disallows any access to locations above a set URL.  (Andrew

      Bennetts)

  BUG FIXES:

    * Now _KnitIndex properly decode revision ids when loading index data.

      And optimize the knit index parsing code.  (Dmitry Vasiliev, John

      Arbash Meinel)

    * ``bzrlib/bzrdir.py`` was directly referencing ``bzrlib.workingtree``,

      without importing it. This prevented ``bzr upgrade`` from working

      unless a plugin already imported ``bzrlib.workingtree``

      (John Arbash Meinel, #70716)

    * Suppress the traceback on invalid URLs (Vincent Ladeuil, #70803).

    * Give nicer error message when an http server returns a 403

      error code. (Vincent Ladeuil, #57644).

    * When a multi-range http GET request fails, try a single

      range one. If it fails too, forget about ranges. Remember that until 

      the death of the transport and propagates that to the clones.

      (Vincent Ladeuil, #62276, #62029).

    * Handles user/passwords supplied in url from command

      line (for the urllib implementation). Don't request already

      known passwords (Vincent Ladeuil, #42383, #44647, #48527)

    * _KnitIndex.add_versions() dictionary compresses revision ids as they

      are added. This fixes bug where fetching remote revisions records

      them as full references rather than integers. (John Arbash Meinel,

      #64789)

    * ``bzr ignore`` strips trailing slashes in patterns.

      Also ``bzr ignore`` rejects absolute paths. (Kent Gibson, #4559)

    * ``bzr ignore`` takes multiple arguments. (Cheuksan Edward Wang, #29488)

    * mv correctly handles paths that traverse symlinks. 

      (Aaron Bentley, #66964)

    * Give nicer looking error messages when failing to connect over ssh.

      (John Arbash Meinel, #49172)

    * Pushing to a remote branch does not currently update the remote working

      tree. After a remote push, ``bzr status`` and ``bzr diff`` on the remote

      machine now show that the working tree is out of date.

      (Cheuksan Edward Wang #48136)

    * Use patiencediff instead of difflib for determining deltas to insert

      into knits. This avoids the O(N^3) behavior of difflib. Patience

      diff should be O(N^2). (Cheuksan Edward Wang, #65714)

    * Running ``bzr log`` on nonexistent file gives an error instead of the

      entire log history. (Cheuksan Edward Wang #50793)

    * ``bzr cat`` can look up contents of removed or renamed files. If the

      pathname is ambiguous, i.e. the files in the old and new trees have

      different id's, the default is the file in the new tree. The user can

      use "--name-from-revision" to select the file in the old tree.

      (Cheuksan Edward Wang, #30190)

  TESTING:

    * TestingHTTPRequestHandler really handles the Range header

      (previously it was ignoring it and returning the whole file,).

bzr 0.12  2006-10-30

  INTERNALS:

    * Clean up ``bzr selftest --benchmark bundle`` to correct an import,

      and remove benchmarks that take longer than 10min to run.

      (John Arbash Meinel)

bzr 0.12rc1  2006-10-23

  IMPROVEMENTS:

    * ``bzr log`` now shows dotted-decimal revision numbers for all revisions,

      rather than just showing a decimal revision number for revisions on the

      mainline. These revision numbers are not yet accepted as input into bzr

      commands such as log, diff etc. (Robert Collins)

    * revisions can now be specified using dotted-decimal revision numbers.

      For instance, ``bzr diff -r 1.2.1..1.2.3. (Robert Collins)

    * ``bzr help commands`` output is now shorter (Aaron Bentley)

    * ``bzr`` now uses lazy importing to reduce the startup time. This has

      a moderate effect on lots of actions, especially ones that have

      little to do. For example ``bzr rocks`` time is down to 116ms from

      283ms. (John Arbash Meinel)

    * New Registry class to provide name-to-object registry-like support,

      for example for schemes where plugins can register new classes to

      do certain tasks (e.g. log formatters). Also provides lazy registration

      to allow modules to be loaded on request. (John Arbash Meinel, Adeodato

      Simó)

  API INCOMPATABILITY:

    * LogFormatter subclasses show now expect the 'revno' parameter to 

      show() to be a string rather than an int. (Robert Collins)

  INTERNALS:

    * ``TestCase.run_bzr``, ``run_bzr_captured``, and ``run_bzr_subprocess``

      can take a ``working_dir='foo'`` parameter, which will change directory 

      for the command. (John Arbash Meinel)

    * ``bzrlib.lazy_regex.lazy_compile`` can be used to create a proxy

      around a regex, which defers compilation until first use. 

      (John Arbash Meinel)

    * ``TestCase.run_bzr_subprocess`` defaults to supplying the

      ``--no-plugins`` parameter to ensure test reproducability, and avoid

      problems with system-wide installed plugins. (John Arbash Meinel)

    * Unique tree root ids are now supported. Newly created trees still

      use the common root id for compatibility with bzr versions before 0.12.

      (Aaron Bentley)

    * ``WorkingTree.set_root_id(None)`` is now deprecated. Please

      pass in inventory.ROOT_ID if you want the default root id value.

      (Robert Collins, John Arbash Meinel)

    * New method ``WorkingTree.flush()`` which will write the current memory

      inventory out to disk. At the same time, read_working_inventory will

      no longer trash the current tree inventory if it has been modified within

      the current lock, and the tree will now ``flush()`` automatically on

      ``unlock()``. ``WorkingTree.set_root_id()`` has been updated to take

      advantage of this functionality. (Robert Collins, John Arbash Meinel)

    * ``bzrlib.tsort.merge_sorted`` now accepts ``generate_revnos``. This

      parameter will cause it to add another column to its output, which

      contains the dotted-decimal revno for each revision, as a tuple.

      (Robert Collins)

    * ``LogFormatter.show_merge`` is deprecated in favour of

      ``LogFormatter.show_merge_revno``. (Robert Collins)

  BUG FIXES:

    * Avoid circular imports by creating a deprecated function for

      ``bzrlib.tree.RevisionTree``. Callers should have been using

      ``bzrlib.revisontree.RevisionTree`` anyway. (John Arbash Meinel,

      #63360, #66349)

    * Don't use ``socket.MSG_WAITALL`` as it doesn't exist on all

      platforms. (Martin Pool, #66356)

    * Don't require ``Content-Type`` in range responses. Assume they are a

      single range if ``Content-Type`` does not exist.

      (John Arbash Meinel, #62473)

    * bzr branch/pull no longer complain about progress bar cleanup when

      interrupted during fetch.  (Aaron Bentley, #54000)

    * ``WorkingTree.set_parent_trees()`` uses the trees to directly write

      the basis inventory, rather than going through the repository. This

      allows us to have 1 inventory read, and 2 inventory writes when

      committing a new tree. (John Arbash Meinel)

    * When reverting, files that are not locally modified that do not exist

      in the target are deleted, not just unversioned (Aaron Bentley)

    * When trying to acquire a lock, don't fail immediately. Instead, try

      a few times (up to 1 hour) before timing out. Also, report why the

      lock is unavailable (John Arbash Meinel, #43521, #49556)

    * Leave HttpTransportBase daughter classes decides how they

      implement cloning. (Vincent Ladeuil, #61606)

    * diff3 does not indicate conflicts on clean merge. (Aaron Bentley)

    * If a commit fails, the commit message is stored in a file at the root of

      the tree for later commit. (Cheuksan Edward Wang, Stefan Metzmacher,

      #32054)

  TESTING:

    * New test base class TestCaseWithMemoryTransport offers memory-only

      testing facilities: its not suitable for tests that need to mutate disk

      state, but most tests should not need that and should be converted to

      TestCaseWithMemoryTransport. (Robert Collins)

    * ``TestCase.make_branch_and_memory_tree`` now takes a format

      option to set the BzrDir, Repository and Branch formats of the

      created objects. (Robert Collins, John Arbash Meinel)

bzr 0.11  2006-10-02

    * Smart server transport test failures on windows fixed. (Lukáš Lalinský).

bzr 0.11rc2  2006-09-27

  BUG FIXES:

    * Test suite hangs on windows fixed. (Andrew Bennets, Alexander Belchenko).

    * Commit performance regression fixed. (Aaron Bentley, Robert Collins, John

      Arbash Meinel).

bzr 0.11rc1  2006-09-25

  IMPROVEMENTS:

    * Knit files now wait to create their contents until the first data is

      added. The old code used to create an empty .knit and a .kndx with just

      the header. However, this caused a lot of extra round trips over sftp.

      This can change the time for ``bzr push`` to create a new remote branch

      from 160s down to 100s. This also affects ``bzr commit`` performance when

      adding new files, ``bzr commit`` on a new kernel-like tree drops from 50s

      down to 40s (John Arbash Meinel, #44692)

    * When an entire subtree has been deleted, commit will now report that

      just the top of the subtree has been deleted, rather than reporting

      all the individual items. (Robert Collins)

    * Commit performs one less XML parse. (Robert Collins)

    * ``bzr checkout`` now operates on readonly branches as well

      as readwrite branches. This fixes bug #39542. (Robert Collins)

    * ``bzr bind`` no longer synchronises history with the master branch.

      Binding should be followed by an update or push to synchronise the 

      two branches. This is closely related to the fix for bug #39542.

      (Robert Collins)

    * ``bzrlib.lazy_import.lazy_import`` function to create on-demand 

      objects.  This allows all imports to stay at the global scope, but

      modules will not actually be imported if they are not used.

      (John Arbash Meinel)

    * Support bzr:// and bzr+ssh:// urls to work with the new RPC-based

      transport which will be used with the upcoming high-performance smart

      server. The new command ``bzr serve`` will invoke bzr in server mode,

      which processes these requests. (Andrew Bennetts, Robert Collins, Martin

      Pool)

    * New command ``bzr version-info`` which can be used to get a summary

      of the current state of the tree. This is especially useful as part

      of a build commands. See ``doc/version_info.txt`` for more information 

      (John Arbash Meinel)

  BUG FIXES:

    * 'bzr inventory [FILE...]' allows restricting the file list to a

      specific set of files. (John Arbash Meinel, #3631)

    * Don't abort when annotating empty files (John Arbash Meinel, #56814)

    * Add ``Stanza.to_unicode()`` which can be passed to another Stanza

      when nesting stanzas. Also, add ``read_stanza_unicode`` to handle when

      reading a nested Stanza. (John Arbash Meinel)

    * Transform._set_mode() needs to stat the right file. 

      (John Arbash Meinel, #56549)

    * Raise WeaveFormatError rather than StopIteration when trying to read

      an empty Weave file. (John Arbash Meinel, #46871)

    * Don't access e.code for generic URLErrors, only HTTPErrors have .code.

      (Vincent Ladeuil, #59835)

    * Handle boundary="" lines properly to allow access through a Squid proxy.

      (John Arbash Meinel, #57723)

    * revert now removes newly-added directories (Aaron Bentley, #54172)

    * ``bzr upgrade sftp://`` shouldn't fail to upgrade v6 branches if there 

      isn't a working tree. (David Allouche, #40679)

    * Give nicer error messages when a user supplies an invalid --revision

      parameter. (John Arbash Meinel, #55420)

    * Handle when LANG is not recognized by python. Emit a warning, but

      just revert to using 'ascii'. (John Arbash Meinel, #35392)

    * Don't use preexec_fn on win32, as it is not supported by subprocess.

      (John Arbash Meinel)

    * Skip specific tests when the dependencies aren't met. This includes

      some ``setup.py`` tests when ``python-dev`` is not available, and

      some tests that depend on paramiko. (John Arbash Meinel, Mattheiu Moy)

    * Fallback to Paramiko properly, if no ``ssh`` executable exists on

      the system. (Andrew Bennetts, John Arbash Meinel)

    * ``Branch.bind(other_branch)`` no longer takes a write lock on the

      other branch, and will not push or pull between the two branches.

      API users will need to perform a push or pull or update operation if they

      require branch synchronisation to take place. (Robert Collins, #47344)

    * When creating a tarball or zipfile export, export unicode names as utf-8

      paths. This may not work perfectly on all platforms, but has the best

      chance of working in the common case. (John Arbash Meinel, #56816)

    * When committing, only files that exist in working tree or basis tree

      may be specified (Aaron Bentley, #50793)

  PORTABILITY:

    * Fixes to run on Python 2.5 (Brian M. Carlson, Martin Pool, Marien Zwart)

  INTERNALS:

    * TestCaseInTempDir now creates a separate directory for HOME, rather

      than having HOME set to the same location as the working directory.

      (John Arbash Meinel)

    * run_bzr_subprocess() can take an optional 'env_changes={}' parameter,

      which will update os.environ inside the spawned child. It also can

      take a 'universal_newlines=True', which helps when checking the output

      of the command. (John Arbash Meinel)

    * Refactor SFTP vendors to allow easier re-use when ssh is used. 

      (Andrew Bennetts)

    * Transport.list_dir() and Transport.iter_files_recursive() should always

      return urlescaped paths. This is now tested (there were bugs in a few

      of the transports) (Andrew Bennetts, David Allouche, John Arbash Meinel)

    * New utility function symbol_versioning.deprecation_string. Returns the

      formatted string for a callable, deprecation format pair. (Robert Collins)

    * New TestCase helper applyDeprecated. This allows you to call a callable

      which is deprecated without it spewing to the screen, just by supplying

      the deprecation format string issued for it. (Robert Collins)

    * Transport.append and Transport.put have been deprecated in favor of

      .append_bytes, .append_file, .put_bytes, and .put_file. This removes the

      ambiguity in what type of object the functions take.

      Transport.non_atomic_put_{bytes,file} has also been added. Which works

      similarly to Transport.append() except for SFTP, it doesn't have a round

      trip when opening the file. Also, it provides functionality for creating

      a parent directory when trying to create a file, rather than raise

      NoSuchFile and forcing the caller to repeat their request.

      (John Arbash Meinel)

    * WorkingTree has a new api ``unversion`` which allow the unversioning of

      entries by their file id. (Robert Collins)

    * WorkingTree.pending_merges is deprecated.  Please use the get_parent_ids

      (introduced in 0.10) method instead. (Robert Collins)

    * WorkingTree has a new lock_tree_write method which locks the branch for

      read rather than write. This is appropriate for actions which only need

      the branch data for reference rather than mutation. A new decorator

      needs_tree_write_lock is provided in the workingtree module. Like the

      needs_read_lock and needs_write_lock decorators this allows static 

      declaration of the locking requirements of a function to ensure that

      a lock is taken out for casual scripts. (Robert Collins, #54107)

    * All WorkingTree methods which write to the tree, but not to the branch

      have been converted to use ``needs_tree_write_lock`` rather than 

      ``needs_write_lock``. Also converted is the revert, conflicts and tree

      transform modules. This provides a modest performance improvement on 

      metadir style trees, due to the reduce lock-acquisition, and a more

      significant performance improvement on lightweight checkouts from 

      remote branches, where trivial operations used to pay a significant 

      penalty. It also provides the basis for allowing readonly checkouts.

      (Robert Collins)

    * Special case importing the standard library 'copy' module. This shaves

      off 40ms of startup time, while retaining compatibility. See:

      ``bzrlib/inspect_for_copy.py`` for more details. (John Arbash Meinel)

    * WorkingTree has a new parent class MutableTree which represents the 

      specialisations of Tree which are able to be altered. (Robert Collins)

    * New methods mkdir and put_file_bytes_non_atomic on MutableTree that

      mutate the tree and its contents. (Robert Collins)

    * Transport behaviour at the root of the URL is now defined and tested.

      (Andrew Bennetts, Robert Collins)

  TESTING:

    * New test helper classs MemoryTree. This is typically accessed via

      ``self.make_branch_and_memory_tree()`` in test cases. (Robert Collins)

    * Add start_bzr_subprocess and stop_bzr_subprocess to allow test code to

      continue running concurrently with a subprocess of bzr. (Andrew Bennetts,

      Robert Collins)

    * Add a new method ``Transport.get_smart_client()``. This is provided to

      allow upgrades to a richer interface than the VFS one provided by

      Transport. (Andrew Bennetts, Martin Pool)

bzr 0.10  2006-08-29

  IMPROVEMENTS:

    * 'merge' now takes --uncommitted, to apply uncommitted changes from a

      tree.  (Aaron Bentley)

    * 'bzr add --file-ids-from' can be used to specify another path to use

      for creating file ids, rather than generating all new ones. Internally,

      the 'action' passed to smart_add_tree() can return file_ids that

      will be used, rather than having bzrlib generate new ones.

      (John Arbash Meinel, #55781)

    * ``bzr selftest --benchmark`` now allows a ``--cache-dir`` parameter.

      This will cache some of the intermediate trees, and decrease the

      setup time for benchmark tests. (John Arbash Meinel)

    * Inverse forms are provided for all boolean options.  For example,

      --strict has --no-strict, --no-recurse has --recurse (Aaron Bentley)

    * Serialize out Inventories directly, rather than using ElementTree.

      Writing out a kernel sized inventory drops from 2s down to ~350ms.

      (Robert Collins, John Arbash Meinel)

  BUG FIXES:

    * Help diffutils 2.8.4 get along with binary tests (Marien Zwart: #57614)