git Archives: as promised, docs: git for the confused

2005-12-08

A good rule of thumb is that the commands with one-word names (git-diff, git-commit, git-merge, git-push, git-pull, git-status, git-tag, etc.) are designed for end-user use.  Multi-word names (git-count-objects, git-write-tree, git-cat-file) are generally designed for use from a script.

A good rule of thumb is that the commands with one-word names (git-diff, git-commit, git-merge, git-push, git-pull, git-status, git-tag, etc.) are designed for end-user use.  Multi-word names (git-count-objects, git-write-tree, git-cat-file) are generally designed for use from a script.  This isn't ironclad.  The first command to start using git is git-init-db, and git-show-branch is pure porcelain, while git-mktag is a primitive. And you don't often run git-daemon by hand.  But still, it's a useful guideline.

One thing that's confusing is why git allows you to have one version of a file in the current HEAD, a second version in the index, and possibly a third in the working directory.  Why doesn't the index just contain a copy of the current HEAD until you commit a new one?  The answer is merging, which does all its work in the index.  Neither the object database nor the working directory let you have multiple files with the same name.  The index is really very simple.  It's a series of structures, each describing one file.  There's an object ID (SHA1) of the contents, some file metadata to detect changes (time-stamps, inode number, size, permissions, owner, etc.), and the path name relative to the root of the working directory.  It's always stored sorted by path name, for efficient merging.  At (almost) any time, you can take a snapshot of the index and write it as a tree object.  The only interesting feature is that each entry has a 2-bit stage number. Normally, this is always zero, but each path name is allowed up to three different versions (object IDs) in the index at once.  This is used to represent an incomplete merge, and an unmerged index entry (with more than one version) prevents committing the index to the object database.

The most common object needed by git primitives is a tree.  Since a commit points to a tree and a tag points to a commit, both of these are acceptable "tree-ish" objects and can be used interchangeably.  Likewise, a tag is "commit-ish" and can be used where a commit is required.  As soon as you get to the porcelain, the most commonly used object is a commit.  Also known as a revision, this is a tree plus a history.

While you can always use the full object ID, you can also use a reference. A reference is a file that contains a 40-character hex SHA1 object ID (and a trailing newline).

When you specify the name of a reference, it is searched for in one of the directories: 	.git/			(or $GIT_DIR) 	.git/refs/		(or $GIT_DIR/refs/) 	.git/refs/heads/	(or $GIT_DIR/refs/heads/) 	.git/refs/tags/		(or $GIT_DIR/refs/tags/)  You may use subdirectories by including slashes in the reference name. There is no search order; if searching the above four path prefixes produces more than one match for the reference name (it's ambiguous), then the name is not valid.

(commit^0 gives the commit object itself.  A no-op if you're starting from a commit, but it lets you get the commit object from a tag object.)

although the most primitive git tools don't care, a convention among all the porcelain is that the current head of development is .git/HEAD, a symbolic link to a reference under refs/heads/.  git-init-db creates HEAD pointing to refs/heads/master, and that is traditionally the name used for the "main trunk" of development. Note that initially refs/heads/master doesn't exist - HEAD is a dangling symlink!  This is okay, and will cause the initial commit to have zero parents.

A "head" is mostly synonymous with a "branch", but the terms have different emphasis.  The "head" is particularly the tip of the branch, where future development will be appended.  A "branch" is the entire development history leading to the head.  However, as far as git is concerned, they're both references to commit objects, referred to from refs/heads/.

When you actually do more (with git-commit, or git-merge), then the current HEAD reference is overwritten with the new commit's id, and the old HEAD becomes HEAD^.  Since HEAD is a symlink, it's the file in refs/heads/ that's actually overwritten.  (See the git-update-ref documentation for further details.)

The git-checkout command actually changes the HEAD symlink.  git-checkout enforces the rule that it will only check out a branch under refs/heads. You can use refs/tags as a source for git-diff or any other command that only examines the revision, but if you want to check it out, you have to copy it to refs/heads.

The "undo" command for commits to the object database is git-reset. Like all deletion-type commands, be careful or you'll hurt yourself. Given a commit (using any of the syntaxes mentioned above), this sets the current HEAD to refer to the given commit.  This does NOT alter the HEAD symlink (as git-checkout <branch> will do), but actually changes the reference pointed to by HEAD (e.g. refs/heads/master) to contain a new object ID.  The classic example is to undo an erroneous commit, use "git-reset HEAD^".

git-reset --soft: Only overwrite the reference.  If you can find the 	old object ID, you can put everything back with a second 	git-reset --soft OLD_HEAD. git-reset --mixed: This is the default, which I always think of 	as "--medium".  Overwrite the reference, and (using 	git-read-tree) read the commit into the index.  The 	working directory is unchanged. git-reset --hard: Do everything --mixed does, and also check out 	the index into the working directory.  This really erases 	all traces of the previous version.  (One caveat: this 	will not delete any files in the working directory that 	were added as part of the changes being undone.)

git-reset with no commit specified is "git-reset HEAD", which is much safer because the object reference is not actually changed.  This can be used to undo changes in the index or working directory that you did not intend.

Like being sure what directory you're in when typing "rm -r", think carefully about what branch you're on when typing "git-reset <commit>".

There is an undelete: git-reset stores the previous HEAD commit in OLD_HEAD.  And git-lost-found can find leftover commits until you do a git-prune.

Merging is central to git operations.  Indeed, a big difference between git and other version control systems is that git assumes that a change will be merged more often than it's written, as it's passed around different developers' repositories.  Even "git checkout" is a merge.

As mentioned, the index and the working directory versions of a file could both be different from the HEAD.  Git lets you merge "under" your current working directory edits, as long as the merge doesn't change the files you're editing.

the procedure for the general 3-way merge case: merging branch B into branch A (the current HEAD).

1) Given two commits, find a common ancestor O to server as the origin    of the merge.

2) Add all three input trees (the Origin, A, and B) to the index by    "git-read-tree -m O A B".  The index now contains up to three copies    of every file.

Then, for each file in the index, git-read-tree does the following:

2a) For each file, git-merge-tree tries to collapse the various        versions into one using the "trivial in-index merge".  This just        uses the file blob object names to see if the file contents        are identical, and if two or more of the three trees contain an        identical copy of this file, it is merged.  A missing (deleted)        file matches another missing file.         Note that this is NOT a majority vote.  If A and B agree on the        contents of the file, that's what is used.  (Whether O agrees is        irrelevant in this case.)  But if O and A agree, then the change        made in B is taken as the final value.  Likewise, if O and B agree,        then A is used.

2b) If this is possible, then a check is made to see if the merge would        conflict with any uncommitted work in the index or change the index        out from under a modified working directory file.  If either of        those cases happen, the entire merge is backed out and fails.

If the merge is possible and safe, the versions are collapsed        into one final result version.

2c) If all three versions differ, the trivial in-index merge is        not possible, and the three source versions are left in the        index unmerged.  Again, if there was uncommitted work in the        index or the working directory, the entire merge fails.

3) Use git-merge-index to iterate over the remaining unmerged files, and    apply an intra-file merge.  The intra-file merge is usually done with    git-merge-one-file, which does a standard RCS-style three-way merge    (see "man merge").

4) Check out all the successfully merged files into the working directory.

5) If automatic merging was successful on every file, commit the merged    version immediately and stop.

6) If automatic merging was not complete, then replace the working    directory copies of any remaining unmerged files with a merged copy    with conflict markers (again, just like RCS or CVS) in the working    directory.  All three source versions are available in the index for    diffing against.

7) Manually edit the conflicts and resolve the merge.  As long as an    unmerged, multi-version file exists in the index, committing the    index is forbidden.

8) Commit the final merged version of the conflicting file(s), replacing    the unmerged versions with the single finished version.

Note that if the merge is simple, with no one file edited on both branches, git never has to open a single file.  It reads three tree objects (recursively) and stat(2)s some working directory files to verify that they haven't changed.  Also note that this aborts and backs out rather than overwrite anything not committed.  You can merge "under" uncommitted edits only if those edits are to files not affected by the merge.

A "2-way merge" is basically a 3-way merge with the contents of the index as the "current HEAD", and the original HEAD as the Origin. However, this merge is designed only for simple cases and only supports the "trivial merge" cases.  It does not fall back to an intra-file merge.

This merge is used by git-checkout to switch between two branches, while preserving any changes in the working directory and index.  Like the 3-way case, if a particular file hasn't changed between the two heads, then git will preserve any uncommitted edits. If the file has changed in any way, git doesn't try to perform any sort of intra-file merge, it just fails.

1-way merging  This is not actually used by the git-core porcelain, and so is only useful to someone writing more porcelain, but I'll describe it for completeness.  Plain (non-merging) git-read-tree will overwrite the index entries with those from the tree.  This invalidates the cached stat data, causing git to think all the working directory files are "potentially changed" until you do a git-update-index --refresh.  By specifying a 1-way merge, any index entry whose contents (object ID) matches the incoming tree will have its cached stat data preserved. Thus, git will know if the working directory file is not changed, and will not overwrite if you execute git-checkout-index.  This is purely an efficiency hack.

There are two cases of 3-way or 2-way merging that are special. Recall that the basic merge pattern is     B--------> A+B   /        /  /        / O -----> A  The two special cases arise if one of A or B is a direct ancestor of the other.  In that case, the common ancestor of both A and B is the older of the two commits.  And the merged result is simply the newer of the two, unchanged.

Recalling that we are merging B into A, if B is a direct ancestor of A, then A already includes all of B.  A is "already up to date" and not changed at all by the merge.  The other case you'll hear mentioned, because it happens a lot when pulling, is when A is a direct ancestor of B.  In this case, the result of the merge is a "fast-forward" to B.  Both of these cases are handled very efficiently by the in-index merge done by git-read-tree.

A pack is a file, built all at once, which contains many delta-compressed objects.  With each .pack file, there's an accompanying .idx file that indexes the pack so that individual objects can be retrieved quickly.  You can reduce the disk space used by your repositories by periodically repacking them with git-repack.  Normally, this makes a new incremental pack of everything not already packed.  With the -a flag, this repacks everything for even greater compression (but takes longer).

The git wire protocol basically consists of negotiation over what objects needs to be transferred followed by sending a custom-built pack. The .idx file can be reconstructed from the .pack file, so it's never transferred.

A major source of git's speed is that it tries to avoid accessing files unnecessarily.  In particular, files can be compared based on their object IDs without needing to open and read them.  As part of this, the responsibility for finding file differences (printing diffs) is divided into finding what files have changed, and finding the changes within those files.  This is all explained in the Documentation/diffcore.txt in the git distribution, but the basics is that many of the primitives spit out a line like this: :100755 100755 68838f3fad1d22ab4f14977434e9ce73365fb304 0000000000000000000000000000000000000000 M	git-bisect.sh when asked for a diff.  This is known as a "raw diff".  They can be told to generate a human-readable diff with the "-p" (patch) flag. The git-diff command includes this by default.

* Advice on using git

If you're used to CVS, where branches and merges are "advanced" features that you can go a long time, you need to learn to use branches in git a lot more.

Branch early and often.  Every time you think about developing a feature or fixing a bug, create a branch to do it on.

In fact, avoid doing any development on the master branch. Merges only.

A branch is the git equivalent of a patch, and merging a branch is the equivalent of applying that patch.  A branch gives it a name that you can use to refer to it.  This is particularly useful if you're sharing your changes.

Once you're done with a branch, you can delete it.  This is basically just removing the refs/heads/<branch> file, but "git-branch -d" adds a few extra safety checks.  Assuming you merged the branch in, you can still find all the commits in the history, it's just the name that's been deleted.

Periodically merge all of the branches you're working on into a testing branch to see if everything works.  Blow away and re-create the testing branch whenever you do this.  When you like the result, merge them into the master.

* The .git directory

index - The actual index file.  objects/ - The object database.  Can be overridden by $GIT_OBJECT_DIRECTORY  hooks/ - where the hook scripts are kept.  The standard git template includes   examples, but disabled by being marked non-executable.  info/exclude - Default project-wide list of file patterns to exclude from notice.   To this is added the per-directory list in .gitignore.   See the git-ls-files docs for full details.  refs/ - References to development heads (branches) and tags.  remotes/ - Short names of remote repositories we pull from or push to.   Details are in the "git-fetch" man page.

HEAD - The current default development head. - Created by git-init-db and never deleted - Changed by git-checkout - Used by git-commit and any other command that commits changes. - May be a dangling pointer, in which case git-commit   does an "initial checkin" with no parent.  COMMIT_EDITMSG - Temp used by git-commit to edit a commit message. COMMIT_MSG - Temp used by git-commit to form a commit message,   post-processed from COMMIT_EDITMSG.  FETCH_HEAD - Just-fetched commits, to be merged into the local trunk. - Created by git-fetch. - Used by git-pull as the source of data to merge.

MERGE_HEAD - Keeps track of what heads are currently being merged into HEAD. - Created by git-merge --no-commit with the heads used  - Deleted by git-checkout and git-reset (since you're abandoning   the merge) - Used by git-commit to supply additional parents to the current commit.   (And deleted when done.)  MERGE_MSG - Generated by git-merge --no-commit. - Used by git-commit as the commit message for a merge   (If present, git-commit doesn't prompt.)  MERGE_SAVE - cpio archive of all locally modified files created by   "git-merge" before starting to do anything, if multiple   merge strategies are being attempted.   Used to rewind the tree in case a merge fails.

ORIG_HEAD - Previous HEAD commit prior to a merge or reset operation.  LAST_MERGE - Set by the "resolve" strategy to the most recently merged-in branch.   Basically, a copy of MERGE_HEAD.  Not used by the other merge strategies,   and resolve is no longer the default, so its utility is very limited.  BISECT_LOG - History of a git-bisect operation. - Can be replayed (or, more usefully, a prefix can) by "git-bisect replay" BISECT_NAMES - The list of files to be modified by git-bisect. - Set by "git-bisect start"  TMP_HEAD (used by git-fetch) TMP_ALT (used by git-fetch)

* Git command summary  There are slightly over a hundred git commands.  This section tries to classify them by purpose, so you can know which commands are intended to be used for what.  You can always use the low-level plumbing directly, but that's inconvenient and error-prone.

I include ".sh", ".perl", etc. suffixes to show what the programs are written in, so you can read those scripts written in languages you're familiar with.  These are the names in the git source tree, but the suffix is not included in the /usr/bin copies.

* Detailed list  Here's a repeat, including descriptions.  I don't try to include every detail you can find on the man page, but to explain when you'd want to use a command.

+ Administrative commands git-init-db 	This creates an empty git repository in ./.git (or $GIT_DIR if 	that is non-null) using a system-wide template. 	It won't hurt an existing repository.

git-fsck-objects 	Validate the object database.  Checks that all references 	point somewhere, all the SHA1 hashes are correct, and that 	sort of thing.  	This walks the entire repository, uncompressing and hashing 	every object, so it takes a while.  Note that by default, 	it skips over packs, which can make it seem misleadingly fast.

git-lost-found.sh 	Find (using git-fsck-objects) any unreferenced commits and 	tags in the object database, and place them in a .git/lost-found 	directory.  This can be used to recover from accidentally 	deleting a tag or branch reference that you wanted to keep.  	This is the opposite of git-prune.

git-prune.sh 	Delete all unreachable objects from the object database. 	It deletes useless packs, but does not remove useless 	objects from the middle of partially useful packs.  	Git leaks objects in a number of cases, such as unsuccessful 	merges.  The leak rate is generally a small fraction of 	the rate at which the desired history grows, so it's not 	very alarming

+ Pack maintenance 	The classic git format is to compress and store each object 	separately.  This is still used for all newly created changes. 	However, objects can also be stored en masse in "packs" which 	contain many objects and tan take advantage of delta-compressing. 	Repacking your repositories periodically can save space. 	(Repacking is pretty quick but not quick enough to be 	comfortable doing every commit.)

git-count-objects.sh 	Print the number and total size of unpacked objects in the 	repository, to help you decide when is a good time to repack.

git-repack.sh 	Make a new pack with all the unpacked objects. 	With -a, include already-packed objects in the new pack. 	With -d as well, deletes all the old packs thereby made redundant.

+ Important primitives 	Although these primitives are not used directly very frequently, 	understanding them will help you understand other git commands 	that wrap them.

git-commit-tree 	Create a new commit object from a tree and a list of parent 	commits.  This is the primitive that's the heart of git-commit. 	(It's also used by git-am, git-applypatch, git-merge, etc.)

git-rev-parse 	This is a very widely used command line canonicalizer for git 	scripts.  It converts relative commit references (e.g. master~3) 	to absolute SHA1 hashes, and can also pass through arguments 	not recognizable as references, so the script can interpret them.  	It is important because it defines the <rev> syntax.

+ Useful primitives These primitives are potentially useful directly.

git-ls-files 	List files in the index and/or working directory.  A variety of 	options control which files to list, based on whether they 	are the same in both places or have been modified.  This command 	is the start of most check-in scripts.

+ General script helpers (used only by scripts) 	These are almost exclusively helpers for use in porcelain 	scripts and have little use by themselves from the command line.

git-cat-file 	Extract a file from the object database.  You can ask for 	an object's type or size given only an object ID, but 	to get its contents, you have to specify the type.  This 	is a deliberate safety measure.

git-hash-object 	Very primitive helper to turn an arbitrary file into an object, 	returning just the ID or actually adding it to the database. 	Used by the cvs-to-git and svn-to-git import filters.

git-ls-tree 	List the contents of a tree object.  Will tell you all the files 	in a commit.  Used by the checkout scripts git-checkout and git-reset.

git-symbolic-ref 	This queries or creates symlinks to references such as HEAD. 	Basically equivalent to readlink(1) or ln -s, this also works 	on platforms that don't have symlinks.  See the man page.

+ Code browsing

git-diff.sh 	Show changes between various trees.  Takes up to two tree 	specifications, and shows the difference between the versions. 	Zero arguments: index vs. working directory (git-diff-files) 	One: tree vs. working directory (git-diff-index) 	One, --cached: tree vs. index (git-diff-index) 	Two: tree vs. tree (git-diff-tree)  	This wrapper always produces human-readable patch output. 	The helpers all produce "diff-raw" format unless you supply 	the -p option.  	There are some interesting options.  Unfortunately, the git-diff 	man page is annoyingly sparse, and refers to the helper scripts' 	documentation rather than describing the many useful options 	they all have in common.  Please do read the man pages of the 	helpers to see what's available. 	 	In particular, although git does not explicitly record file 	renames, it has some pretty good heuristics to notice things. 	-M tries to detect renamed files by matching up deleted files 	with similar newly created files.  -C tries to detect copies 	as well.  By default, -C only looks among the modified files for 	the copy source.  For common cases like splitting a file in two, 	this works well.  The --find-copies-harder searches ALL files 	in the tree for the copy source.  This can be slow on large trees!  	See Documentation/diffcore.txt for an explanation of how all 	this works.

git-diff-tree 	Compare two trees.  This is the git equivalent of the two-operand 	form of "cvs diff".  This command is sometimes useful by itself 	to see the changes made by a single commit.  If you give it 	only one commit on the command line, it shows the diff between 	that commit and its first parent.  If the commit specification 	is long and awkward to type, using "git-diff-tree -p <commit>" 	can be easier than "git-diff <commit>^ <commit>".

git-grep.sh 	A very simple wrapper that runs git-ls-files and greps the 	output looking for a file name.  Does nothing fancy except 	saves typing.

git-log.sh 	Wrapper around git-rev-list --pretty.  Shows a history 	of changes made to the repository.  Takes all of git-rev-list's 	options for specifying which revisions to list.

git-shortlog.perl 	This is a filter for the output of "git-log --pretty=short" 	to generate a one-line-per-change "shortlog" as Linus likes.

git-show-branch 	Visually show the merge history of the references given as 	arguments.  Prints one column per reference and one line per 	commit showing whether that commit is an ancestor of each 	reference.

git-whatchanged.sh 	A simple wrapper around git-rev-list and git-diff-tree, 	this shows the change history of a repository.  Specify a 	directory or file on the command line to limit the 	output to changes affecting those files.  This isn't 	the same as "cvs annotate", but it serves a similar purpose 	among git folks.  	You can add the -p option to include patches as well as log 	comments.  You can also add the -M or -C option to follow 	history back through file renames.  	-S is interesting: it's the "pickaxe" option.  Given a string, 	this limits the output to changes that make that string appear 	or disappear.  This is for "digging through history" to see when 	a piece of code was introduced.  The string may (and often does) 	contain embedded newlines.  See Documentation/cvs-migration.txt.

+ Making local changes 	All of these are examples of "porcelain" scripts.  Reading the 	scripts themselves can be informative; they're generally not 	too confusing.

git-add.sh 	A simple wrapper around "git-ls-files | git-update-index --add" 	to add new files to the index.   You may specify directories. 	You need to invoke this for every new file you want git to 	track.

git-bisect.sh 	Utility to do a binary search to find the change that broke something. 	The heart of this is in "git-rev-list --bisect" 	A very handy little utility!  Kernel developers love it 	when you tell them exactly which patch broke something.

There are three steps: 	git-bisect start [<files>] 		- Reset to start bisecting.  If any files are specified, 		  only they will be checked out as bisection proceeds. 	git-bisect good [<revision>] 		- Record the revision as "good".  The change being sought 		  must be after this revision. 	git-bisect bad [<revision>] 		- Record the revision as "bad".  The change being sought 		  must be before or equal to this revision. 	As soon as you have specified one good version and one bad version, 	git-bisect will find a halfway point and check out that 	revision.  Build and test it, then report it as good or bad, 	and git-bisect will narrow the search.  Finally, git-bisect 	will tell you exactly which change caused the problem. 	git-bisect log 		- Show a history of revisions. 	git-bisect replay 		- Replay (part of) a git-bisect log.  Generally used 		  to recover from a mistake, you can truncate the log 		  before the mistake and replay it to continue. 	If git-bisect chooses a version that cannot build, or you 	are otherwise unable to determine whether it is good or bad, 	you can change revisions with "git-reset --hard <revision>" 	to another checkout between the current good and bad limits, and 	continue from there.  "git-reset --hard <revision>" is generally 	dangerous, but you are on a scratch branch.  	This can, of course, be used to look for any change, even 	one for the better, if you can avoid being confused by the 	terms "good" and "bad".

git-branch.sh 	Most commonly used bare, to show the available branches. 	Show, create, or delete a branch.  The current branches 	are simply the contents of .git/refs/heads/. 	Note that this does NOT switch to the created branch! 	For the common case of creating a branch and immediately 	switching to it, "git-checkout -b <branch>" is simpler.

git-checkout.sh 	This does two superficially similar but very different things 	depending on whether any files or paths are specified on the 	command line.  	git-checkout [-f] [-b <new-branch>] <branch> 		This switches (changes the HEAD symlink to) the specified 		branch, updating the index and working directory to 		reflect the change.  This preserves changes in the 		working directory unless -f is specified. 		If -b is specified, a new branch is started from the 		specified point and switched to.  If <branch> is omitted, 		it defaults to HEAD.  This is the usual way to start a 		new branch.  	git-checkout [<branch>] [--] <paths>... 		This replaces the files specified by the given paths with 		the versions from the index or the specified branch. 		It does NOT affect the HEAD symlink, just replaces the 		specified paths.  This form is like a selective form of 		"git-reset".  Normally, this can guess whether the first 		argument is a branch name or a path, but you can use 		"--" to force the latter interpretation.  	With no branch, this is used to revert a botched edit of a 	particular file.  	Both forms use git-read-tree internally, but the net effect is 	quite different.

git-commit.sh 	Commit changes to the revision history.  In terms of primitives, 	this does three things: 	1) Updates the index file with the working directory files 	   specified on the command line, or -a for all 	   (using git-diff-files --name-only | git-update_index), 	2) Prompts for or generates a commit message, and then 	3) Creates a commit object with the current index contents.  	This also executes the pre-commit, commit-msg, and 	post-commit hooks if present.  	This will remove deleted files from the index, but will not 	add new files to the index, even if explicitly specified on the 	command line; you must use git-add for that.

git-reset.sh 	Explained in detail in "resetting", above.  This modifies the 	current branch head reference (as pointed to by .git/HEAD) 	to refer to the given commit.  It does not modify .git/HEAD  	Reset the current HEAD to the specified commit, so that future 	checkins will be relative to it.  There are three 	variations: 	--soft: Just move the HEAD link.  The index is unchanged. 	--mixed (default): Move the HEAD link and update the index file. 		Any local changes will appear not checked in. 	--hard: Move the HEAD links, update the index file, and 		check out the index, overwriting the working 		directory.  Like "cvs update -C". 	In case of accidents, this copies the previous head 	object ID to ORIG_HEAD (which is NOT a symlink).

git-status.sh 	Show all files in the directory not current with respect 	to the git HEAD.  The basic categories are: 	1) Changed in the index, will be included in the next commit. 	2) Changed in the working directory but NOT in the index; will 	   be committed only if added via git-update-index or the 	   git-commit command line. 	3) Not tracked by git.

+ Cherry-picking 	Cherry-picking is the process of taking part of the changes 	introduced on one tree and applying those changes to another. 	This doesn't produce a parent/descendant relationship in the 	commit history.  	To produce that relationship, there's a special type of merge you 	can do if you've taken everything you want off a branch and 	want to show it in the merge history without actually importing 	any changes from it: ours.  "git-merge -s ours" will generate a 	commit that shows some branches were merged in, but not 	actually alter the current HEAD source code in any way.  	One thing cherry-picking is sometimes used for is taking a 	development branch and re-organizing the changes into a patch 	series for submission to the Linux kernel.

git-cherry.sh 	This searches a branch for patches which have not been applied 	to another.  Basically, it finds the unpicked cherries. 	It searches back to the common ancestor of the named branch and 	the current head using git-patch-id to identify similarity 	in patches.

git-cherry-pick.sh 	Given a commit (on a different branch), compute a diff between 	it and its immediate parent, and apply it to the current HEAD.   	This is actually the same script as "git revert", but 	works forward.  git-cherry finds the patches, this merges them. 	Handles failures gracefully.

git-rebase.sh 	Move a branch to a more recent "base" release.	This just extracts 	all the patches applied on the local head since the last merge 	with upstream (using git-format-patch) and re-applies them 	relative to the current upstream with git-am (explained under 	"accepting changes by e-mail").  Finally, it deletes the old 	branch and gives its name to the new one, so your branch now 	contains all the same changes, but relative to a different base. 	Basically the same as cherry-picking an entire branch.

git-revert.sh 	Undo a commit.	Basically "patch -R" followed by a commit. 	This is actually the same script as "git-cherry-pick", just 	applies the patch in reverse, undoing a change that you don't 	wish to back up to using git-reset.  Handles failures gracefully 	by telling the user what to do.

+ Accepting changes by e-mail

git-apply 	Apply a (git-style extended) patch to the current index 	and working directory.

git-am.sh 	The new and improved "apply an mbox" script.  Takes an 	mbox-style concatenation of e-mails as input and batch-applies 	them, generating one commit per message.  Can resume after 	stopping on a patch problem. 	(Invoke it as "git-am --skip" or "git-am --resolved" to 	deal with the problematic patch and continue.)

+ Publishing changes by e-mail

git-format-patch.sh 	Generate a series of patches, in the preferred Linux kernel 	(Documentation/SubmittingPatches) format, for posting to lkml 	or the like.  This formats every commit on a branch as a separate 	patch.

git-send-email.perl Actually e-mail the output of git-format-patch.

+ Merging

git-merge.sh 	Merge one or more "remote" heads into the current head. 	Some changes, when there has been change only on one branch or 	the same change has been made to all branches, can be resolved 	by the "trivial in-index" merge done by git-read-tree.	For more 	complex cases, git provides a number of different merge strategies 	(with reasonable defaults).  	Note that merges are done on a filename basis.	While git tries 	to detect renames when generating diffs, most merge strategies 	don't track them by renaming.  (The "recursive" strategy, which 	recently became the default, is a notable exception.)

git-merge-ours.sh 	A "dummy" merge strategy helper.  Claims that we did the merge, but 	actually takes the current tree unmodified.  This is used to 	cleanly terminate side branches that heve been cherry-picked in.

+ Making releases

git-tag.sh 	Create a tag in the refs/tags directory.  There are two kinds: 	"lightweight tags" are just references to commits.  More 	serious tags are GPG-signed tag objects, and people receiving 	the git tree can verify that it is the version that you released.

+ Accepting changes by network 	Pulling consists of two steps: retrieving the remote commit 	objects and everything they point to (including ancestors), 	then merging that into the desired tree.  There are still 	separate fetch and merge commands, but it's more commonly done 	with a single "git-pull" command.  git-fetch leaves the commit 	objects, one per line, in .git/FETCH_HEAD.  git-merge will 	merge those in if that file exists when it is run.  	References to remote repositories can be made with long URLs, 	or with files in the .git/remotes/ directory.  The latter 	also specifies the local branches to merge the fetched data into, 	making it very easy to track a remote repository.

git-clone.sh 	Create a new local clone of a remote repository. 	(Can do a couple of space-sharing hacks when "remote" is on 	a local machine.)  	You only do this once

git-fetch.sh 	Fetch the named refs and all linked objects from a remote repository. 	The resultant refs (tags and commits) are stored in .git/FETCH_HEAD, 	which is used by a later git-resolve or git octopus.  	This is the first half of a "git pull" operation.

git-ls-remote.sh 	Show the contents of the refs/heads/ and/or refs/tags/ directories 	of a remote repository.  Useful to see what's available.

git-pull.sh 	Fetches specific commits from the given remote repository, 	and merges everything into the current branch.	If a remote 	commit is named as src:dst, this merges the remote head "src" 	into the branch "dst" as well as the trunk.  Typically, the "dst" 	branch is not modified locally, but is kept as a pristine copy 	of the remote branch.  	One very standard example of this contention is that 	a repository that is tracking another specifies "master:origin" 	to provide a pristine local copy of the remote "master" 	branch in the local branch named "origin".

git-shell 	A shell that can be used for git-only users.  Allows git 	push (git-receive-pack) and pull (git-upload-pack) only.

+ Publishing changes by network

git-daemon 	A daemon that serves up the git native protocol so anonymous 	clients can fetch data.  For it to allow export of a directory, 	the magic file name "git-daemon-export-ok" must exist in it.  	This does not accept (receive) data under any circumstances.

git-push.sh 	Git-pull, only backwards.  Send local changes to a remote 	repository.  The same .git/remotes/ short-cuts can be used, 	and the same src:dst syntax.  (But this time, the src is local 	and the dst is remote.)

git-request-pull.sh 	Generate an e-mail summarizing the changes between two commits, 	and request that the recipient pull them from your repository. 	Just a little helper to generate a consistent and informative 	format.

git-update-server-info 	To run git over http, auxiliary info files are required that 	describes what objects are in the repository (since git-upload-pack 	can't generate this on the fly).  If you want to publish a 	repository via http, run this after every commit.  (Typically 	via the hooks/post-update script.)

git Archives: as promised, docs: git for the confused - The Diigo Meta page

Would you like to comment?

Top Tags

Check out another URL