1.5.1

GIT v1.5.1 Release Notes ========================

Updates since v1.5.0 --------------------

• Deprecated commands and options.

• git-diff-stages and git-resolve have been removed.

• New commands and options.

• "git log" and friends take --reverse, which instructs them to give their output in the order opposite from their usual. They typically output from new to old, but with this option their output would read from old to new. "git shortlog" usually lists older commits first, but with this option, they are shown from new to old.

• "git log --pretty=format:<string>" to allow more flexible custom log output.

• "git diff" learned --ignore-space-at-eol. This is a weaker form of --ignore-space-change.

• "git diff --no-index pathA pathB" can be used as diff replacement with git specific enhancements.

• "git diff --no-index" can read from ’-’ (standard input).

• "git diff" also learned --exit-code to exit with non-zero status when it found differences. In the future we might want to make this the default but that would be a rather big backward incompatible change; it will stay as an option for now.

• "git diff --quiet" is --exit-code with output turned off, meant for scripted use to quickly determine if there is any tree-level difference.

• Textual patch generation with "git diff" without -w/-b option has been significantly optimized. "git blame" got faster because of the same change.

• "git log" and "git rev-list" has been optimized significantly when they are used with pathspecs.

• "git branch --track" can be used to set up configuration variables to help it easier to base your work on branches you track from a remote site.

• "git format-patch --attach" now emits attachments. Use --inline to get an inlined multipart/mixed.

• "git name-rev" learned --refs=<pattern>, to limit the tags used for naming the given revisions only to the ones matching the given pattern.

• "git remote update" is to run "git fetch" for defined remotes to update tracking branches.

• "git cvsimport" can now take ’-d’ to talk with a CVS repository different from what are recorded in CVS/Root (overriding it with environment CVSROOT does not work).

• "git bundle" can help sneaker-netting your changes between repositories.

• "git mergetool" can help 3-way file-level conflict resolution with your favorite graphical merge tools.

• A new configuration "core.symlinks" can be used to disable symlinks on filesystems that do not support them; they are checked out as regular files instead.

• You can name a commit object with its first line of the message. The syntax to use is ’:/message text’. E.g.

$ git show ":/object name: introduce ':/<oneline prefix>' notation"

means the same thing as:

$ git show 28a4d940443806412effa246ecc7768a21553ec7

• "git bisect" learned a new command "run" that takes a script to run after each revision is checked out to determine if it is good or bad, to automate the bisection process.

• "git log" family learned a new traversal option --first-parent, which does what the name suggests.

• Updated behavior of existing commands.

• "git-merge-recursive" used to barf when there are more than one common ancestors for the merge, and merging them had a rename/rename conflict. This has been fixed.

• "git fsck" does not barf on corrupt loose objects.

• "git rm" does not remove newly added files without -f.

• "git archimport" allows remapping when coming up with git branch names from arch names.

• git-svn got almost a rewrite.

• core.autocrlf configuration, when set to ’true’, makes git to convert CRLF at the end of lines in text files to LF when reading from the filesystem, and convert in reverse when writing to the filesystem. The variable can be set to ’input’, in which case the conversion happens only while reading from the filesystem but files are written out with LF at the end of lines. Currently, which paths to consider ’text’ (i.e. be subjected to the autocrlf mechanism) is decided purely based on the contents, but the plan is to allow users to explicitly override this heuristic based on paths.

• The behavior of ’git-apply’, when run in a subdirectory, without --index nor --cached were inconsistent with that of the command with these options. This was fixed to match the behavior with --index. A patch that is meant to be applied with -p1 from the toplevel of the project tree can be applied with any custom -p<n> option. A patch that is not relative to the toplevel needs to be applied with -p<n> option with or without --index (or --cached).

• "git diff" outputs a trailing HT when pathnames have embedded SP on +++/--- header lines, in order to help "GNU patch" to parse its output. "git apply" was already updated to accept this modified output format since ce74618d (Sep 22, 2006).

• "git cvsserver" runs hooks/update and honors its exit status.

• "git cvsserver" can be told to send everything with -kb.

• "git diff --check" also honors the --color output option.

• "git name-rev" used to stress the fact that a ref is a tag too much, by saying something like "v1.2.3^022". It now says "v1.2.322" in such a case (it still says "v1.2.3^0" if it does not talk about an ancestor of the commit that is tagged, which makes sense).

• "git rev-list --boundary" now shows boundary markers for the commits omitted by --max-age and --max-count condition.

• The configuration mechanism now reads $(prefix)/etc/gitconfig.

• "git apply --verbose" shows what preimage lines were wanted when it couldn’t find them.

• "git status" in a read-only repository got a bit saner.

• "git fetch" (hence "git clone" and "git pull") are less noisy when the output does not go to tty.

• "git fetch" between repositories with many refs were slow even when there are not many changes that needed transferring. This has been sped up by partially rewriting the heaviest parts in C.

• "git mailinfo" which splits an e-mail into a patch and the meta-information was rewritten, thanks to Don Zickus. It handles nested multipart better. The command was broken for a brief period on ’master’ branch since 1.5.0 but the breakage is fixed now.

• send-email learned configurable bcc and chain-reply-to.

• "git remote show $remote" also talks about branches that would be pushed if you run "git push remote".

• Using objects from packs is now seriously optimized by clever use of a cache. This should be most noticeable in git-log family of commands that involve reading many tree objects. In addition, traversing revisions while filtering changes with pathspecs is made faster by terminating the comparison between the trees as early as possible.

• Hooks

• The part to send out notification e-mails was removed from the sample update hook, as it was not an appropriate place to do so. The proper place to do this is the new post-receive hook. An example hook has been added to contrib/hooks/.

• Others

• git-revert, git-gc and git-cherry-pick are now built-ins.

Fixes since v1.5.0 ------------------

These are all in v1.5.0.x series.

• Documentation updates

• Clarifications and corrections to 1.5.0 release notes.

• The main documentation did not link to git-remote documentation.

• Clarified introductory text of git-rebase documentation.

• Converted remaining mentions of update-index on Porcelain documents to git-add/git-rm.

• Some i18n.* configuration variables were incorrectly described as core.*; fixed.

• added and clarified core.bare, core.legacyheaders configurations.

• updated "git-clone --depth" documentation.

• user-manual updates.

• Options to ’git remote add’ were described insufficiently.

• Configuration format.suffix was not documented.

• Other formatting and spelling fixes.

• user-manual has better cross references.

• gitweb installation/deployment procedure is now documented.

• Bugfixes

• git-upload-pack closes unused pipe ends; earlier this caused many zombies to hang around.

• git-rerere was recording the contents of earlier hunks duplicated in later hunks. This prevented resolving the same conflict when performing the same merge the other way around.

• git-add and git-update-index on a filesystem on which executable bits are unreliable incorrectly reused st_mode bits even when the path changed between symlink and regular file.

• git-daemon marks the listening sockets with FD_CLOEXEC so that it won’t be leaked into the children.

• segfault from git-blame when the mandatory pathname parameter was missing was fixed; usage() message is given instead.

• git-rev-list did not read $GIT_DIR/config file, which means that did not honor i18n.logoutputencoding correctly.

• Automated merge conflict handling when changes to symbolic links conflicted were completely broken. The merge-resolve strategy created a regular file with conflict markers in it in place of the symbolic link. The default strategy, merge-recursive was even more broken. It removed the path that was pointed at by the symbolic link. Both of these problems have been fixed.

• ’git diff maint master next’ did not correctly give combined diff across three trees.

• ’git fast-import’ portability fix for Solaris.

• ’git show-ref --verify’ without arguments did not error out but segfaulted.

• ’git diff :tracked-file pwd/an-untracked-file’ gave an extra slashes after a/ and b/.

• ’git format-patch’ produced too long filenames if the commit message had too long line at the beginning.

• Running ’make all’ and then without changing anything running ’make install’ still rebuilt some files. This was inconvenient when building as yourself and then installing as root (especially problematic when the source directory is on NFS and root is mapped to nobody).

• ’git-rerere’ failed to deal with two unconflicted paths that sorted next to each other.

• ’git-rerere’ attempted to open(2) a symlink and failed if there was a conflict. Since a conflicting change to a symlink would not benefit from rerere anyway, the command now ignores conflicting changes to symlinks.

• ’git-repack’ did not like to pass more than 64 arguments internally to underlying ’rev-list’ logic, which made it impossible to repack after accumulating many (small) packs in the repository.

• ’git-diff’ to review the combined diff during a conflicted merge were not reading the working tree version correctly when changes to a symbolic link conflicted. It should have read the data using readlink(2) but read from the regular file the symbolic link pointed at.

• ’git-remote’ did not like period in a remote’s name.

• ’git.el’ honors the commit coding system from the configuration.

• ’blameview’ in contrib/ correctly digs deeper when a line is clicked.

• ’http-push’ correctly makes sure the remote side has leading path. Earlier it started in the middle of the path, and incorrectly.

• ’git-merge’ did not exit with non-zero status when the working tree was dirty and cannot fast forward. It does now.

• ’cvsexportcommit’ does not lose yet-to-be-used message file.

• int-vs-size_t typefix when running combined diff on files over 2GB long.

• ’git apply --whitespace=strip’ should not touch unmodified lines.

• ’git-mailinfo’ choke when a logical header line was too long.

• ’git show A..B’ did not error out. Negative ref ("not A" in this example) does not make sense for the purpose of the command, so now it errors out.

• ’git fmt-merge-msg --file’ without file parameter did not correctly error out.

• ’git archimport’ barfed upon encountering a commit without summary.

• ’git index-pack’ did not protect itself from getting a short read out of pread(2).

• ’git http-push’ had a few buffer overruns.

• Build dependency fixes to rebuild fetch.o when other headers change.

• git.el does not add duplicate sign-off lines.

• git-commit shows the full stat of the resulting commit, not just about the files in the current directory, when run from a subdirectory.

• "git-checkout -m ’@{8 hours ago}’" had a funny failure from eval; fixed.

• git-merge (hence git-pull) did not refuse fast-forwarding when the working tree had local changes that would have conflicted with it.

• a handful small fixes to gitweb.

• build procedure for user-manual is fixed not to require locally installed stylesheets.

• "git commit $paths" on paths whose earlier contents were already updated in the index were failing out.

• Tweaks

• sliding mmap() inefficiently mmaped the same region of a packfile with an access pattern that used objects in the reverse order. This has been made more efficient.