D7net Mini Sh3LL v1
Current File : //usr/share/systemd/../doc/info/../librtmp1/../bolt/../git/RelNotes/1.8.2.txt |
Git v1.8.2 Release Notes
========================
Backward compatibility notes (this release)
-------------------------------------------
"git push $there tag v1.2.3" used to allow replacing a tag v1.2.3
that already exists in the repository $there, if the rewritten tag
you are pushing points at a commit that is a descendant of a commit
that the old tag v1.2.3 points at. This was found to be error prone
and starting with this release, any attempt to update an existing
ref under refs/tags/ hierarchy will fail, without "--force".
When "git add -u" and "git add -A" that does not specify what paths
to add on the command line is run from inside a subdirectory, the
scope of the operation has always been limited to the subdirectory.
Many users found this counter-intuitive, given that "git commit -a"
and other commands operate on the entire tree regardless of where you
are. In this release, these commands give a warning message that
suggests the users to use "git add -u/-A ." when they want to limit
the scope to the current directory; doing so will squelch the message,
while training their fingers.
Backward compatibility notes (for Git 2.0)
------------------------------------------
When "git push [$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 the current branch to the branch with the same
name, only when the current branch is set to integrate with that
remote branch. There is a 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 it 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", that does not specify what paths
to add on the command line is run from inside a subdirectory, these
commands will operate on the entire tree in Git 2.0 for consistency
with "git commit -a" and other commands. Because there will be no
mechanism to make "git add -u" behave as if "git add -u .", it is
important for those who are used to "git add -u" (without pathspec)
updating the index only for paths in the current subdirectory to start
training their fingers to explicitly say "git add -u ." when they mean
it before Git 2.0 comes.
Updates since v1.8.1
--------------------
UI, Workflows & Features
* Initial ports to QNX and z/OS UNIX System Services have started.
* Output from the tests is coloured using "green is okay, yellow is
questionable, red is bad and blue is informative" scheme.
* Mention of "GIT/Git/git" in the documentation have been updated to
be more uniform and consistent. The name of the system and the
concept it embodies is "Git"; the command the users type is "git".
All-caps "GIT" was merely a way to imitate "Git" typeset in small
caps in our ASCII text only documentation and to be avoided.
* The completion script (in contrib/completion) used to let the
default completer to suggest pathnames, which gave too many
irrelevant choices (e.g. "git add" would not want to add an
unmodified path). It learnt to use a more git-aware logic to
enumerate only relevant ones.
* In bare repositories, "git shortlog" and other commands now read
mailmap files from the tip of the history, to help running these
tools in server settings.
* Color specifiers, e.g. "%C(blue)Hello%C(reset)", used in the
"--format=" option of "git log" and friends can be disabled when
the output is not sent to a terminal by prefixing them with
"auto,", e.g. "%C(auto,blue)Hello%C(auto,reset)".
* Scripts can ask Git that wildcard patterns in pathspecs they give do
not have any significance, i.e. take them as literal strings.
* The patterns in .gitignore and .gitattributes files can have **/,
as a pattern that matches 0 or more levels of subdirectory.
E.g. "foo/**/bar" matches "bar" in "foo" itself or in a
subdirectory of "foo".
* When giving arguments without "--" disambiguation, object names
that come earlier on the command line must not be interpretable as
pathspecs and pathspecs that come later on the command line must
not be interpretable as object names. This disambiguation rule has
been tweaked so that ":/" (no other string before or after) is
always interpreted as a pathspec; "git cmd -- :/" is no longer
needed, you can just say "git cmd :/".
* Various "hint" lines Git gives when it asks the user to edit
messages in the editor are commented out with '#' by default. The
core.commentchar configuration variable can be used to customize
this '#' to a different character.
* "git add -u" and "git add -A" without pathspec issues warning to
make users aware that they are only operating on paths inside the
subdirectory they are in. Use ":/" (everything from the top) or
"." (everything from the $cwd) to disambiguate.
* "git blame" (and "git diff") learned the "--no-follow" option.
* "git branch" now rejects some nonsense combinations of command line
arguments (e.g. giving more than one branch name to rename) with
more case-specific error messages.
* "git check-ignore" command to help debugging .gitignore files has
been added.
* "git cherry-pick" can be used to replay a root commit to an unborn
branch.
* "git commit" can be told to use --cleanup=whitespace by setting the
configuration variable commit.cleanup to 'whitespace'.
* "git diff" and other Porcelain commands can be told to use a
non-standard algorithm by setting diff.algorithm configuration
variable.
* "git fetch --mirror" and fetch that uses other forms of refspec
with wildcard used to attempt to update a symbolic ref that match
the wildcard on the receiving end, which made little sense (the
real ref that is pointed at by the symbolic ref would be updated
anyway). Symbolic refs no longer are affected by such a fetch.
* "git format-patch" now detects more cases in which a whole branch
is being exported, and uses the description for the branch, when
asked to write a cover letter for the series.
* "git format-patch" learned "-v $count" option, and prepends a
string "v$count-" to the names of its output files, and also
automatically sets the subject prefix to "PATCH v$count". This
allows patches from rerolled series to be stored under different
names and makes it easier to reuse cover letter messages.
* "git log" and friends can be told with --use-mailmap option to
rewrite the names and email addresses of people using the mailmap
mechanism.
* "git log --cc --graph" now shows the combined diff output with the
ancestry graph.
* "git log --grep=<pattern>" honors i18n.logoutputencoding to look
for the pattern after fixing the log message to the specified
encoding.
* "git mergetool" and "git difftool" learned to list the available
tool backends in a more consistent manner.
* "git mergetool" is aware of TortoiseGitMerge now and uses it over
TortoiseMerge when available.
* "git push" now requires "-f" to update a tag, even if it is a
fast-forward, as tags are meant to be fixed points.
* Error messages from "git push" when it stops to prevent remote refs
from getting overwritten by mistake have been improved to explain
various situations separately.
* "git push" will stop without doing anything if the new "pre-push"
hook exists and exits with a failure.
* When "git rebase" fails to generate patches to be applied (e.g. due
to oom), it failed to detect the failure and instead behaved as if
there were nothing to do. A workaround to use a temporary file has
been applied, but we probably would want to revisit this later, as
it hurts the common case of not failing at all.
* Input and preconditions to "git reset" has been loosened where
appropriate. "git reset $fromtree Makefile" requires $fromtree to
be any tree (it used to require it to be a commit), for example.
"git reset" (without options or parameters) used to error out when
you do not have any commits in your history, but it now gives you
an empty index (to match non-existent commit you are not even on).
* "git status" says what branch is being bisected or rebased when
able, not just "bisecting" or "rebasing".
* "git submodule" started learning a new mode to integrate with the
tip of the remote branch (as opposed to integrating with the commit
recorded in the superproject's gitlink).
* "git upload-pack" which implements the service "ls-remote" and
"fetch" talk to can be told to hide ref hierarchies the server
side internally uses (and that clients have no business learning
about) with transfer.hiderefs configuration.
Foreign Interface
* "git fast-export" has been updated for its use in the context of
the remote helper interface.
* A new remote helper to interact with bzr has been added to contrib/.
* "git p4" got various bugfixes around its branch handling. It is
also made usable with Python 2.4/2.5. In addition, its various
portability issues for Cygwin have been addressed.
* The remote helper to interact with Hg in contrib/ has seen a few
fixes.
Performance, Internal Implementation, etc.
* "git fsck" has been taught to be pickier about entries in tree
objects that should not be there, e.g. ".", ".git", and "..".
* Matching paths with common forms of pathspecs that contain wildcard
characters has been optimized further.
* We stopped paying attention to $GIT_CONFIG environment that points
at a single configuration file from any command other than "git config"
quite a while ago, but "git clone" internally set, exported, and
then unexported the variable during its operation unnecessarily.
* "git reset" internals has been reworked and should be faster in
general. We tried to be careful not to break any behaviour but
there could be corner cases, especially when running the command
from a conflicted state, that we may have missed.
* The implementation of "imap-send" has been updated to reuse xml
quoting code from http-push codepath, and lost a lot of unused
code.
* There is a simple-minded checker for the test scripts in t/
directory to catch most common mistakes (it is not enabled by
default).
* You can build with USE_WILDMATCH=YesPlease to use a replacement
implementation of pattern matching logic used for pathname-like
things, e.g. refnames and paths in the repository. This new
implementation is not expected change the existing behaviour of Git
in this release, except for "git for-each-ref" where you can now
say "refs/**/master" and match with both refs/heads/master and
refs/remotes/origin/master. We plan to use this new implementation
in wider places (e.g. "git ls-files '**/Makefile' may find Makefile
at the top-level, and "git log '**/t*.sh'" may find commits that
touch a shell script whose name begins with "t" at any level) in
future versions of Git, but we are not there yet. By building with
USE_WILDMATCH, using the resulting Git daily and reporting when you
find breakages, you can help us get closer to that goal.
* Some reimplementations of Git do not write all the stat info back
to the index due to their implementation limitations (e.g. jgit).
A configuration option can tell Git to ignore changes to most of
the stat fields and only pay attention to mtime and size, which
these implementations can reliably update. This can be used to
avoid excessive revalidation of contents.
* Some platforms ship with old version of expat where xmlparse.h
needs to be included instead of expat.h; the build procedure has
been taught about this.
* "make clean" on platforms that cannot compute header dependencies
on the fly did not work with implementations of "rm" that do not
like an empty argument list.
Also contains minor documentation updates and code clean-ups.
Fixes since v1.8.1
------------------
Unless otherwise noted, all the fixes since v1.8.1 in the maintenance
track are contained in this release (see release notes to them for
details).
* An element on GIT_CEILING_DIRECTORIES list that does not name the
real path to a directory (i.e. a symbolic link) could have caused
the GIT_DIR discovery logic to escape the ceiling.
* When attempting to read the XDG-style $HOME/.config/git/config and
finding that $HOME/.config/git is a file, we gave a wrong error
message, instead of treating the case as "a custom config file does
not exist there" and moving on.
* The behaviour visible to the end users was confusing, when they
attempt to kill a process spawned in the editor that was in turn
launched by Git with SIGINT (or SIGQUIT), as Git would catch that
signal and die. We ignore these signals now.
(merge 0398fc34 pf/editor-ignore-sigint later to maint).
* A child process that was killed by a signal (e.g. SIGINT) was
reported in an inconsistent way depending on how the process was
spawned by us, with or without a shell in between.
* After failing to create a temporary file using mkstemp(), failing
pathname was not reported correctly on some platforms.
* We used to stuff "user@" and then append what we read from
/etc/mailname to come up with a default e-mail ident, but a bug
lost the "user@" part.
* The attribute mechanism didn't allow limiting attributes to be
applied to only a single directory itself with "path/" like the
exclude mechanism does. The initial implementation of this that
was merged to 'maint' and 1.8.1.2 was with a severe performance
degradations and needs to merge a fix-up topic.
* The smart HTTP clients forgot to verify the content-type that comes
back from the server side to make sure that the request is being
handled properly.
* "git am" did not parse datestamp correctly from Hg generated patch,
when it is run in a locale outside C (or en).
* "git apply" misbehaved when fixing whitespace breakages by removing
excess trailing blank lines.
* "git apply --summary" has been taught to make sure the similarity
value shown in its output is sensible, even when the input had a
bogus value.
* A tar archive created by "git archive" recorded a directory in a
way that made NetBSD's implementation of "tar" sometimes unhappy.
* "git archive" did not record uncompressed size in the header when
streaming a zip archive, which confused some implementations of unzip.
* "git archive" did not parse configuration values in tar.* namespace
correctly.
(merge b3873c3 jk/config-parsing-cleanup later to maint).
* Attempt to "branch --edit-description" an existing branch, while
being on a detached HEAD, errored out.
* "git clean" showed what it was going to do, but sometimes end up
finding that it was not allowed to do so, which resulted in a
confusing output (e.g. after saying that it will remove an
untracked directory, it found an embedded git repository there
which it is not allowed to remove). It now performs the actions
and then reports the outcome more faithfully.
* When "git clone --separate-git-dir=$over_there" is interrupted, it
failed to remove the real location of the $GIT_DIR it created.
This was most visible when interrupting a submodule update.
* "git cvsimport" mishandled timestamps at DST boundary.
* We used to have an arbitrary 32 limit for combined diff input,
resulting in incorrect number of leading colons shown when showing
the "--raw --cc" output.
* "git fetch --depth" was broken in at least three ways. The
resulting history was deeper than specified by one commit, it was
unclear how to wipe the shallowness of the repository with the
command, and documentation was misleading.
(merge cfb70e1 nd/fetch-depth-is-broken later to maint).
* "git log --all -p" that walked refs/notes/textconv/ ref can later
try to use the textconv data incorrectly after it gets freed.
* We forgot to close the file descriptor reading from "gpg" output,
killing "git log --show-signature" on a long history.
* The way "git svn" asked for password using SSH_ASKPASS and
GIT_ASKPASS was not in line with the rest of the system.
* The --graph code fell into infinite loop when asked to do what the
code did not expect.
* http transport was wrong to ask for the username when the
authentication is done by certificate identity.
* "git pack-refs" that ran in parallel to another process that
created new refs had a nasty race.
* Rebasing the history of superproject with change in the submodule
has been broken since v1.7.12.
* After "git add -N" and then writing a tree object out of the
index, the cache-tree data structure got corrupted.
* "git clone" used to allow --bare and --separate-git-dir=$there
options at the same time, which was nonsensical.
* "git rebase --preserve-merges" lost empty merges in recent versions
of Git.
* "git merge --no-edit" computed who were involved in the work done
on the side branch, even though that information is to be discarded
without getting seen in the editor.
* "git merge" started calling prepare-commit-msg hook like "git
commit" does some time ago, but forgot to pay attention to the exit
status of the hook.
* A failure to push due to non-ff while on an unborn branch
dereferenced a NULL pointer when showing an error message.
* When users spell "cc:" in lowercase in the fake "header" in the
trailer part, "git send-email" failed to pick up the addresses from
there. As e-mail headers field names are case insensitive, this
script should follow suit and treat "cc:" and "Cc:" the same way.
* Output from "git status --ignored" showed an unexpected interaction
with "--untracked".
* "gitweb", when sorting by age to show repositories with new
activities first, used to sort repositories with absolutely
nothing in it early, which was not very useful.
* "gitweb"'s code to sanitize control characters before passing it to
"highlight" filter lost known-to-be-safe control characters by
mistake.
* "gitweb" pages served over HTTPS, when configured to show picon or
gravatar, referred to these external resources to be fetched via
HTTP, resulting in mixed contents warning in browsers.
* When a line to be wrapped has a solid run of non space characters
whose length exactly is the wrap width, "git shortlog -w" failed
to add a newline after such a line.
* Command line completion leaked an unnecessary error message while
looking for possible matches with paths in <tree-ish>.
* Command line completion for "tcsh" emitted an unwanted space
after completing a single directory name.
* Command line completion code was inadvertently made incompatible with
older versions of bash by using a newer array notation.
* "git push" was taught to refuse updating the branch that is
currently checked out long time ago, but the user manual was left
stale.
(merge 50995ed wk/man-deny-current-branch-is-default-these-days later to maint).
* Some shells do not behave correctly when IFS is unset; work it
around by explicitly setting it to the default value.
* Some scripted programs written in Python did not get updated when
PYTHON_PATH changed.
(cherry-pick 96a4647fca54031974cd6ad1 later to maint).
* When autoconf is used, any build on a different commit always ran
"config.status --recheck" even when unnecessary.
* A fix was added to the build procedure to work around buggy
versions of ccache broke the auto-generation of dependencies, which
unfortunately is still relevant because some people use ancient
distros.
* The autoconf subsystem passed --mandir down to generated
config.mak.autogen but forgot to do the same for --htmldir.
(merge 55d9bf0 ct/autoconf-htmldir later to maint).
* A change made on v1.8.1.x maintenance track had a nasty regression
to break the build when autoconf is used.
(merge 7f1b697 jn/less-reconfigure later to maint).
* We have been carrying a translated and long-unmaintained copy of an
old version of the tutorial; removed.
* t0050 had tests expecting failures from a bug that was fixed some
time ago.
* t4014, t9502 and t0200 tests had various portability issues that
broke on OpenBSD.
* t9020 and t3600 tests had various portability issues.
* t9200 runs "cvs init" on a directory that already exists, but a
platform can configure this fail for the current user (e.g. you
need to be in the cvsadmin group on NetBSD 6.0).
* t9020 and t9810 had a few non-portable shell script construct.
* Scripts to test bash completion was inherently flaky as it was
affected by whatever random things the user may have on $PATH.
* An element on GIT_CEILING_DIRECTORIES could be a "logical" pathname
that uses a symbolic link to point at somewhere else (e.g. /home/me
that points at /net/host/export/home/me, and the latter directory
is automounted). Earlier when Git saw such a pathname e.g. /home/me
on this environment variable, the "ceiling" mechanism did not take
effect. With this release (the fix has also been merged to the
v1.8.1.x maintenance series), elements on GIT_CEILING_DIRECTORIES
are by default checked for such aliasing coming from symbolic
links. As this needs to actually resolve symbolic links for each
element on the GIT_CEILING_DIRECTORIES, you can disable this
mechanism for some elements by listing them after an empty element
on the GIT_CEILING_DIRECTORIES. e.g. Setting /home/me::/home/him to
GIT_CEILING_DIRECTORIES makes Git resolve symbolic links in
/home/me when checking if the current directory is under /home/me,
but does not do so for /home/him.
(merge 7ec30aa mh/maint-ceil-absolute later to maint).
AnonSec - 2021 | Recode By D7net