get_expanded_map(): avoid memory leak

[git/mjg.git] / Documentation / RelNotes / 1.8.5.txt

Git v1.8.5 Release Notes
========================

Backward compatibility notes (for Git 2.0)
------------------------------------------

When "git push [$there]" does not say what to push, we have used the
traditional "matching" semantics so far (all your branches were sent
to the remote as long as there already are branches of the same name
over there).  In Git 2.0, the default will change to the "simple"
semantics that pushes:

 - only the current branch to the branch with the same name, and only
   when the current branch is set to integrate with that remote
   branch, if you are pushing to the same remote as you fetch from; or

 - only the current branch to the branch with the same name, if you
   are pushing to a remote that is not where you usually fetch from.

Use the user preference configuration variable "push.default" to
change this.  If you are an old-timer who is used to the "matching"
semantics, you can set the variable to "matching" to keep the
traditional behaviour.  If you want to live in the future early, you
can set it to "simple" today without waiting for Git 2.0.

When "git add -u" (and "git add -A") is run inside a subdirectory and
does not specify which paths to add on the command line, it
will operate on the entire tree in Git 2.0 for consistency
with "git commit -a" and other commands.  There will be no
mechanism to make plain "git add -u" behave like "git add -u .".
Current users of "git add -u" (without a pathspec) should start
training their fingers to explicitly say "git add -u ."
before Git 2.0 comes.  A warning is issued when these commands are
run without a pathspec and when you have local changes outside the
current directory, because the behaviour in Git 2.0 will be different
from today's version in such a situation.

In Git 2.0, "git add <path>" will behave as "git add -A <path>", so
that "git add dir/" will notice paths you removed from the directory
and record the removal.  Versions before Git 2.0, including this
release, will keep ignoring removals, but the users who rely on this
behaviour are encouraged to start using "git add --ignore-removal <path>"
now before 2.0 is released.

The default prefix for "git svn" will change in Git 2.0.  For a long
time, "git svn" created its remote-tracking branches directly under
refs/remotes, but it will place them under refs/remotes/origin/ unless
it is told otherwise with its --prefix option.


Updates since v1.8.4
--------------------

Foreign interfaces, subsystems and ports.

 * "git-svn" used with SVN 1.8.0 when talking over https:// connection
   dumped core due to a bug in the serf library that SVN uses.  Work
   it around on our side, even though the SVN side is being fixed.

 * On MacOS X, we detected if the filesystem needs the "pre-composed
   unicode strings" workaround, but did not automatically enable it.
   Now we do.

 * remote-hg remote helper misbehaved when interacting with a local Hg
   repository relative to the home directory, e.g. "clone hg::~/there".

 * imap-send ported to OS X uses Apple's security framework instead of
   OpenSSL one.

 * Subversion 1.8.0 that was recently released breaks older subversion
   clients coming over http/https in various ways.

 * "git fast-import" treats an empty path given to "ls" as the root of
   the tree.


UI, Workflows & Features

 * "git grep" and "git show" pays attention to "--textconv" option
   when these commands are told to operate on blob objects (e.g. "git
   grep -e pattern HEAD:Makefile").

 * "git replace" helper no longer allows an object to be replaced with
   another object of a different type to avoid confusion (you can
   still manually craft such replacement using "git update-ref", as an
   escape hatch).

 * "git status" no longer prints dirty status information for
   submodules for which submodule.$name.ignore is set to "all".

 * "git rebase -i" honours core.abbrev when preparing the insn sheet
   for editing.

 * "git status" during a cherry-pick shows what original commit is
   being picked.

 * Instead of typing four capital letters "HEAD", you can say "@" now,
   e.g. "git log @".

 * "git check-ignore" follows the same rule as "git add" and "git
   status" in that the ignore/exclude mechanism does not take effect
   on paths that are already tracked.  With "--no-index" option, it
   can be used to diagnose which paths that should have been ignored
   have been mistakenly added to the index.

 * Some irrelevant "advice" messages that are shared with "git status"
   output have been removed from the commit log template.

 * "update-refs" learnt a "--stdin" option to read multiple update
   requests and perform them in an all-or-none fashion.

 * Just like "make -C <directory>", "git -C <directory> ..." tells Git
   to go there before doing anything else.

 * Just like "git checkout -" knows to check out and "git merge -"
   knows to merge the branch you were previously on, "git cherry-pick"
   now understands "git cherry-pick -" to pick from the previous
   branch.

 * "git status" now omits the prefix to make its output a comment in a
   commit log editor, which is not necessary for human consumption.
   Scripts that parse the output of "git status" are advised to use
   "git status --porcelain" instead, as its format is stable and easier
   to parse.

 * Make "foo^{tag}" to peel a tag to itself, i.e. no-op., and fail if
   "foo" is not a tag.  "git rev-parse --verify v1.0^{tag}" would be
   a more convenient way to say "test $(git cat-file -t v1.0) = tag".

 * "git branch -v -v" (and "git status") did not distinguish among a
   branch that does not build on any other branch, a branch that is in
   sync with the branch it builds on, and a branch that is configured
   to build on some other branch that no longer exists.

 * A packfile that stores the same object more than once is broken and
   will be rejected by "git index-pack" that is run when receiving
   data over the wire.

 * Earlier we started rejecting an attempt to add 0{40} object name to
   the index and to tree objects, but it sometimes is necessary to
   allow so to be able to use tools like filter-branch to correct such
   broken tree objects.  "filter-branch" can again be used to to do
   so.

 * "git config" did not provide a way to set or access numbers larger
   than a native "int" on the platform; it now provides 64-bit signed
   integers on all platforms.

 * "git pull --rebase" always chose to do the bog-standard flattening
   rebase.  You can tell it to run "rebase --preserve-merges" by
   setting "pull.rebase" configuration to "preserve".

 * "git push --no-thin" actually disables the "thin pack transfer"
   optimization.

 * Magic pathspecs like ":(icase)makefile" that matches both
   Makefile and makefile can be used in more places.

 * The "http.*" variables can now be specified per URL that the
   configuration applies.  For example,

   [http]
       sslVerify = true
   [http "https://weak.example.com/"]
       sslVerify = false

   would flip http.sslVerify off only when talking to that specified
   site.

 * "git mv A B" when moving a submodule A has been taught to
   relocate its working tree and to adjust the paths in the
   .gitmodules file.

 * "git blame" can now take more than one -L option to discover the
   origin of multiple blocks of the lines.

 * The http transport clients can optionally ask to save cookies
   with http.savecookies configuration variable.

 * "git push" learned a more fine grained control over a blunt
   "--force" when requesting a non-fast-forward update with the
   "--force-with-lease=<refname>:<expected object name>" option.

 * "git diff --diff-filter=<classes of changes>" can now take
   lowercase letters (e.g. "--diff-filter=d") to mean "show
   everything but these classes".  "git diff-files -q" is now a
   deprecated synonym for "git diff-files --diff-filter=d".

 * "git fetch" (hence "git pull" as well) learned to check
   "fetch.prune" and "remote.*.prune" configuration variables and
   to behave as if the "--prune" command line option was given.

 * "git check-ignore -z" applied the NUL termination to both its input
   (with --stdin) and its output, but "git check-attr -z" ignored the
   option on the output side. Make both honor -z on the input and
   output side the same way.

 * "git whatchanged" may still be used by old timers, but mention of
   it in documents meant for new users will only waste readers' time
   wonderig what the difference is between it and "git log".  Make it
   less prominent in the general part of the documentation and explain
   that it is merely a "git log" with different default behaviour in
   its own document.


Performance, Internal Implementation, etc.

 * "git repack" is now written in C.

 * Build procedure for MSVC has been updated.

 * If a build-time fallback is set to "cat" instead of "less", we
   should apply the same "no subprocess or pipe" optimization as we
   apply to user-supplied GIT_PAGER=cat.

 * Many commands use --dashed-option as a operation mode selector
   (e.g. "git tag --delete") that the user can use at most one
   (e.g. "git tag --delete --verify" is a nonsense) and you cannot
   negate (e.g. "git tag --no-delete" is a nonsense).  parse-options
   API learned a new OPT_CMDMODE macro to make it easier to implement
   such a set of options.

 * OPT_BOOLEAN() in parse-options API was misdesigned to be "counting
   up" but many subcommands expect it to behave as "on/off". Update
   them to use OPT_BOOL() which is a proper boolean.

 * "git gc" exits early without doing a double-work when it detects
   that another instance of itself is already running.

 * Under memory pressure and/or file descriptor pressure, we used to
   close pack windows that are not used and also closed filehandle to
   an open but unused packfiles. These are now controlled separately
   to better cope with the load.

Also contains various documentation updates and code clean-ups.


Fixes since v1.8.4
------------------

Unless otherwise noted, all the fixes since v1.8.4 in the maintenance
track are contained in this release (see release notes to them for
details).

 * "git clone" gave some progress messages to the standard output, not
   to the standard error, and did not allow suppressing them with the
   --no-progress option.
   (merge 643f918 jk/clone-progress-to-stderr later to maint).

 * "format-patch --from=<whom>" forgot to omit unnecessary in-body
   from line, i.e. when <whom> is the same as the real author.
   (merge 662cc30 jk/format-patch-from later to maint).

 * "git shortlog" used to choke and die when there is a malformed
   commit (e.g. missing authors); it now simply ignore such a commit
   and keeps going.
   (merge cd4f09e jk/shortlog-tolerate-broken-commit later to maint).

 * "git merge-recursive" did not parse its "--diff-algorithm=" command
   line option correctly.
   (merge 6562928 jk/diff-algo later to maint).

 * When running "fetch -q", a long silence while the sender side
   computes the set of objects to send can be mistaken by proxies as
   dropped connection.  The server side has been taught to send a
   small empty messages to keep the connection alive.
   (merge 115dedd jk/upload-pack-keepalive later to maint).

 * "git rebase" had a portability regression in v1.8.4 to trigger a
   bug in some BSD shell implementations.
   (merge 99855dd mm/rebase-continue-freebsd-WB later to maint).

 * "git branch --track" had a minor regression in v1.8.3.2 and later
   that made it impossible to base your local work on anything but a
   local branch of the upstream repository you are tracking from.
   (merge b0f49ff jh/checkout-auto-tracking later to maint).

 * When the webserver responds with "405 Method Not Allowed", "git
   http-backend" should tell the client what methods are allowed with
   the "Allow" header.
   (merge 9247be0 bc/http-backend-allow-405 later to maint).

 * When there is no sufficient overlap between old and new history
   during a "git fetch" into a shallow repository, objects that the
   sending side knows the receiving end has were unnecessarily sent.
   (merge f21d2a7 nd/fetch-into-shallow later to maint).

 * "git cvsserver" computed the permission mode bits incorrectly for
   executable files.
   (merge 1b48d56 jc/cvsserver-perm-bit-fix later to maint).

 * When send-email comes up with an error message to die with upon
   failure to start an SSL session, it tried to read the error string
   from a wrong place.
   (merge 6cb0c88 bc/send-email-ssl-die-message-fix later to maint).

 * The implementation of "add -i" has a crippling code to work around
   ActiveState Perl limitation but it by mistake also triggered on Git
   for Windows where MSYS perl is used.
   (merge df17e77 js/add-i-mingw later to maint).

 * We made sure that we notice the user-supplied GIT_DIR is actually a
   gitfile, but did not do the same when the default ".git" is a
   gitfile.
   (merge 487a2b7 nd/git-dir-pointing-at-gitfile later to maint).

 * When an object is not found after checking the packfiles and then
   loose object directory, read_sha1_file() re-checks the packfiles to
   prevent racing with a concurrent repacker; teach the same logic to
   has_sha1_file().
   (merge 45e8a74 jk/has-sha1-file-retry-packed later to maint).

 * "git commit --author=$name", when $name is not in the canonical
   "A. U. Thor <au.thor@example.xz>" format, looks for a matching name
   from existing history, but did not consult mailmap to grab the
   preferred author name.
   (merge ea16794 ap/commit-author-mailmap later to maint).

 * "git ls-files -k" needs to crawl only the part of the working tree
   that may overlap the paths in the index to find killed files, but
   shared code with the logic to find all the untracked files, which
   made it unnecessarily inefficient.
   (merge 680be04 jc/ls-files-killed-optim later to maint).

 * The commit object names in the insn sheet that was prepared at the
   beginning of "rebase -i" session can become ambiguous as the
   rebasing progresses and the repository gains more commits. Make
   sure the internal record is kept with full 40-hex object names.
   (merge 75c6976 es/rebase-i-no-abbrev later to maint).

 * "git rebase --preserve-merges" internally used the merge machinery
   and as a side effect, left merge summary message in the log, but
   when rebasing, there should not be a need for merge summary.
   (merge a9f739c rt/rebase-p-no-merge-summary later to maint).

 * A call to xread() was used without a loop around to cope with short
   read in the codepath to stream new contents to a pack.
   (merge e92527c js/xread-in-full later to maint).

 * "git rebase -i" forgot that the comment character can be
   configurable while reading its insn sheet.
   (merge 7bca7af es/rebase-i-respect-core-commentchar later to maint).

 * The mailmap support code read past the allocated buffer when the
   mailmap file ended with an incomplete line.
   (merge f972a16 jk/mailmap-incomplete-line later to maint).

 * We used to send a large request to read(2)/write(2) as a single
   system call, which was bad from the latency point of view when
   the operation needs to be killed, and also triggered an error on
   broken 64-bit systems that refuse to take more than 2GB read or
   write in one go.
   (merge a487916 sp/clip-read-write-to-8mb later to maint).

 * "git fetch" that auto-followed tags incorrectly reused the
   connection with Git-aware transport helper (like the sample "ext::"
   helper shipped with Git).
   (merge 0f73f8b jc/transport-do-not-use-connect-twice-in-fetch later to maint).

 * "git log --full-diff -- <pathspec>" showed a huge diff for paths
   outside the given <pathspec> for each commit, instead of showing
   the change relative to the parent of the commit.  "git reflog -p"
   had a similar problem.
   (merge 838f9a1 tr/log-full-diff-keep-true-parents later to maint).

 * Setting submodule.*.path configuration variable to true (without
   giving "= value") caused Git to segfault.
   (merge 4b05440 jl/some-submodule-config-are-not-boolean later to maint).

 * "git rebase -i" (there could be others, as the root cause is pretty
   generic) fed a random, data dependeant string to 'echo' and
   expects it to come out literally, corrupting its error message.
   (merge 89b0230 mm/no-shell-escape-in-die-message later to maint).

 * Some people still use rather old versions of bash, which cannot
   grok some constructs like 'printf -v varname' the prompt and
   completion code started to use recently.
   (merge a44aa69 bc/completion-for-bash-3.0 later to maint).

 * Code to read configuration from a blob object did not compile on
   platforms with fgetc() etc. implemented as macros.
   (merge 49d6cfa hv/config-from-blob later to maint-1.8.3).

 * The recent "short-cut clone connectivity check" topic broke a
   shallow repository when a fetch operation tries to auto-follow tags.
   (merge 6da8bdc nd/fetch-pack-shallow-fix later to maint-1.8.3).