Merge branch 'sv/objfixes'
[git/git.git] / Documentation / user-manual.txt
<
CommitLineData
71f4b183
CW
1Git User's Manual (for version 1.5.1 or newer)
2______________________________________________
d19fbc3c 3
99eaefdd
BF
4
5Git is a fast distributed revision control system.
6
d19fbc3c 7This manual is designed to be readable by someone with basic unix
79c96c57 8command-line skills, but no previous knowledge of git.
d19fbc3c 9
2624d9a5
BF
10<<repositories-and-branches>> and <<exploring-git-history>> explain how
11to fetch and study a project using git--read these chapters to learn how
12to build and test a particular version of a software project, search for
13regressions, and so on.
ef89f701 14
2624d9a5
BF
15People needing to do actual development will also want to read
16<<Developing-with-git>> and <<sharing-development>>.
6bd9b682
BF
17
18Further chapters cover more specialized topics.
19
d19fbc3c
BF
20Comprehensive reference documentation is available through the man
21pages. For a command such as "git clone", just use
22
23------------------------------------------------
24$ man git-clone
25------------------------------------------------
26
2624d9a5
BF
27See also <<git-quick-start>> for a brief overview of git commands,
28without any explanation.
b181d57f 29
2624d9a5
BF
30Also, see <<todo>> for ways that you can help make this manual more
31complete.
b181d57f 32
b181d57f 33
e34caace 34[[repositories-and-branches]]
d19fbc3c
BF
35Repositories and Branches
36=========================
37
e34caace 38[[how-to-get-a-git-repository]]
d19fbc3c
BF
39How to get a git repository
40---------------------------
41
42It will be useful to have a git repository to experiment with as you
43read this manual.
44
45The best way to get one is by using the gitlink:git-clone[1] command
46to download a copy of an existing repository for a project that you
47are interested in. If you don't already have a project in mind, here
48are some interesting examples:
49
50------------------------------------------------
51 # git itself (approx. 10MB download):
52$ git clone git://git.kernel.org/pub/scm/git/git.git
53 # the linux kernel (approx. 150MB download):
54$ git clone git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6.git
55------------------------------------------------
56
57The initial clone may be time-consuming for a large project, but you
58will only need to clone once.
59
60The clone command creates a new directory named after the project
61("git" or "linux-2.6" in the examples above). After you cd into this
62directory, you will see that it contains a copy of the project files,
63together with a special top-level directory named ".git", which
64contains all the information about the history of the project.
65
d5cd5de4 66In most of the following, examples will be taken from one of the two
d19fbc3c
BF
67repositories above.
68
e34caace 69[[how-to-check-out]]
d19fbc3c
BF
70How to check out a different version of a project
71-------------------------------------------------
72
73Git is best thought of as a tool for storing the history of a
74collection of files. It stores the history as a compressed
75collection of interrelated snapshots (versions) of the project's
76contents.
77
81b6c950
BF
78A single git repository may contain multiple branches. It keeps track
79of them by keeping a list of <<def_head,heads>> which reference the
80latest version on each branch; the gitlink:git-branch[1] command shows
81you the list of branch heads:
d19fbc3c
BF
82
83------------------------------------------------
84$ git branch
85* master
86------------------------------------------------
87
4f752407
BF
88A freshly cloned repository contains a single branch head, by default
89named "master", with the working directory initialized to the state of
90the project referred to by that branch head.
d19fbc3c 91
81b6c950
BF
92Most projects also use <<def_tag,tags>>. Tags, like heads, are
93references into the project's history, and can be listed using the
d19fbc3c
BF
94gitlink:git-tag[1] command:
95
96------------------------------------------------
97$ git tag -l
98v2.6.11
99v2.6.11-tree
100v2.6.12
101v2.6.12-rc2
102v2.6.12-rc3
103v2.6.12-rc4
104v2.6.12-rc5
105v2.6.12-rc6
106v2.6.13
107...
108------------------------------------------------
109
fe4b3e59 110Tags are expected to always point at the same version of a project,
81b6c950 111while heads are expected to advance as development progresses.
fe4b3e59 112
81b6c950 113Create a new branch head pointing to one of these versions and check it
d19fbc3c
BF
114out using gitlink:git-checkout[1]:
115
116------------------------------------------------
117$ git checkout -b new v2.6.13
118------------------------------------------------
119
120The working directory then reflects the contents that the project had
121when it was tagged v2.6.13, and gitlink:git-branch[1] shows two
122branches, with an asterisk marking the currently checked-out branch:
123
124------------------------------------------------
125$ git branch
126 master
127* new
128------------------------------------------------
129
130If you decide that you'd rather see version 2.6.17, you can modify
131the current branch to point at v2.6.17 instead, with
132
133------------------------------------------------
134$ git reset --hard v2.6.17
135------------------------------------------------
136
81b6c950 137Note that if the current branch head was your only reference to a
d19fbc3c 138particular point in history, then resetting that branch may leave you
81b6c950
BF
139with no way to find the history it used to point to; so use this command
140carefully.
d19fbc3c 141
e34caace 142[[understanding-commits]]
d19fbc3c
BF
143Understanding History: Commits
144------------------------------
145
146Every change in the history of a project is represented by a commit.
147The gitlink:git-show[1] command shows the most recent commit on the
148current branch:
149
150------------------------------------------------
151$ git show
152commit 2b5f6dcce5bf94b9b119e9ed8d537098ec61c3d2
153Author: Jamal Hadi Salim <hadi@cyberus.ca>
154Date: Sat Dec 2 22:22:25 2006 -0800
155
156 [XFRM]: Fix aevent structuring to be more complete.
157
158 aevents can not uniquely identify an SA. We break the ABI with this
159 patch, but consensus is that since it is not yet utilized by any
160 (known) application then it is fine (better do it now than later).
161
162 Signed-off-by: Jamal Hadi Salim <hadi@cyberus.ca>
163 Signed-off-by: David S. Miller <davem@davemloft.net>
164
165diff --git a/Documentation/networking/xfrm_sync.txt b/Documentation/networking/xfrm_sync.txt
166index 8be626f..d7aac9d 100644
167--- a/Documentation/networking/xfrm_sync.txt
168+++ b/Documentation/networking/xfrm_sync.txt
169@@ -47,10 +47,13 @@ aevent_id structure looks like:
170
171 struct xfrm_aevent_id {
172 struct xfrm_usersa_id sa_id;
173+ xfrm_address_t saddr;
174 __u32 flags;
175+ __u32 reqid;
176 };
177...
178------------------------------------------------
179
180As you can see, a commit shows who made the latest change, what they
181did, and why.
182
35121930
BF
183Every commit has a 40-hexdigit id, sometimes called the "object name" or the
184"SHA1 id", shown on the first line of the "git show" output. You can usually
185refer to a commit by a shorter name, such as a tag or a branch name, but this
186longer name can also be useful. Most importantly, it is a globally unique
187name for this commit: so if you tell somebody else the object name (for
188example in email), then you are guaranteed that name will refer to the same
189commit in their repository that it does in yours (assuming their repository
190has that commit at all). Since the object name is computed as a hash over the
191contents of the commit, you are guaranteed that the commit can never change
192without its name also changing.
193
194In fact, in <<git-internals>> we shall see that everything stored in git
195history, including file data and directory contents, is stored in an object
196with a name that is a hash of its contents.
d19fbc3c 197
e34caace 198[[understanding-reachability]]
d19fbc3c
BF
199Understanding history: commits, parents, and reachability
200~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
201
202Every commit (except the very first commit in a project) also has a
203parent commit which shows what happened before this commit.
204Following the chain of parents will eventually take you back to the
205beginning of the project.
206
207However, the commits do not form a simple list; git allows lines of
208development to diverge and then reconverge, and the point where two
209lines of development reconverge is called a "merge". The commit
210representing a merge can therefore have more than one parent, with
211each parent representing the most recent commit on one of the lines
212of development leading to that point.
213
214The best way to see how this works is using the gitlink:gitk[1]
215command; running gitk now on a git repository and looking for merge
216commits will help understand how the git organizes history.
217
218In the following, we say that commit X is "reachable" from commit Y
219if commit X is an ancestor of commit Y. Equivalently, you could say
220that Y is a descendent of X, or that there is a chain of parents
221leading from commit Y to commit X.
222
e34caace 223[[history-diagrams]]
3dff5379
PR
224Understanding history: History diagrams
225~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
d19fbc3c
BF
226
227We will sometimes represent git history using diagrams like the one
228below. Commits are shown as "o", and the links between them with
229lines drawn with - / and \. Time goes left to right:
230
1dc71a91
BF
231
232................................................
d19fbc3c
BF
233 o--o--o <-- Branch A
234 /
235 o--o--o <-- master
236 \
237 o--o--o <-- Branch B
1dc71a91 238................................................
d19fbc3c
BF
239
240If we need to talk about a particular commit, the character "o" may
241be replaced with another letter or number.
242
e34caace 243[[what-is-a-branch]]
d19fbc3c
BF
244Understanding history: What is a branch?
245~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
246
81b6c950
BF
247When we need to be precise, we will use the word "branch" to mean a line
248of development, and "branch head" (or just "head") to mean a reference
249to the most recent commit on a branch. In the example above, the branch
250head named "A" is a pointer to one particular commit, but we refer to
251the line of three commits leading up to that point as all being part of
d19fbc3c
BF
252"branch A".
253
81b6c950
BF
254However, when no confusion will result, we often just use the term
255"branch" both for branches and for branch heads.
d19fbc3c 256
e34caace 257[[manipulating-branches]]
d19fbc3c
BF
258Manipulating branches
259---------------------
260
261Creating, deleting, and modifying branches is quick and easy; here's
262a summary of the commands:
263
264git branch::
265 list all branches
266git branch <branch>::
267 create a new branch named <branch>, referencing the same
268 point in history as the current branch
269git branch <branch> <start-point>::
270 create a new branch named <branch>, referencing
271 <start-point>, which may be specified any way you like,
272 including using a branch name or a tag name
273git branch -d <branch>::
274 delete the branch <branch>; if the branch you are deleting
c64415e2
BF
275 points to a commit which is not reachable from the current
276 branch, this command will fail with a warning.
d19fbc3c
BF
277git branch -D <branch>::
278 even if the branch points to a commit not reachable
279 from the current branch, you may know that that commit
280 is still reachable from some other branch or tag. In that
281 case it is safe to use this command to force git to delete
282 the branch.
283git checkout <branch>::
284 make the current branch <branch>, updating the working
285 directory to reflect the version referenced by <branch>
286git checkout -b <new> <start-point>::
287 create a new branch <new> referencing <start-point>, and
288 check it out.
289
72a76c95
BF
290The special symbol "HEAD" can always be used to refer to the current
291branch. In fact, git uses a file named "HEAD" in the .git directory to
292remember which branch is current:
293
294------------------------------------------------
295$ cat .git/HEAD
296ref: refs/heads/master
297------------------------------------------------
298
25d9f3fa 299[[detached-head]]
72a76c95
BF
300Examining an old version without creating a new branch
301------------------------------------------------------
302
303The git-checkout command normally expects a branch head, but will also
304accept an arbitrary commit; for example, you can check out the commit
305referenced by a tag:
306
307------------------------------------------------
308$ git checkout v2.6.17
309Note: moving to "v2.6.17" which isn't a local branch
310If you want to create a new branch from this checkout, you may do so
311(now or later) by using -b with the checkout command again. Example:
312 git checkout -b <new_branch_name>
313HEAD is now at 427abfa... Linux v2.6.17
314------------------------------------------------
315
316The HEAD then refers to the SHA1 of the commit instead of to a branch,
317and git branch shows that you are no longer on a branch:
318
319------------------------------------------------
320$ cat .git/HEAD
321427abfa28afedffadfca9dd8b067eb6d36bac53f
953f3d6f 322$ git branch
72a76c95
BF
323* (no branch)
324 master
325------------------------------------------------
326
327In this case we say that the HEAD is "detached".
328
953f3d6f
BF
329This is an easy way to check out a particular version without having to
330make up a name for the new branch. You can still create a new branch
331(or tag) for this version later if you decide to.
d19fbc3c 332
e34caace 333[[examining-remote-branches]]
d19fbc3c
BF
334Examining branches from a remote repository
335-------------------------------------------
336
337The "master" branch that was created at the time you cloned is a copy
338of the HEAD in the repository that you cloned from. That repository
339may also have had other branches, though, and your local repository
340keeps branches which track each of those remote branches, which you
341can view using the "-r" option to gitlink:git-branch[1]:
342
343------------------------------------------------
344$ git branch -r
345 origin/HEAD
346 origin/html
347 origin/maint
348 origin/man
349 origin/master
350 origin/next
351 origin/pu
352 origin/todo
353------------------------------------------------
354
355You cannot check out these remote-tracking branches, but you can
356examine them on a branch of your own, just as you would a tag:
357
358------------------------------------------------
359$ git checkout -b my-todo-copy origin/todo
360------------------------------------------------
361
362Note that the name "origin" is just the name that git uses by default
363to refer to the repository that you cloned from.
364
365[[how-git-stores-references]]
f60b9642
BF
366Naming branches, tags, and other references
367-------------------------------------------
d19fbc3c
BF
368
369Branches, remote-tracking branches, and tags are all references to
f60b9642
BF
370commits. All references are named with a slash-separated path name
371starting with "refs"; the names we've been using so far are actually
372shorthand:
d19fbc3c 373
f60b9642
BF
374 - The branch "test" is short for "refs/heads/test".
375 - The tag "v2.6.18" is short for "refs/tags/v2.6.18".
376 - "origin/master" is short for "refs/remotes/origin/master".
d19fbc3c 377
f60b9642
BF
378The full name is occasionally useful if, for example, there ever
379exists a tag and a branch with the same name.
d19fbc3c 380
c64415e2
BF
381As another useful shortcut, the "HEAD" of a repository can be referred
382to just using the name of that repository. So, for example, "origin"
383is usually a shortcut for the HEAD branch in the repository "origin".
d19fbc3c
BF
384
385For the complete list of paths which git checks for references, and
f60b9642
BF
386the order it uses to decide which to choose when there are multiple
387references with the same shorthand name, see the "SPECIFYING
388REVISIONS" section of gitlink:git-rev-parse[1].
d19fbc3c
BF
389
390[[Updating-a-repository-with-git-fetch]]
391Updating a repository with git fetch
392------------------------------------
393
394Eventually the developer cloned from will do additional work in her
395repository, creating new commits and advancing the branches to point
396at the new commits.
397
398The command "git fetch", with no arguments, will update all of the
399remote-tracking branches to the latest version found in her
400repository. It will not touch any of your own branches--not even the
401"master" branch that was created for you on clone.
402
e34caace 403[[fetching-branches]]
d5cd5de4
BF
404Fetching branches from other repositories
405-----------------------------------------
406
407You can also track branches from repositories other than the one you
408cloned from, using gitlink:git-remote[1]:
409
410-------------------------------------------------
411$ git remote add linux-nfs git://linux-nfs.org/pub/nfs-2.6.git
04483524 412$ git fetch linux-nfs
d5cd5de4
BF
413* refs/remotes/linux-nfs/master: storing branch 'master' ...
414 commit: bf81b46
415-------------------------------------------------
416
417New remote-tracking branches will be stored under the shorthand name
418that you gave "git remote add", in this case linux-nfs:
419
420-------------------------------------------------
421$ git branch -r
422linux-nfs/master
423origin/master
424-------------------------------------------------
425
426If you run "git fetch <remote>" later, the tracking branches for the
427named <remote> will be updated.
428
429If you examine the file .git/config, you will see that git has added
430a new stanza:
431
432-------------------------------------------------
433$ cat .git/config
434...
435[remote "linux-nfs"]
923642fe
BF
436 url = git://linux-nfs.org/pub/nfs-2.6.git
437 fetch = +refs/heads/*:refs/remotes/linux-nfs/*
d5cd5de4
BF
438...
439-------------------------------------------------
440
fc90c536
BF
441This is what causes git to track the remote's branches; you may modify
442or delete these configuration options by editing .git/config with a
443text editor. (See the "CONFIGURATION FILE" section of
444gitlink:git-config[1] for details.)
d5cd5de4 445
e34caace 446[[exploring-git-history]]
d19fbc3c
BF
447Exploring git history
448=====================
449
450Git is best thought of as a tool for storing the history of a
451collection of files. It does this by storing compressed snapshots of
452the contents of a file heirarchy, together with "commits" which show
453the relationships between these snapshots.
454
455Git provides extremely flexible and fast tools for exploring the
456history of a project.
457
aacd404e 458We start with one specialized tool that is useful for finding the
d19fbc3c
BF
459commit that introduced a bug into a project.
460
e34caace 461[[using-bisect]]
d19fbc3c
BF
462How to use bisect to find a regression
463--------------------------------------
464
465Suppose version 2.6.18 of your project worked, but the version at
466"master" crashes. Sometimes the best way to find the cause of such a
467regression is to perform a brute-force search through the project's
468history to find the particular commit that caused the problem. The
469gitlink:git-bisect[1] command can help you do this:
470
471-------------------------------------------------
472$ git bisect start
473$ git bisect good v2.6.18
474$ git bisect bad master
475Bisecting: 3537 revisions left to test after this
476[65934a9a028b88e83e2b0f8b36618fe503349f8e] BLOCK: Make USB storage depend on SCSI rather than selecting it [try #6]
477-------------------------------------------------
478
479If you run "git branch" at this point, you'll see that git has
480temporarily moved you to a new branch named "bisect". This branch
481points to a commit (with commit id 65934...) that is reachable from
482v2.6.19 but not from v2.6.18. Compile and test it, and see whether
483it crashes. Assume it does crash. Then:
484
485-------------------------------------------------
486$ git bisect bad
487Bisecting: 1769 revisions left to test after this
488[7eff82c8b1511017ae605f0c99ac275a7e21b867] i2c-core: Drop useless bitmaskings
489-------------------------------------------------
490
491checks out an older version. Continue like this, telling git at each
492stage whether the version it gives you is good or bad, and notice
493that the number of revisions left to test is cut approximately in
494half each time.
495
496After about 13 tests (in this case), it will output the commit id of
497the guilty commit. You can then examine the commit with
498gitlink:git-show[1], find out who wrote it, and mail them your bug
499report with the commit id. Finally, run
500
501-------------------------------------------------
502$ git bisect reset
503-------------------------------------------------
504
505to return you to the branch you were on before and delete the
506temporary "bisect" branch.
507
508Note that the version which git-bisect checks out for you at each
509point is just a suggestion, and you're free to try a different
510version if you think it would be a good idea. For example,
511occasionally you may land on a commit that broke something unrelated;
512run
513
514-------------------------------------------------
04483524 515$ git bisect visualize
d19fbc3c
BF
516-------------------------------------------------
517
518which will run gitk and label the commit it chose with a marker that
519says "bisect". Chose a safe-looking commit nearby, note its commit
520id, and check it out with:
521
522-------------------------------------------------
523$ git reset --hard fb47ddb2db...
524-------------------------------------------------
525
526then test, run "bisect good" or "bisect bad" as appropriate, and
527continue.
528
e34caace 529[[naming-commits]]
d19fbc3c
BF
530Naming commits
531--------------
532
533We have seen several ways of naming commits already:
534
d55ae921 535 - 40-hexdigit object name
d19fbc3c
BF
536 - branch name: refers to the commit at the head of the given
537 branch
538 - tag name: refers to the commit pointed to by the given tag
539 (we've seen branches and tags are special cases of
540 <<how-git-stores-references,references>>).
541 - HEAD: refers to the head of the current branch
542
eb6ae7f4 543There are many more; see the "SPECIFYING REVISIONS" section of the
aec053bb 544gitlink:git-rev-parse[1] man page for the complete list of ways to
d19fbc3c
BF
545name revisions. Some examples:
546
547-------------------------------------------------
d55ae921 548$ git show fb47ddb2 # the first few characters of the object name
d19fbc3c
BF
549 # are usually enough to specify it uniquely
550$ git show HEAD^ # the parent of the HEAD commit
551$ git show HEAD^^ # the grandparent
552$ git show HEAD~4 # the great-great-grandparent
553-------------------------------------------------
554
555Recall that merge commits may have more than one parent; by default,
556^ and ~ follow the first parent listed in the commit, but you can
557also choose:
558
559-------------------------------------------------
560$ git show HEAD^1 # show the first parent of HEAD
561$ git show HEAD^2 # show the second parent of HEAD
562-------------------------------------------------
563
564In addition to HEAD, there are several other special names for
565commits:
566
567Merges (to be discussed later), as well as operations such as
568git-reset, which change the currently checked-out commit, generally
569set ORIG_HEAD to the value HEAD had before the current operation.
570
571The git-fetch operation always stores the head of the last fetched
572branch in FETCH_HEAD. For example, if you run git fetch without
573specifying a local branch as the target of the operation
574
575-------------------------------------------------
576$ git fetch git://example.com/proj.git theirbranch
577-------------------------------------------------
578
579the fetched commits will still be available from FETCH_HEAD.
580
581When we discuss merges we'll also see the special name MERGE_HEAD,
582which refers to the other branch that we're merging in to the current
583branch.
584
aec053bb 585The gitlink:git-rev-parse[1] command is a low-level command that is
d55ae921
BF
586occasionally useful for translating some name for a commit to the object
587name for that commit:
aec053bb
BF
588
589-------------------------------------------------
590$ git rev-parse origin
591e05db0fd4f31dde7005f075a84f96b360d05984b
592-------------------------------------------------
593
e34caace 594[[creating-tags]]
d19fbc3c
BF
595Creating tags
596-------------
597
598We can also create a tag to refer to a particular commit; after
599running
600
601-------------------------------------------------
04483524 602$ git tag stable-1 1b2e1d63ff
d19fbc3c
BF
603-------------------------------------------------
604
605You can use stable-1 to refer to the commit 1b2e1d63ff.
606
c64415e2
BF
607This creates a "lightweight" tag. If you would also like to include a
608comment with the tag, and possibly sign it cryptographically, then you
609should create a tag object instead; see the gitlink:git-tag[1] man page
610for details.
d19fbc3c 611
e34caace 612[[browsing-revisions]]
d19fbc3c
BF
613Browsing revisions
614------------------
615
616The gitlink:git-log[1] command can show lists of commits. On its
617own, it shows all commits reachable from the parent commit; but you
618can also make more specific requests:
619
620-------------------------------------------------
621$ git log v2.5.. # commits since (not reachable from) v2.5
622$ git log test..master # commits reachable from master but not test
623$ git log master..test # ...reachable from test but not master
624$ git log master...test # ...reachable from either test or master,
625 # but not both
626$ git log --since="2 weeks ago" # commits from the last 2 weeks
627$ git log Makefile # commits which modify Makefile
628$ git log fs/ # ... which modify any file under fs/
629$ git log -S'foo()' # commits which add or remove any file data
630 # matching the string 'foo()'
631-------------------------------------------------
632
633And of course you can combine all of these; the following finds
634commits since v2.5 which touch the Makefile or any file under fs:
635
636-------------------------------------------------
637$ git log v2.5.. Makefile fs/
638-------------------------------------------------
639
640You can also ask git log to show patches:
641
642-------------------------------------------------
643$ git log -p
644-------------------------------------------------
645
646See the "--pretty" option in the gitlink:git-log[1] man page for more
647display options.
648
649Note that git log starts with the most recent commit and works
650backwards through the parents; however, since git history can contain
3dff5379 651multiple independent lines of development, the particular order that
d19fbc3c
BF
652commits are listed in may be somewhat arbitrary.
653
e34caace 654[[generating-diffs]]
d19fbc3c
BF
655Generating diffs
656----------------
657
658You can generate diffs between any two versions using
659gitlink:git-diff[1]:
660
661-------------------------------------------------
662$ git diff master..test
663-------------------------------------------------
664
665Sometimes what you want instead is a set of patches:
666
667-------------------------------------------------
668$ git format-patch master..test
669-------------------------------------------------
670
671will generate a file with a patch for each commit reachable from test
672but not from master. Note that if master also has commits which are
673not reachable from test, then the combined result of these patches
674will not be the same as the diff produced by the git-diff example.
675
e34caace 676[[viewing-old-file-versions]]
d19fbc3c
BF
677Viewing old file versions
678-------------------------
679
680You can always view an old version of a file by just checking out the
681correct revision first. But sometimes it is more convenient to be
682able to view an old version of a single file without checking
683anything out; this command does that:
684
685-------------------------------------------------
686$ git show v2.5:fs/locks.c
687-------------------------------------------------
688
689Before the colon may be anything that names a commit, and after it
690may be any path to a file tracked by git.
691
e34caace 692[[history-examples]]
aec053bb
BF
693Examples
694--------
695
46acd3fa
BF
696[[counting-commits-on-a-branch]]
697Counting the number of commits on a branch
698~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
699
700Suppose you want to know how many commits you've made on "mybranch"
701since it diverged from "origin":
702
703-------------------------------------------------
704$ git log --pretty=oneline origin..mybranch | wc -l
705-------------------------------------------------
706
707Alternatively, you may often see this sort of thing done with the
708lower-level command gitlink:git-rev-list[1], which just lists the SHA1's
709of all the given commits:
710
711-------------------------------------------------
712$ git rev-list origin..mybranch | wc -l
713-------------------------------------------------
714
e34caace 715[[checking-for-equal-branches]]
aec053bb 716Check whether two branches point at the same history
2f99710c 717~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
aec053bb
BF
718
719Suppose you want to check whether two branches point at the same point
720in history.
721
722-------------------------------------------------
723$ git diff origin..master
724-------------------------------------------------
725
69f7ad73
BF
726will tell you whether the contents of the project are the same at the
727two branches; in theory, however, it's possible that the same project
728contents could have been arrived at by two different historical
d55ae921 729routes. You could compare the object names:
aec053bb
BF
730
731-------------------------------------------------
732$ git rev-list origin
733e05db0fd4f31dde7005f075a84f96b360d05984b
734$ git rev-list master
735e05db0fd4f31dde7005f075a84f96b360d05984b
736-------------------------------------------------
737
69f7ad73
BF
738Or you could recall that the ... operator selects all commits
739contained reachable from either one reference or the other but not
740both: so
aec053bb
BF
741
742-------------------------------------------------
743$ git log origin...master
744-------------------------------------------------
745
746will return no commits when the two branches are equal.
747
e34caace 748[[finding-tagged-descendants]]
b181d57f
BF
749Find first tagged version including a given fix
750~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
aec053bb 751
69f7ad73
BF
752Suppose you know that the commit e05db0fd fixed a certain problem.
753You'd like to find the earliest tagged release that contains that
754fix.
755
756Of course, there may be more than one answer--if the history branched
757after commit e05db0fd, then there could be multiple "earliest" tagged
758releases.
759
760You could just visually inspect the commits since e05db0fd:
761
762-------------------------------------------------
763$ gitk e05db0fd..
764-------------------------------------------------
765
b181d57f
BF
766Or you can use gitlink:git-name-rev[1], which will give the commit a
767name based on any tag it finds pointing to one of the commit's
768descendants:
769
770-------------------------------------------------
04483524 771$ git name-rev --tags e05db0fd
b181d57f
BF
772e05db0fd tags/v1.5.0-rc1^0~23
773-------------------------------------------------
774
775The gitlink:git-describe[1] command does the opposite, naming the
776revision using a tag on which the given commit is based:
777
778-------------------------------------------------
779$ git describe e05db0fd
04483524 780v1.5.0-rc0-260-ge05db0f
b181d57f
BF
781-------------------------------------------------
782
783but that may sometimes help you guess which tags might come after the
784given commit.
785
786If you just want to verify whether a given tagged version contains a
787given commit, you could use gitlink:git-merge-base[1]:
788
789-------------------------------------------------
790$ git merge-base e05db0fd v1.5.0-rc1
791e05db0fd4f31dde7005f075a84f96b360d05984b
792-------------------------------------------------
793
794The merge-base command finds a common ancestor of the given commits,
795and always returns one or the other in the case where one is a
796descendant of the other; so the above output shows that e05db0fd
797actually is an ancestor of v1.5.0-rc1.
798
799Alternatively, note that
800
801-------------------------------------------------
4a7979ca 802$ git log v1.5.0-rc1..e05db0fd
b181d57f
BF
803-------------------------------------------------
804
4a7979ca 805will produce empty output if and only if v1.5.0-rc1 includes e05db0fd,
b181d57f 806because it outputs only commits that are not reachable from v1.5.0-rc1.
aec053bb 807
4a7979ca
BF
808As yet another alternative, the gitlink:git-show-branch[1] command lists
809the commits reachable from its arguments with a display on the left-hand
810side that indicates which arguments that commit is reachable from. So,
811you can run something like
812
813-------------------------------------------------
814$ git show-branch e05db0fd v1.5.0-rc0 v1.5.0-rc1 v1.5.0-rc2
815! [e05db0fd] Fix warnings in sha1_file.c - use C99 printf format if
816available
817 ! [v1.5.0-rc0] GIT v1.5.0 preview
818 ! [v1.5.0-rc1] GIT v1.5.0-rc1
819 ! [v1.5.0-rc2] GIT v1.5.0-rc2
820...
821-------------------------------------------------
822
823then search for a line that looks like
824
825-------------------------------------------------
826+ ++ [e05db0fd] Fix warnings in sha1_file.c - use C99 printf format if
827available
828-------------------------------------------------
829
830Which shows that e05db0fd is reachable from itself, from v1.5.0-rc1, and
831from v1.5.0-rc2, but not from v1.5.0-rc0.
832
629d9f78
BF
833[[showing-commits-unique-to-a-branch]]
834Showing commits unique to a given branch
835~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
4a7979ca 836
629d9f78
BF
837Suppose you would like to see all the commits reachable from the branch
838head named "master" but not from any other head in your repository.
d19fbc3c 839
629d9f78
BF
840We can list all the heads in this repository with
841gitlink:git-show-ref[1]:
d19fbc3c 842
629d9f78
BF
843-------------------------------------------------
844$ git show-ref --heads
845bf62196b5e363d73353a9dcf094c59595f3153b7 refs/heads/core-tutorial
846db768d5504c1bb46f63ee9d6e1772bd047e05bf9 refs/heads/maint
847a07157ac624b2524a059a3414e99f6f44bebc1e7 refs/heads/master
84824dbc180ea14dc1aebe09f14c8ecf32010690627 refs/heads/tutorial-2
8491e87486ae06626c2f31eaa63d26fc0fd646c8af2 refs/heads/tutorial-fixes
850-------------------------------------------------
d19fbc3c 851
629d9f78
BF
852We can get just the branch-head names, and remove "master", with
853the help of the standard utilities cut and grep:
854
855-------------------------------------------------
856$ git show-ref --heads | cut -d' ' -f2 | grep -v '^refs/heads/master'
857refs/heads/core-tutorial
858refs/heads/maint
859refs/heads/tutorial-2
860refs/heads/tutorial-fixes
861-------------------------------------------------
862
863And then we can ask to see all the commits reachable from master
864but not from these other heads:
865
866-------------------------------------------------
867$ gitk master --not $( git show-ref --heads | cut -d' ' -f2 |
868 grep -v '^refs/heads/master' )
869-------------------------------------------------
870
871Obviously, endless variations are possible; for example, to see all
872commits reachable from some head but not from any tag in the repository:
873
874-------------------------------------------------
c78974f7 875$ gitk $( git show-ref --heads ) --not $( git show-ref --tags )
629d9f78
BF
876-------------------------------------------------
877
878(See gitlink:git-rev-parse[1] for explanations of commit-selecting
879syntax such as `--not`.)
880
82c8bf28
BF
881[[making-a-release]]
882Creating a changelog and tarball for a software release
883~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
884
885The gitlink:git-archive[1] command can create a tar or zip archive from
886any version of a project; for example:
887
888-------------------------------------------------
889$ git archive --format=tar --prefix=project/ HEAD | gzip >latest.tar.gz
890-------------------------------------------------
891
892will use HEAD to produce a tar archive in which each filename is
893preceded by "prefix/".
894
895If you're releasing a new version of a software project, you may want
896to simultaneously make a changelog to include in the release
897announcement.
898
899Linus Torvalds, for example, makes new kernel releases by tagging them,
900then running:
901
902-------------------------------------------------
903$ release-script 2.6.12 2.6.13-rc6 2.6.13-rc7
904-------------------------------------------------
905
906where release-script is a shell script that looks like:
907
908-------------------------------------------------
909#!/bin/sh
910stable="$1"
911last="$2"
912new="$3"
913echo "# git tag v$new"
914echo "git archive --prefix=linux-$new/ v$new | gzip -9 > ../linux-$new.tar.gz"
915echo "git diff v$stable v$new | gzip -9 > ../patch-$new.gz"
916echo "git log --no-merges v$new ^v$last > ../ChangeLog-$new"
917echo "git shortlog --no-merges v$new ^v$last > ../ShortLog"
918echo "git diff --stat --summary -M v$last v$new > ../diffstat-$new"
919-------------------------------------------------
920
921and then he just cut-and-pastes the output commands after verifying that
922they look OK.
4a7979ca 923
187b0d80
BF
924Finding commits referencing a file with given content
925-----------------------------------------------------
926
927Somebody hands you a copy of a file, and asks which commits modified a
928file such that it contained the given content either before or after the
929commit. You can find out with this:
930
931-------------------------------------------------
932$ git log --raw -r --abbrev=40 --pretty=oneline -- filename |
933 grep -B 1 `git hash-object filename`
934-------------------------------------------------
935
936Figuring out why this works is left as an exercise to the (advanced)
937student. The gitlink:git-log[1], gitlink:git-diff-tree[1], and
938gitlink:git-hash-object[1] man pages may prove helpful.
939
e34caace 940[[Developing-with-git]]
d19fbc3c
BF
941Developing with git
942===================
943
e34caace 944[[telling-git-your-name]]
d19fbc3c
BF
945Telling git your name
946---------------------
947
948Before creating any commits, you should introduce yourself to git. The
58c19d1f
BF
949easiest way to do so is to make sure the following lines appear in a
950file named .gitconfig in your home directory:
d19fbc3c
BF
951
952------------------------------------------------
d19fbc3c
BF
953[user]
954 name = Your Name Comes Here
955 email = you@yourdomain.example.com
d19fbc3c
BF
956------------------------------------------------
957
fc90c536
BF
958(See the "CONFIGURATION FILE" section of gitlink:git-config[1] for
959details on the configuration file.)
960
d19fbc3c 961
e34caace 962[[creating-a-new-repository]]
d19fbc3c
BF
963Creating a new repository
964-------------------------
965
966Creating a new repository from scratch is very easy:
967
968-------------------------------------------------
969$ mkdir project
970$ cd project
f1d2b477 971$ git init
d19fbc3c
BF
972-------------------------------------------------
973
974If you have some initial content (say, a tarball):
975
976-------------------------------------------------
977$ tar -xzvf project.tar.gz
978$ cd project
f1d2b477 979$ git init
d19fbc3c
BF
980$ git add . # include everything below ./ in the first commit:
981$ git commit
982-------------------------------------------------
983
984[[how-to-make-a-commit]]
ae25c67a 985How to make a commit
d19fbc3c
BF
986--------------------
987
988Creating a new commit takes three steps:
989
990 1. Making some changes to the working directory using your
991 favorite editor.
992 2. Telling git about your changes.
993 3. Creating the commit using the content you told git about
994 in step 2.
995
996In practice, you can interleave and repeat steps 1 and 2 as many
997times as you want: in order to keep track of what you want committed
998at step 3, git maintains a snapshot of the tree's contents in a
999special staging area called "the index."
1000
01997b4a
BF
1001At the beginning, the content of the index will be identical to
1002that of the HEAD. The command "git diff --cached", which shows
1003the difference between the HEAD and the index, should therefore
1004produce no output at that point.
eb6ae7f4 1005
d19fbc3c
BF
1006Modifying the index is easy:
1007
1008To update the index with the new contents of a modified file, use
1009
1010-------------------------------------------------
1011$ git add path/to/file
1012-------------------------------------------------
1013
1014To add the contents of a new file to the index, use
1015
1016-------------------------------------------------
1017$ git add path/to/file
1018-------------------------------------------------
1019
eb6ae7f4 1020To remove a file from the index and from the working tree,
d19fbc3c
BF
1021
1022-------------------------------------------------
1023$ git rm path/to/file
1024-------------------------------------------------
1025
1026After each step you can verify that
1027
1028-------------------------------------------------
1029$ git diff --cached
1030-------------------------------------------------
1031
1032always shows the difference between the HEAD and the index file--this
1033is what you'd commit if you created the commit now--and that
1034
1035-------------------------------------------------
1036$ git diff
1037-------------------------------------------------
1038
1039shows the difference between the working tree and the index file.
1040
1041Note that "git add" always adds just the current contents of a file
1042to the index; further changes to the same file will be ignored unless
1043you run git-add on the file again.
1044
1045When you're ready, just run
1046
1047-------------------------------------------------
1048$ git commit
1049-------------------------------------------------
1050
1051and git will prompt you for a commit message and then create the new
3dff5379 1052commit. Check to make sure it looks like what you expected with
d19fbc3c
BF
1053
1054-------------------------------------------------
1055$ git show
1056-------------------------------------------------
1057
1058As a special shortcut,
1059
1060-------------------------------------------------
1061$ git commit -a
1062-------------------------------------------------
1063
1064will update the index with any files that you've modified or removed
1065and create a commit, all in one step.
1066
1067A number of commands are useful for keeping track of what you're
1068about to commit:
1069
1070-------------------------------------------------
1071$ git diff --cached # difference between HEAD and the index; what
1072 # would be commited if you ran "commit" now.
1073$ git diff # difference between the index file and your
1074 # working directory; changes that would not
1075 # be included if you ran "commit" now.
c64415e2
BF
1076$ git diff HEAD # difference between HEAD and working tree; what
1077 # would be committed if you ran "commit -a" now.
d19fbc3c
BF
1078$ git status # a brief per-file summary of the above.
1079-------------------------------------------------
1080
e34caace 1081[[creating-good-commit-messages]]
ae25c67a 1082Creating good commit messages
d19fbc3c
BF
1083-----------------------------
1084
1085Though not required, it's a good idea to begin the commit message
1086with a single short (less than 50 character) line summarizing the
1087change, followed by a blank line and then a more thorough
1088description. Tools that turn commits into email, for example, use
1089the first line on the Subject line and the rest of the commit in the
1090body.
1091
2dc53617
JH
1092[[ignoring-files]]
1093Ignoring files
1094--------------
1095
1096A project will often generate files that you do 'not' want to track with git.
1097This typically includes files generated by a build process or temporary
1098backup files made by your editor. Of course, 'not' tracking files with git
1099is just a matter of 'not' calling "`git add`" on them. But it quickly becomes
1100annoying to have these untracked files lying around; e.g. they make
1101"`git add .`" and "`git commit -a`" practically useless, and they keep
1102showing up in the output of "`git status`", etc.
1103
1104Git therefore provides "exclude patterns" for telling git which files to
1105actively ignore. Exclude patterns are thoroughly explained in the
cedb8d5d
JT
1106gitlink:gitignore[5] manual page, but the heart of the concept is simply
1107a list of files which git should ignore. Entries in the list may contain
1108globs to specify multiple files, or may be prefixed by "`!`" to
1109explicitly include (un-ignore) a previously excluded (ignored) file
1110(i.e. later exclude patterns override earlier ones). The following
1111example should illustrate such patterns:
2dc53617
JH
1112
1113-------------------------------------------------
1114# Lines starting with '#' are considered comments.
1115# Ignore foo.txt.
1116foo.txt
1117# Ignore (generated) html files,
1118*.html
1119# except foo.html which is maintained by hand.
1120!foo.html
1121# Ignore objects and archives.
1122*.[oa]
1123-------------------------------------------------
1124
1125The next question is where to put these exclude patterns so that git can
1126find them. Git looks for exclude patterns in the following files:
1127
1128`.gitignore` files in your working tree:::
1129 You may store multiple `.gitignore` files at various locations in your
1130 working tree. Each `.gitignore` file is applied to the directory where
1131 it's located, including its subdirectories. Furthermore, the
1132 `.gitignore` files can be tracked like any other files in your working
1133 tree; just do a "`git add .gitignore`" and commit. `.gitignore` is
1134 therefore the right place to put exclude patterns that are meant to
1135 be shared between all project participants, such as build output files
1136 (e.g. `\*.o`), etc.
1137`.git/info/exclude` in your repo:::
1138 Exclude patterns in this file are applied to the working tree as a
1139 whole. Since the file is not located in your working tree, it does
1140 not follow push/pull/clone like `.gitignore` can do. This is therefore
1141 the place to put exclude patterns that are local to your copy of the
1142 repo (i.e. 'not' shared between project participants), such as
1143 temporary backup files made by your editor (e.g. `\*~`), etc.
1144The file specified by the `core.excludesfile` config directive:::
1145 By setting the `core.excludesfile` config directive you can tell git
1146 where to find more exclude patterns (see gitlink:git-config[1] for
1147 more information on configuration options). This config directive
1148 can be set in the per-repo `.git/config` file, in which case the
1149 exclude patterns will apply to that repo only. Alternatively, you
1150 can set the directive in the global `~/.gitconfig` file to apply
1151 the exclude pattern to all your git repos. As with the above
1152 `.git/info/exclude` (and, indeed, with git config directives in
1153 general), this directive does not follow push/pull/clone, but remain
1154 local to your repo(s).
1155
1156[NOTE]
1157In addition to the above alternatives, there are git commands that can take
1158exclude patterns directly on the command line. See gitlink:git-ls-files[1]
1159for an example of this.
1160
e34caace 1161[[how-to-merge]]
ae25c67a 1162How to merge
d19fbc3c
BF
1163------------
1164
1165You can rejoin two diverging branches of development using
1166gitlink:git-merge[1]:
1167
1168-------------------------------------------------
1169$ git merge branchname
1170-------------------------------------------------
1171
1172merges the development in the branch "branchname" into the current
1173branch. If there are conflicts--for example, if the same file is
1174modified in two different ways in the remote branch and the local
1175branch--then you are warned; the output may look something like this:
1176
1177-------------------------------------------------
fabbd8f6
BF
1178$ git merge next
1179 100% (4/4) done
1180Auto-merged file.txt
d19fbc3c
BF
1181CONFLICT (content): Merge conflict in file.txt
1182Automatic merge failed; fix conflicts and then commit the result.
1183-------------------------------------------------
1184
1185Conflict markers are left in the problematic files, and after
1186you resolve the conflicts manually, you can update the index
1187with the contents and run git commit, as you normally would when
1188creating a new file.
1189
1190If you examine the resulting commit using gitk, you will see that it
1191has two parents, one pointing to the top of the current branch, and
1192one to the top of the other branch.
1193
d19fbc3c
BF
1194[[resolving-a-merge]]
1195Resolving a merge
1196-----------------
1197
1198When a merge isn't resolved automatically, git leaves the index and
1199the working tree in a special state that gives you all the
1200information you need to help resolve the merge.
1201
1202Files with conflicts are marked specially in the index, so until you
ef561ac7
BF
1203resolve the problem and update the index, gitlink:git-commit[1] will
1204fail:
d19fbc3c
BF
1205
1206-------------------------------------------------
1207$ git commit
1208file.txt: needs merge
1209-------------------------------------------------
1210
ef561ac7
BF
1211Also, gitlink:git-status[1] will list those files as "unmerged", and the
1212files with conflicts will have conflict markers added, like this:
1213
1214-------------------------------------------------
1215<<<<<<< HEAD:file.txt
1216Hello world
1217=======
1218Goodbye
1219>>>>>>> 77976da35a11db4580b80ae27e8d65caf5208086:file.txt
1220-------------------------------------------------
1221
1222All you need to do is edit the files to resolve the conflicts, and then
1223
1224-------------------------------------------------
1225$ git add file.txt
1226$ git commit
1227-------------------------------------------------
1228
1229Note that the commit message will already be filled in for you with
1230some information about the merge. Normally you can just use this
1231default message unchanged, but you may add additional commentary of
1232your own if desired.
1233
1234The above is all you need to know to resolve a simple merge. But git
1235also provides more information to help resolve conflicts:
1236
e34caace 1237[[conflict-resolution]]
ef561ac7
BF
1238Getting conflict-resolution help during a merge
1239~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
d19fbc3c
BF
1240
1241All of the changes that git was able to merge automatically are
1242already added to the index file, so gitlink:git-diff[1] shows only
ef561ac7 1243the conflicts. It uses an unusual syntax:
d19fbc3c
BF
1244
1245-------------------------------------------------
1246$ git diff
1247diff --cc file.txt
1248index 802992c,2b60207..0000000
1249--- a/file.txt
1250+++ b/file.txt
1251@@@ -1,1 -1,1 +1,5 @@@
1252++<<<<<<< HEAD:file.txt
1253 +Hello world
1254++=======
1255+ Goodbye
1256++>>>>>>> 77976da35a11db4580b80ae27e8d65caf5208086:file.txt
1257-------------------------------------------------
1258
1259Recall that the commit which will be commited after we resolve this
1260conflict will have two parents instead of the usual one: one parent
1261will be HEAD, the tip of the current branch; the other will be the
1262tip of the other branch, which is stored temporarily in MERGE_HEAD.
1263
ef561ac7
BF
1264During the merge, the index holds three versions of each file. Each of
1265these three "file stages" represents a different version of the file:
1266
1267-------------------------------------------------
1268$ git show :1:file.txt # the file in a common ancestor of both branches
1269$ git show :2:file.txt # the version from HEAD, but including any
1270 # nonconflicting changes from MERGE_HEAD
1271$ git show :3:file.txt # the version from MERGE_HEAD, but including any
1272 # nonconflicting changes from HEAD.
1273-------------------------------------------------
1274
1275Since the stage 2 and stage 3 versions have already been updated with
1276nonconflicting changes, the only remaining differences between them are
1277the important ones; thus gitlink:git-diff[1] can use the information in
1278the index to show only those conflicts.
1279
1280The diff above shows the differences between the working-tree version of
1281file.txt and the stage 2 and stage 3 versions. So instead of preceding
1282each line by a single "+" or "-", it now uses two columns: the first
1283column is used for differences between the first parent and the working
1284directory copy, and the second for differences between the second parent
1285and the working directory copy. (See the "COMBINED DIFF FORMAT" section
1286of gitlink:git-diff-files[1] for a details of the format.)
1287
1288After resolving the conflict in the obvious way (but before updating the
1289index), the diff will look like:
d19fbc3c
BF
1290
1291-------------------------------------------------
1292$ git diff
1293diff --cc file.txt
1294index 802992c,2b60207..0000000
1295--- a/file.txt
1296+++ b/file.txt
1297@@@ -1,1 -1,1 +1,1 @@@
1298- Hello world
1299 -Goodbye
1300++Goodbye world
1301-------------------------------------------------
1302
1303This shows that our resolved version deleted "Hello world" from the
1304first parent, deleted "Goodbye" from the second parent, and added
1305"Goodbye world", which was previously absent from both.
1306
ef561ac7
BF
1307Some special diff options allow diffing the working directory against
1308any of these stages:
1309
1310-------------------------------------------------
1311$ git diff -1 file.txt # diff against stage 1
1312$ git diff --base file.txt # same as the above
1313$ git diff -2 file.txt # diff against stage 2
1314$ git diff --ours file.txt # same as the above
1315$ git diff -3 file.txt # diff against stage 3
1316$ git diff --theirs file.txt # same as the above.
1317-------------------------------------------------
1318
1319The gitlink:git-log[1] and gitk[1] commands also provide special help
1320for merges:
d19fbc3c
BF
1321
1322-------------------------------------------------
1323$ git log --merge
ef561ac7 1324$ gitk --merge
d19fbc3c
BF
1325-------------------------------------------------
1326
ef561ac7
BF
1327These will display all commits which exist only on HEAD or on
1328MERGE_HEAD, and which touch an unmerged file.
d19fbc3c 1329
61d72564 1330You may also use gitlink:git-mergetool[1], which lets you merge the
c64415e2
BF
1331unmerged files using external tools such as emacs or kdiff3.
1332
ef561ac7 1333Each time you resolve the conflicts in a file and update the index:
d19fbc3c
BF
1334
1335-------------------------------------------------
1336$ git add file.txt
d19fbc3c
BF
1337-------------------------------------------------
1338
ef561ac7
BF
1339the different stages of that file will be "collapsed", after which
1340git-diff will (by default) no longer show diffs for that file.
d19fbc3c
BF
1341
1342[[undoing-a-merge]]
ae25c67a 1343Undoing a merge
d19fbc3c
BF
1344---------------
1345
1346If you get stuck and decide to just give up and throw the whole mess
1347away, you can always return to the pre-merge state with
1348
1349-------------------------------------------------
1350$ git reset --hard HEAD
1351-------------------------------------------------
1352
1353Or, if you've already commited the merge that you want to throw away,
1354
1355-------------------------------------------------
1c73bb0e 1356$ git reset --hard ORIG_HEAD
d19fbc3c
BF
1357-------------------------------------------------
1358
1359However, this last command can be dangerous in some cases--never
1360throw away a commit you have already committed if that commit may
1361itself have been merged into another branch, as doing so may confuse
1362further merges.
1363
e34caace 1364[[fast-forwards]]
d19fbc3c
BF
1365Fast-forward merges
1366-------------------
1367
1368There is one special case not mentioned above, which is treated
1369differently. Normally, a merge results in a merge commit, with two
1370parents, one pointing at each of the two lines of development that
1371were merged.
1372
59723040
BF
1373However, if the current branch is a descendant of the other--so every
1374commit present in the one is already contained in the other--then git
1375just performs a "fast forward"; the head of the current branch is moved
1376forward to point at the head of the merged-in branch, without any new
1377commits being created.
d19fbc3c 1378
e34caace 1379[[fixing-mistakes]]
b684f830
BF
1380Fixing mistakes
1381---------------
1382
1383If you've messed up the working tree, but haven't yet committed your
1384mistake, you can return the entire working tree to the last committed
1385state with
1386
1387-------------------------------------------------
1388$ git reset --hard HEAD
1389-------------------------------------------------
1390
1391If you make a commit that you later wish you hadn't, there are two
1392fundamentally different ways to fix the problem:
1393
1394 1. You can create a new commit that undoes whatever was done
1395 by the previous commit. This is the correct thing if your
1396 mistake has already been made public.
1397
1398 2. You can go back and modify the old commit. You should
1399 never do this if you have already made the history public;
1400 git does not normally expect the "history" of a project to
1401 change, and cannot correctly perform repeated merges from
1402 a branch that has had its history changed.
1403
e34caace 1404[[reverting-a-commit]]
b684f830
BF
1405Fixing a mistake with a new commit
1406~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1407
1408Creating a new commit that reverts an earlier change is very easy;
1409just pass the gitlink:git-revert[1] command a reference to the bad
1410commit; for example, to revert the most recent commit:
1411
1412-------------------------------------------------
1413$ git revert HEAD
1414-------------------------------------------------
1415
1416This will create a new commit which undoes the change in HEAD. You
1417will be given a chance to edit the commit message for the new commit.
1418
1419You can also revert an earlier change, for example, the next-to-last:
1420
1421-------------------------------------------------
1422$ git revert HEAD^
1423-------------------------------------------------
1424
1425In this case git will attempt to undo the old change while leaving
1426intact any changes made since then. If more recent changes overlap
1427with the changes to be reverted, then you will be asked to fix
1428conflicts manually, just as in the case of <<resolving-a-merge,
1429resolving a merge>>.
1430
365aa199 1431[[fixing-a-mistake-by-editing-history]]
b684f830
BF
1432Fixing a mistake by editing history
1433~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1434
1435If the problematic commit is the most recent commit, and you have not
1436yet made that commit public, then you may just
1437<<undoing-a-merge,destroy it using git-reset>>.
1438
1439Alternatively, you
1440can edit the working directory and update the index to fix your
1441mistake, just as if you were going to <<how-to-make-a-commit,create a
1442new commit>>, then run
1443
1444-------------------------------------------------
1445$ git commit --amend
1446-------------------------------------------------
1447
1448which will replace the old commit by a new commit incorporating your
1449changes, giving you a chance to edit the old commit message first.
1450
1451Again, you should never do this to a commit that may already have
1452been merged into another branch; use gitlink:git-revert[1] instead in
1453that case.
1454
1455It is also possible to edit commits further back in the history, but
1456this is an advanced topic to be left for
1457<<cleaning-up-history,another chapter>>.
1458
e34caace 1459[[checkout-of-path]]
b684f830
BF
1460Checking out an old version of a file
1461~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1462
1463In the process of undoing a previous bad change, you may find it
1464useful to check out an older version of a particular file using
1465gitlink:git-checkout[1]. We've used git checkout before to switch
1466branches, but it has quite different behavior if it is given a path
1467name: the command
1468
1469-------------------------------------------------
1470$ git checkout HEAD^ path/to/file
1471-------------------------------------------------
1472
1473replaces path/to/file by the contents it had in the commit HEAD^, and
1474also updates the index to match. It does not change branches.
1475
1476If you just want to look at an old version of the file, without
1477modifying the working directory, you can do that with
1478gitlink:git-show[1]:
1479
1480-------------------------------------------------
ed4eb0d8 1481$ git show HEAD^:path/to/file
b684f830
BF
1482-------------------------------------------------
1483
1484which will display the given version of the file.
1485
e34caace 1486[[ensuring-good-performance]]
d19fbc3c
BF
1487Ensuring good performance
1488-------------------------
1489
1490On large repositories, git depends on compression to keep the history
1491information from taking up to much space on disk or in memory.
1492
1493This compression is not performed automatically. Therefore you
17217090 1494should occasionally run gitlink:git-gc[1]:
d19fbc3c
BF
1495
1496-------------------------------------------------
1497$ git gc
1498-------------------------------------------------
1499
17217090
BF
1500to recompress the archive. This can be very time-consuming, so
1501you may prefer to run git-gc when you are not doing other work.
d19fbc3c 1502
e34caace
BF
1503
1504[[ensuring-reliability]]
11e016a3
BF
1505Ensuring reliability
1506--------------------
1507
e34caace 1508[[checking-for-corruption]]
11e016a3
BF
1509Checking the repository for corruption
1510~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1511
1191ee18
BF
1512The gitlink:git-fsck[1] command runs a number of self-consistency checks
1513on the repository, and reports on any problems. This may take some
21dcb3b7
BF
1514time. The most common warning by far is about "dangling" objects:
1515
1516-------------------------------------------------
04e50e94 1517$ git fsck
21dcb3b7
BF
1518dangling commit 7281251ddd2a61e38657c827739c57015671a6b3
1519dangling commit 2706a059f258c6b245f298dc4ff2ccd30ec21a63
1520dangling commit 13472b7c4b80851a1bc551779171dcb03655e9b5
1521dangling blob 218761f9d90712d37a9c5e36f406f92202db07eb
1522dangling commit bf093535a34a4d35731aa2bd90fe6b176302f14f
1523dangling commit 8e4bec7f2ddaa268bef999853c25755452100f8e
1524dangling tree d50bb86186bf27b681d25af89d3b5b68382e4085
1525dangling tree b24c2473f1fd3d91352a624795be026d64c8841f
1526...
1527-------------------------------------------------
1528
59723040
BF
1529Dangling objects are not a problem. At worst they may take up a little
1530extra disk space. They can sometimes provide a last-resort method of
1531recovery lost work--see <<dangling-objects>> for details. However, if
1532you want, you may remove them with gitlink:git-prune[1] or the --prune
1191ee18 1533option to gitlink:git-gc[1]:
21dcb3b7
BF
1534
1535-------------------------------------------------
1536$ git gc --prune
1537-------------------------------------------------
1538
1191ee18
BF
1539This may be time-consuming. Unlike most other git operations (including
1540git-gc when run without any options), it is not safe to prune while
1541other git operations are in progress in the same repository.
21dcb3b7 1542
e34caace 1543[[recovering-lost-changes]]
11e016a3
BF
1544Recovering lost changes
1545~~~~~~~~~~~~~~~~~~~~~~~
1546
e34caace 1547[[reflogs]]
559e4d7a
BF
1548Reflogs
1549^^^^^^^
1550
1551Say you modify a branch with gitlink:git-reset[1] --hard, and then
1552realize that the branch was the only reference you had to that point in
1553history.
1554
1555Fortunately, git also keeps a log, called a "reflog", of all the
1556previous values of each branch. So in this case you can still find the
1557old history using, for example,
1558
1559-------------------------------------------------
1560$ git log master@{1}
1561-------------------------------------------------
1562
1563This lists the commits reachable from the previous version of the head.
1564This syntax can be used to with any git command that accepts a commit,
1565not just with git log. Some other examples:
1566
1567-------------------------------------------------
1568$ git show master@{2} # See where the branch pointed 2,
1569$ git show master@{3} # 3, ... changes ago.
1570$ gitk master@{yesterday} # See where it pointed yesterday,
1571$ gitk master@{"1 week ago"} # ... or last week
953f3d6f
BF
1572$ git log --walk-reflogs master # show reflog entries for master
1573-------------------------------------------------
1574
1575A separate reflog is kept for the HEAD, so
1576
1577-------------------------------------------------
1578$ git show HEAD@{"1 week ago"}
559e4d7a
BF
1579-------------------------------------------------
1580
953f3d6f
BF
1581will show what HEAD pointed to one week ago, not what the current branch
1582pointed to one week ago. This allows you to see the history of what
1583you've checked out.
1584
559e4d7a 1585The reflogs are kept by default for 30 days, after which they may be
036be17e 1586pruned. See gitlink:git-reflog[1] and gitlink:git-gc[1] to learn
559e4d7a
BF
1587how to control this pruning, and see the "SPECIFYING REVISIONS"
1588section of gitlink:git-rev-parse[1] for details.
1589
1590Note that the reflog history is very different from normal git history.
1591While normal history is shared by every repository that works on the
1592same project, the reflog history is not shared: it tells you only about
1593how the branches in your local repository have changed over time.
1594
59723040 1595[[dangling-object-recovery]]
559e4d7a
BF
1596Examining dangling objects
1597^^^^^^^^^^^^^^^^^^^^^^^^^^
1598
59723040
BF
1599In some situations the reflog may not be able to save you. For example,
1600suppose you delete a branch, then realize you need the history it
1601contained. The reflog is also deleted; however, if you have not yet
1602pruned the repository, then you may still be able to find the lost
1603commits in the dangling objects that git-fsck reports. See
1604<<dangling-objects>> for the details.
559e4d7a
BF
1605
1606-------------------------------------------------
1607$ git fsck
1608dangling commit 7281251ddd2a61e38657c827739c57015671a6b3
1609dangling commit 2706a059f258c6b245f298dc4ff2ccd30ec21a63
1610dangling commit 13472b7c4b80851a1bc551779171dcb03655e9b5
1611...
1612-------------------------------------------------
1613
aacd404e 1614You can examine
559e4d7a
BF
1615one of those dangling commits with, for example,
1616
1617------------------------------------------------
1618$ gitk 7281251ddd --not --all
1619------------------------------------------------
1620
1621which does what it sounds like: it says that you want to see the commit
1622history that is described by the dangling commit(s), but not the
1623history that is described by all your existing branches and tags. Thus
1624you get exactly the history reachable from that commit that is lost.
1625(And notice that it might not be just one commit: we only report the
1626"tip of the line" as being dangling, but there might be a whole deep
79c96c57 1627and complex commit history that was dropped.)
559e4d7a
BF
1628
1629If you decide you want the history back, you can always create a new
1630reference pointing to it, for example, a new branch:
1631
1632------------------------------------------------
1633$ git branch recovered-branch 7281251ddd
1634------------------------------------------------
1635
59723040
BF
1636Other types of dangling objects (blobs and trees) are also possible, and
1637dangling objects can arise in other situations.
1638
11e016a3 1639
e34caace 1640[[sharing-development]]
d19fbc3c 1641Sharing development with others
b684f830 1642===============================
d19fbc3c
BF
1643
1644[[getting-updates-with-git-pull]]
1645Getting updates with git pull
b684f830 1646-----------------------------
d19fbc3c
BF
1647
1648After you clone a repository and make a few changes of your own, you
1649may wish to check the original repository for updates and merge them
1650into your own work.
1651
1652We have already seen <<Updating-a-repository-with-git-fetch,how to
1653keep remote tracking branches up to date>> with gitlink:git-fetch[1],
1654and how to merge two branches. So you can merge in changes from the
1655original repository's master branch with:
1656
1657-------------------------------------------------
1658$ git fetch
1659$ git merge origin/master
1660-------------------------------------------------
1661
1662However, the gitlink:git-pull[1] command provides a way to do this in
1663one step:
1664
1665-------------------------------------------------
1666$ git pull origin master
1667-------------------------------------------------
1668
1669In fact, "origin" is normally the default repository to pull from,
1670and the default branch is normally the HEAD of the remote repository,
1671so often you can accomplish the above with just
1672
1673-------------------------------------------------
1674$ git pull
1675-------------------------------------------------
1676
c64415e2
BF
1677See the descriptions of the branch.<name>.remote and branch.<name>.merge
1678options in gitlink:git-config[1] to learn how to control these defaults
1679depending on the current branch. Also note that the --track option to
1680gitlink:git-branch[1] and gitlink:git-checkout[1] can be used to
1681automatically set the default remote branch to pull from at the time
1682that a branch is created:
1683
1684-------------------------------------------------
1685$ git checkout --track -b origin/maint maint
1686-------------------------------------------------
d19fbc3c
BF
1687
1688In addition to saving you keystrokes, "git pull" also helps you by
1689producing a default commit message documenting the branch and
1690repository that you pulled from.
1691
1692(But note that no such commit will be created in the case of a
1693<<fast-forwards,fast forward>>; instead, your branch will just be
79c96c57 1694updated to point to the latest commit from the upstream branch.)
d19fbc3c 1695
1191ee18
BF
1696The git-pull command can also be given "." as the "remote" repository,
1697in which case it just merges in a branch from the current repository; so
4c63ff45
BF
1698the commands
1699
1700-------------------------------------------------
1701$ git pull . branch
1702$ git merge branch
1703-------------------------------------------------
1704
1705are roughly equivalent. The former is actually very commonly used.
1706
e34caace 1707[[submitting-patches]]
d19fbc3c 1708Submitting patches to a project
b684f830 1709-------------------------------
d19fbc3c
BF
1710
1711If you just have a few changes, the simplest way to submit them may
1712just be to send them as patches in email:
1713
036be17e 1714First, use gitlink:git-format-patch[1]; for example:
d19fbc3c
BF
1715
1716-------------------------------------------------
eb6ae7f4 1717$ git format-patch origin
d19fbc3c
BF
1718-------------------------------------------------
1719
1720will produce a numbered series of files in the current directory, one
1721for each patch in the current branch but not in origin/HEAD.
1722
1723You can then import these into your mail client and send them by
1724hand. However, if you have a lot to send at once, you may prefer to
1725use the gitlink:git-send-email[1] script to automate the process.
1726Consult the mailing list for your project first to determine how they
1727prefer such patches be handled.
1728
e34caace 1729[[importing-patches]]
d19fbc3c 1730Importing patches to a project
b684f830 1731------------------------------
d19fbc3c
BF
1732
1733Git also provides a tool called gitlink:git-am[1] (am stands for
1734"apply mailbox"), for importing such an emailed series of patches.
1735Just save all of the patch-containing messages, in order, into a
1736single mailbox file, say "patches.mbox", then run
1737
1738-------------------------------------------------
eb6ae7f4 1739$ git am -3 patches.mbox
d19fbc3c
BF
1740-------------------------------------------------
1741
1742Git will apply each patch in order; if any conflicts are found, it
1743will stop, and you can fix the conflicts as described in
01997b4a
BF
1744"<<resolving-a-merge,Resolving a merge>>". (The "-3" option tells
1745git to perform a merge; if you would prefer it just to abort and
1746leave your tree and index untouched, you may omit that option.)
1747
1748Once the index is updated with the results of the conflict
1749resolution, instead of creating a new commit, just run
d19fbc3c
BF
1750
1751-------------------------------------------------
1752$ git am --resolved
1753-------------------------------------------------
1754
1755and git will create the commit for you and continue applying the
1756remaining patches from the mailbox.
1757
1758The final result will be a series of commits, one for each patch in
1759the original mailbox, with authorship and commit log message each
1760taken from the message containing each patch.
1761
eda69449
BF
1762[[public-repositories]]
1763Public git repositories
1764-----------------------
d19fbc3c 1765
eda69449
BF
1766Another way to submit changes to a project is to tell the maintainer of
1767that project to pull the changes from your repository using git-pull[1].
1768In the section "<<getting-updates-with-git-pull, Getting updates with
1769git pull>>" we described this as a way to get updates from the "main"
1770repository, but it works just as well in the other direction.
d19fbc3c 1771
eda69449
BF
1772If you and the maintainer both have accounts on the same machine, then
1773you can just pull changes from each other's repositories directly;
1774commands that accepts repository URLs as arguments will also accept a
1775local directory name:
d19fbc3c
BF
1776
1777-------------------------------------------------
1778$ git clone /path/to/repository
1779$ git pull /path/to/other/repository
1780-------------------------------------------------
1781
eda69449
BF
1782However, the more common way to do this is to maintain a separate public
1783repository (usually on a different host) for others to pull changes
1784from. This is usually more convenient, and allows you to cleanly
1785separate private work in progress from publicly visible work.
d19fbc3c
BF
1786
1787You will continue to do your day-to-day work in your personal
1788repository, but periodically "push" changes from your personal
1789repository into your public repository, allowing other developers to
1790pull from that repository. So the flow of changes, in a situation
1791where there is one other developer with a public repository, looks
1792like this:
1793
1794 you push
1795 your personal repo ------------------> your public repo
1796 ^ |
1797 | |
1798 | you pull | they pull
1799 | |
1800 | |
1801 | they push V
1802 their public repo <------------------- their repo
1803
eda69449
BF
1804[[setting-up-a-public-repository]]
1805Setting up a public repository
1806~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1807
1808Assume your personal repository is in the directory ~/proj. We
1809first create a new clone of the repository and tell git-daemon that it
1810is meant to be public:
d19fbc3c
BF
1811
1812-------------------------------------------------
52c80037 1813$ git clone --bare ~/proj proj.git
eda69449 1814$ touch proj.git/git-daemon-export-ok
d19fbc3c
BF
1815-------------------------------------------------
1816
52c80037 1817The resulting directory proj.git contains a "bare" git repository--it is
eda69449
BF
1818just the contents of the ".git" directory, without any files checked out
1819around it.
d19fbc3c 1820
c64415e2 1821Next, copy proj.git to the server where you plan to host the
d19fbc3c
BF
1822public repository. You can use scp, rsync, or whatever is most
1823convenient.
1824
eda69449
BF
1825[[exporting-via-git]]
1826Exporting a git repository via the git protocol
1827~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1828
1829This is the preferred method.
1830
1831If someone else administers the server, they should tell you what
1832directory to put the repository in, and what git:// url it will appear
1833at. You can then skip to the section
d19fbc3c
BF
1834"<<pushing-changes-to-a-public-repository,Pushing changes to a public
1835repository>>", below.
1836
eda69449
BF
1837Otherwise, all you need to do is start gitlink:git-daemon[1]; it will
1838listen on port 9418. By default, it will allow access to any directory
1839that looks like a git directory and contains the magic file
1840git-daemon-export-ok. Passing some directory paths as git-daemon
1841arguments will further restrict the exports to those paths.
1842
1843You can also run git-daemon as an inetd service; see the
1844gitlink:git-daemon[1] man page for details. (See especially the
1845examples section.)
d19fbc3c
BF
1846
1847[[exporting-via-http]]
1848Exporting a git repository via http
eda69449 1849~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
d19fbc3c
BF
1850
1851The git protocol gives better performance and reliability, but on a
1852host with a web server set up, http exports may be simpler to set up.
1853
1854All you need to do is place the newly created bare git repository in
1855a directory that is exported by the web server, and make some
1856adjustments to give web clients some extra information they need:
1857
1858-------------------------------------------------
1859$ mv proj.git /home/you/public_html/proj.git
1860$ cd proj.git
c64415e2 1861$ git --bare update-server-info
d19fbc3c
BF
1862$ chmod a+x hooks/post-update
1863-------------------------------------------------
1864
1865(For an explanation of the last two lines, see
1866gitlink:git-update-server-info[1], and the documentation
a2983cb7 1867link:hooks.html[Hooks used by git].)
d19fbc3c
BF
1868
1869Advertise the url of proj.git. Anybody else should then be able to
1870clone or pull from that url, for example with a commandline like:
1871
1872-------------------------------------------------
1873$ git clone http://yourserver.com/~you/proj.git
1874-------------------------------------------------
1875
1876(See also
1877link:howto/setup-git-server-over-http.txt[setup-git-server-over-http]
1878for a slightly more sophisticated setup using WebDAV which also
1879allows pushing over http.)
1880
d19fbc3c
BF
1881[[pushing-changes-to-a-public-repository]]
1882Pushing changes to a public repository
eda69449 1883~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
d19fbc3c 1884
eda69449 1885Note that the two techniques outlined above (exporting via
d19fbc3c
BF
1886<<exporting-via-http,http>> or <<exporting-via-git,git>>) allow other
1887maintainers to fetch your latest changes, but they do not allow write
1888access, which you will need to update the public repository with the
1889latest changes created in your private repository.
1890
1891The simplest way to do this is using gitlink:git-push[1] and ssh; to
1892update the remote branch named "master" with the latest state of your
1893branch named "master", run
1894
1895-------------------------------------------------
1896$ git push ssh://yourserver.com/~you/proj.git master:master
1897-------------------------------------------------
1898
1899or just
1900
1901-------------------------------------------------
1902$ git push ssh://yourserver.com/~you/proj.git master
1903-------------------------------------------------
1904
1905As with git-fetch, git-push will complain if this does not result in
1906a <<fast-forwards,fast forward>>. Normally this is a sign of
1907something wrong. However, if you are sure you know what you're
1908doing, you may force git-push to perform the update anyway by
1909proceeding the branch name by a plus sign:
1910
1911-------------------------------------------------
1912$ git push ssh://yourserver.com/~you/proj.git +master
1913-------------------------------------------------
1914
1915As with git-fetch, you may also set up configuration options to
1916save typing; so, for example, after
1917
1918-------------------------------------------------
c64415e2 1919$ cat >>.git/config <<EOF
d19fbc3c
BF
1920[remote "public-repo"]
1921 url = ssh://yourserver.com/~you/proj.git
1922EOF
1923-------------------------------------------------
1924
1925you should be able to perform the above push with just
1926
1927-------------------------------------------------
1928$ git push public-repo master
1929-------------------------------------------------
1930
1931See the explanations of the remote.<name>.url, branch.<name>.remote,
9d13bda3 1932and remote.<name>.push options in gitlink:git-config[1] for
d19fbc3c
BF
1933details.
1934
e34caace 1935[[setting-up-a-shared-repository]]
d19fbc3c 1936Setting up a shared repository
eda69449 1937~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
d19fbc3c
BF
1938
1939Another way to collaborate is by using a model similar to that
1940commonly used in CVS, where several developers with special rights
1941all push to and pull from a single shared repository. See
a2983cb7 1942link:cvs-migration.html[git for CVS users] for instructions on how to
d19fbc3c
BF
1943set this up.
1944
8fae2225
BF
1945However, while there is nothing wrong with git's support for shared
1946repositories, this mode of operation is not generally recommended,
1947simply because the mode of collaboration that git supports--by
1948exchanging patches and pulling from public repositories--has so many
1949advantages over the central shared repository:
1950
1951 - Git's ability to quickly import and merge patches allows a
1952 single maintainer to process incoming changes even at very
1953 high rates. And when that becomes too much, git-pull provides
1954 an easy way for that maintainer to delegate this job to other
1955 maintainers while still allowing optional review of incoming
1956 changes.
1957 - Since every developer's repository has the same complete copy
1958 of the project history, no repository is special, and it is
1959 trivial for another developer to take over maintenance of a
1960 project, either by mutual agreement, or because a maintainer
1961 becomes unresponsive or difficult to work with.
1962 - The lack of a central group of "committers" means there is
1963 less need for formal decisions about who is "in" and who is
1964 "out".
1965
e34caace 1966[[setting-up-gitweb]]
eda69449
BF
1967Allowing web browsing of a repository
1968~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
d19fbc3c 1969
a8cd1402
BF
1970The gitweb cgi script provides users an easy way to browse your
1971project's files and history without having to install git; see the file
04483524 1972gitweb/INSTALL in the git source tree for instructions on setting it up.
d19fbc3c 1973
e34caace 1974[[sharing-development-examples]]
b684f830
BF
1975Examples
1976--------
d19fbc3c 1977
9e2163ea
BF
1978[[maintaining-topic-branches]]
1979Maintaining topic branches for a Linux subsystem maintainer
1980~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1981
1982This describes how Tony Luck uses git in his role as maintainer of the
1983IA64 architecture for the Linux kernel.
1984
1985He uses two public branches:
1986
1987 - A "test" tree into which patches are initially placed so that they
1988 can get some exposure when integrated with other ongoing development.
1989 This tree is available to Andrew for pulling into -mm whenever he
1990 wants.
1991
1992 - A "release" tree into which tested patches are moved for final sanity
1993 checking, and as a vehicle to send them upstream to Linus (by sending
1994 him a "please pull" request.)
1995
1996He also uses a set of temporary branches ("topic branches"), each
1997containing a logical grouping of patches.
1998
1999To set this up, first create your work tree by cloning Linus's public
2000tree:
2001
2002-------------------------------------------------
2003$ git clone git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6.git work
2004$ cd work
2005-------------------------------------------------
2006
2007Linus's tree will be stored in the remote branch named origin/master,
2008and can be updated using gitlink:git-fetch[1]; you can track other
2009public trees using gitlink:git-remote[1] to set up a "remote" and
2010git-fetch[1] to keep them up-to-date; see <<repositories-and-branches>>.
2011
2012Now create the branches in which you are going to work; these start out
2013at the current tip of origin/master branch, and should be set up (using
2014the --track option to gitlink:git-branch[1]) to merge changes in from
2015Linus by default.
2016
2017-------------------------------------------------
2018$ git branch --track test origin/master
2019$ git branch --track release origin/master
2020-------------------------------------------------
2021
2022These can be easily kept up to date using gitlink:git-pull[1]
2023
2024-------------------------------------------------
2025$ git checkout test && git pull
2026$ git checkout release && git pull
2027-------------------------------------------------
2028
2029Important note! If you have any local changes in these branches, then
2030this merge will create a commit object in the history (with no local
2031changes git will simply do a "Fast forward" merge). Many people dislike
2032the "noise" that this creates in the Linux history, so you should avoid
2033doing this capriciously in the "release" branch, as these noisy commits
2034will become part of the permanent history when you ask Linus to pull
2035from the release branch.
2036
2037A few configuration variables (see gitlink:git-config[1]) can
2038make it easy to push both branches to your public tree. (See
2039<<setting-up-a-public-repository>>.)
2040
2041-------------------------------------------------
2042$ cat >> .git/config <<EOF
2043[remote "mytree"]
2044 url = master.kernel.org:/pub/scm/linux/kernel/git/aegl/linux-2.6.git
2045 push = release
2046 push = test
2047EOF
2048-------------------------------------------------
2049
2050Then you can push both the test and release trees using
2051gitlink:git-push[1]:
2052
2053-------------------------------------------------
2054$ git push mytree
2055-------------------------------------------------
2056
2057or push just one of the test and release branches using:
2058
2059-------------------------------------------------
2060$ git push mytree test
2061-------------------------------------------------
2062
2063or
2064
2065-------------------------------------------------
2066$ git push mytree release
2067-------------------------------------------------
2068
2069Now to apply some patches from the community. Think of a short
2070snappy name for a branch to hold this patch (or related group of
2071patches), and create a new branch from the current tip of Linus's
2072branch:
2073
2074-------------------------------------------------
2075$ git checkout -b speed-up-spinlocks origin
2076-------------------------------------------------
2077
2078Now you apply the patch(es), run some tests, and commit the change(s). If
2079the patch is a multi-part series, then you should apply each as a separate
2080commit to this branch.
2081
2082-------------------------------------------------
2083$ ... patch ... test ... commit [ ... patch ... test ... commit ]*
2084-------------------------------------------------
2085
2086When you are happy with the state of this change, you can pull it into the
2087"test" branch in preparation to make it public:
2088
2089-------------------------------------------------
2090$ git checkout test && git pull . speed-up-spinlocks
2091-------------------------------------------------
2092
2093It is unlikely that you would have any conflicts here ... but you might if you
2094spent a while on this step and had also pulled new versions from upstream.
2095
2096Some time later when enough time has passed and testing done, you can pull the
2097same branch into the "release" tree ready to go upstream. This is where you
2098see the value of keeping each patch (or patch series) in its own branch. It
2099means that the patches can be moved into the "release" tree in any order.
2100
2101-------------------------------------------------
2102$ git checkout release && git pull . speed-up-spinlocks
2103-------------------------------------------------
2104
2105After a while, you will have a number of branches, and despite the
2106well chosen names you picked for each of them, you may forget what
2107they are for, or what status they are in. To get a reminder of what
2108changes are in a specific branch, use:
2109
2110-------------------------------------------------
2111$ git log linux..branchname | git-shortlog
2112-------------------------------------------------
2113
2114To see whether it has already been merged into the test or release branches
2115use:
2116
2117-------------------------------------------------
2118$ git log test..branchname
2119-------------------------------------------------
2120
2121or
2122
2123-------------------------------------------------
2124$ git log release..branchname
2125-------------------------------------------------
2126
2127(If this branch has not yet been merged you will see some log entries.
2128If it has been merged, then there will be no output.)
2129
2130Once a patch completes the great cycle (moving from test to release,
2131then pulled by Linus, and finally coming back into your local
2132"origin/master" branch) the branch for this change is no longer needed.
2133You detect this when the output from:
2134
2135-------------------------------------------------
2136$ git log origin..branchname
2137-------------------------------------------------
2138
2139is empty. At this point the branch can be deleted:
2140
2141-------------------------------------------------
2142$ git branch -d branchname
2143-------------------------------------------------
2144
2145Some changes are so trivial that it is not necessary to create a separate
2146branch and then merge into each of the test and release branches. For
2147these changes, just apply directly to the "release" branch, and then
2148merge that into the "test" branch.
2149
2150To create diffstat and shortlog summaries of changes to include in a "please
2151pull" request to Linus you can use:
2152
2153-------------------------------------------------
2154$ git diff --stat origin..release
2155-------------------------------------------------
2156
2157and
2158
2159-------------------------------------------------
2160$ git log -p origin..release | git shortlog
2161-------------------------------------------------
2162
2163Here are some of the scripts that simplify all this even further.
2164
2165-------------------------------------------------
2166==== update script ====
2167# Update a branch in my GIT tree. If the branch to be updated
2168# is origin, then pull from kernel.org. Otherwise merge
2169# origin/master branch into test|release branch
2170
2171case "$1" in
2172test|release)
2173 git checkout $1 && git pull . origin
2174 ;;
2175origin)
2176 before=$(cat .git/refs/remotes/origin/master)
2177 git fetch origin
2178 after=$(cat .git/refs/remotes/origin/master)
2179 if [ $before != $after ]
2180 then
2181 git log $before..$after | git shortlog
2182 fi
2183 ;;
2184*)
2185 echo "Usage: $0 origin|test|release" 1>&2
2186 exit 1
2187 ;;
2188esac
2189-------------------------------------------------
2190
2191-------------------------------------------------
2192==== merge script ====
2193# Merge a branch into either the test or release branch
2194
2195pname=$0
2196
2197usage()
2198{
2199 echo "Usage: $pname branch test|release" 1>&2
2200 exit 1
2201}
2202
2203if [ ! -f .git/refs/heads/"$1" ]
2204then
2205 echo "Can't see branch <$1>" 1>&2
2206 usage
2207fi
2208
2209case "$2" in
2210test|release)
2211 if [ $(git log $2..$1 | wc -c) -eq 0 ]
2212 then
2213 echo $1 already merged into $2 1>&2
2214 exit 1
2215 fi
2216 git checkout $2 && git pull . $1
2217 ;;
2218*)
2219 usage
2220 ;;
2221esac
2222-------------------------------------------------
2223
2224-------------------------------------------------
2225==== status script ====
2226# report on status of my ia64 GIT tree
2227
2228gb=$(tput setab 2)
2229rb=$(tput setab 1)
2230restore=$(tput setab 9)
2231
2232if [ `git rev-list test..release | wc -c` -gt 0 ]
2233then
2234 echo $rb Warning: commits in release that are not in test $restore
2235 git log test..release
2236fi
2237
2238for branch in `ls .git/refs/heads`
2239do
2240 if [ $branch = test -o $branch = release ]
2241 then
2242 continue
2243 fi
2244
2245 echo -n $gb ======= $branch ====== $restore " "
2246 status=
2247 for ref in test release origin/master
2248 do
2249 if [ `git rev-list $ref..$branch | wc -c` -gt 0 ]
2250 then
2251 status=$status${ref:0:1}
2252 fi
2253 done
2254 case $status in
2255 trl)
2256 echo $rb Need to pull into test $restore
2257 ;;
2258 rl)
2259 echo "In test"
2260 ;;
2261 l)
2262 echo "Waiting for linus"
2263 ;;
2264 "")
2265 echo $rb All done $restore
2266 ;;
2267 *)
2268 echo $rb "<$status>" $restore
2269 ;;
2270 esac
2271 git log origin/master..$branch | git shortlog
2272done
2273-------------------------------------------------
d19fbc3c 2274
d19fbc3c 2275
d19fbc3c 2276[[cleaning-up-history]]
4c63ff45
BF
2277Rewriting history and maintaining patch series
2278==============================================
2279
2280Normally commits are only added to a project, never taken away or
2281replaced. Git is designed with this assumption, and violating it will
2282cause git's merge machinery (for example) to do the wrong thing.
2283
2284However, there is a situation in which it can be useful to violate this
2285assumption.
2286
e34caace 2287[[patch-series]]
4c63ff45
BF
2288Creating the perfect patch series
2289---------------------------------
2290
2291Suppose you are a contributor to a large project, and you want to add a
2292complicated feature, and to present it to the other developers in a way
2293that makes it easy for them to read your changes, verify that they are
2294correct, and understand why you made each change.
2295
b181d57f 2296If you present all of your changes as a single patch (or commit), they
79c96c57 2297may find that it is too much to digest all at once.
4c63ff45
BF
2298
2299If you present them with the entire history of your work, complete with
2300mistakes, corrections, and dead ends, they may be overwhelmed.
2301
2302So the ideal is usually to produce a series of patches such that:
2303
2304 1. Each patch can be applied in order.
2305
2306 2. Each patch includes a single logical change, together with a
2307 message explaining the change.
2308
2309 3. No patch introduces a regression: after applying any initial
2310 part of the series, the resulting project still compiles and
2311 works, and has no bugs that it didn't have before.
2312
2313 4. The complete series produces the same end result as your own
2314 (probably much messier!) development process did.
2315
b181d57f
BF
2316We will introduce some tools that can help you do this, explain how to
2317use them, and then explain some of the problems that can arise because
2318you are rewriting history.
4c63ff45 2319
e34caace 2320[[using-git-rebase]]
4c63ff45
BF
2321Keeping a patch series up to date using git-rebase
2322--------------------------------------------------
2323
79c96c57
MC
2324Suppose that you create a branch "mywork" on a remote-tracking branch
2325"origin", and create some commits on top of it:
4c63ff45
BF
2326
2327-------------------------------------------------
2328$ git checkout -b mywork origin
2329$ vi file.txt
2330$ git commit
2331$ vi otherfile.txt
2332$ git commit
2333...
2334-------------------------------------------------
2335
2336You have performed no merges into mywork, so it is just a simple linear
2337sequence of patches on top of "origin":
2338
1dc71a91 2339................................................
4c63ff45
BF
2340 o--o--o <-- origin
2341 \
2342 o--o--o <-- mywork
1dc71a91 2343................................................
4c63ff45
BF
2344
2345Some more interesting work has been done in the upstream project, and
2346"origin" has advanced:
2347
1dc71a91 2348................................................
4c63ff45
BF
2349 o--o--O--o--o--o <-- origin
2350 \
2351 a--b--c <-- mywork
1dc71a91 2352................................................
4c63ff45
BF
2353
2354At this point, you could use "pull" to merge your changes back in;
2355the result would create a new merge commit, like this:
2356
1dc71a91 2357................................................
4c63ff45
BF
2358 o--o--O--o--o--o <-- origin
2359 \ \
2360 a--b--c--m <-- mywork
1dc71a91 2361................................................
4c63ff45
BF
2362
2363However, if you prefer to keep the history in mywork a simple series of
2364commits without any merges, you may instead choose to use
2365gitlink:git-rebase[1]:
2366
2367-------------------------------------------------
2368$ git checkout mywork
2369$ git rebase origin
2370-------------------------------------------------
2371
b181d57f
BF
2372This will remove each of your commits from mywork, temporarily saving
2373them as patches (in a directory named ".dotest"), update mywork to
2374point at the latest version of origin, then apply each of the saved
2375patches to the new mywork. The result will look like:
4c63ff45
BF
2376
2377
1dc71a91 2378................................................
4c63ff45
BF
2379 o--o--O--o--o--o <-- origin
2380 \
2381 a'--b'--c' <-- mywork
1dc71a91 2382................................................
4c63ff45 2383
b181d57f
BF
2384In the process, it may discover conflicts. In that case it will stop
2385and allow you to fix the conflicts; after fixing conflicts, use "git
2386add" to update the index with those contents, and then, instead of
2387running git-commit, just run
4c63ff45
BF
2388
2389-------------------------------------------------
2390$ git rebase --continue
2391-------------------------------------------------
2392
2393and git will continue applying the rest of the patches.
2394
2395At any point you may use the --abort option to abort this process and
2396return mywork to the state it had before you started the rebase:
2397
2398-------------------------------------------------
2399$ git rebase --abort
2400-------------------------------------------------
2401
e34caace 2402[[modifying-one-commit]]
365aa199
BF
2403Modifying a single commit
2404-------------------------
2405
2406We saw in <<fixing-a-mistake-by-editing-history>> that you can replace the
2407most recent commit using
2408
2409-------------------------------------------------
2410$ git commit --amend
2411-------------------------------------------------
2412
2413which will replace the old commit by a new commit incorporating your
2414changes, giving you a chance to edit the old commit message first.
2415
2416You can also use a combination of this and gitlink:git-rebase[1] to edit
2417commits further back in your history. First, tag the problematic commit with
2418
2419-------------------------------------------------
2420$ git tag bad mywork~5
2421-------------------------------------------------
2422
2423(Either gitk or git-log may be useful for finding the commit.)
2424
25d9f3fa
BF
2425Then check out that commit, edit it, and rebase the rest of the series
2426on top of it (note that we could check out the commit on a temporary
2427branch, but instead we're using a <<detached-head,detached head>>):
365aa199
BF
2428
2429-------------------------------------------------
25d9f3fa 2430$ git checkout bad
365aa199
BF
2431$ # make changes here and update the index
2432$ git commit --amend
25d9f3fa 2433$ git rebase --onto HEAD bad mywork
365aa199
BF
2434-------------------------------------------------
2435
25d9f3fa
BF
2436When you're done, you'll be left with mywork checked out, with the top
2437patches on mywork reapplied on top of your modified commit. You can
365aa199
BF
2438then clean up with
2439
2440-------------------------------------------------
365aa199
BF
2441$ git tag -d bad
2442-------------------------------------------------
2443
2444Note that the immutable nature of git history means that you haven't really
2445"modified" existing commits; instead, you have replaced the old commits with
2446new commits having new object names.
2447
e34caace 2448[[reordering-patch-series]]
4c63ff45
BF
2449Reordering or selecting from a patch series
2450-------------------------------------------
2451
b181d57f
BF
2452Given one existing commit, the gitlink:git-cherry-pick[1] command
2453allows you to apply the change introduced by that commit and create a
2454new commit that records it. So, for example, if "mywork" points to a
2455series of patches on top of "origin", you might do something like:
2456
2457-------------------------------------------------
2458$ git checkout -b mywork-new origin
2459$ gitk origin..mywork &
2460-------------------------------------------------
2461
2462And browse through the list of patches in the mywork branch using gitk,
2463applying them (possibly in a different order) to mywork-new using
2464cherry-pick, and possibly modifying them as you go using commit
2465--amend.
2466
2467Another technique is to use git-format-patch to create a series of
2468patches, then reset the state to before the patches:
4c63ff45 2469
b181d57f
BF
2470-------------------------------------------------
2471$ git format-patch origin
2472$ git reset --hard origin
2473-------------------------------------------------
4c63ff45 2474
b181d57f
BF
2475Then modify, reorder, or eliminate patches as preferred before applying
2476them again with gitlink:git-am[1].
4c63ff45 2477
e34caace 2478[[patch-series-tools]]
4c63ff45
BF
2479Other tools
2480-----------
2481
b181d57f 2482There are numerous other tools, such as stgit, which exist for the
79c96c57 2483purpose of maintaining a patch series. These are outside of the scope of
b181d57f 2484this manual.
4c63ff45 2485
e34caace 2486[[problems-with-rewriting-history]]
4c63ff45
BF
2487Problems with rewriting history
2488-------------------------------
2489
b181d57f
BF
2490The primary problem with rewriting the history of a branch has to do
2491with merging. Suppose somebody fetches your branch and merges it into
2492their branch, with a result something like this:
2493
1dc71a91 2494................................................
b181d57f
BF
2495 o--o--O--o--o--o <-- origin
2496 \ \
2497 t--t--t--m <-- their branch:
1dc71a91 2498................................................
b181d57f
BF
2499
2500Then suppose you modify the last three commits:
2501
1dc71a91 2502................................................
b181d57f
BF
2503 o--o--o <-- new head of origin
2504 /
2505 o--o--O--o--o--o <-- old head of origin
1dc71a91 2506................................................
b181d57f
BF
2507
2508If we examined all this history together in one repository, it will
2509look like:
2510
1dc71a91 2511................................................
b181d57f
BF
2512 o--o--o <-- new head of origin
2513 /
2514 o--o--O--o--o--o <-- old head of origin
2515 \ \
2516 t--t--t--m <-- their branch:
1dc71a91 2517................................................
b181d57f
BF
2518
2519Git has no way of knowing that the new head is an updated version of
2520the old head; it treats this situation exactly the same as it would if
2521two developers had independently done the work on the old and new heads
2522in parallel. At this point, if someone attempts to merge the new head
2523in to their branch, git will attempt to merge together the two (old and
2524new) lines of development, instead of trying to replace the old by the
2525new. The results are likely to be unexpected.
2526
2527You may still choose to publish branches whose history is rewritten,
2528and it may be useful for others to be able to fetch those branches in
2529order to examine or test them, but they should not attempt to pull such
2530branches into their own work.
2531
2532For true distributed development that supports proper merging,
2533published branches should never be rewritten.
2534
e34caace 2535[[advanced-branch-management]]
b181d57f
BF
2536Advanced branch management
2537==========================
4c63ff45 2538
e34caace 2539[[fetching-individual-branches]]
b181d57f
BF
2540Fetching individual branches
2541----------------------------
2542
2543Instead of using gitlink:git-remote[1], you can also choose just
2544to update one branch at a time, and to store it locally under an
2545arbitrary name:
2546
2547-------------------------------------------------
2548$ git fetch origin todo:my-todo-work
2549-------------------------------------------------
2550
2551The first argument, "origin", just tells git to fetch from the
2552repository you originally cloned from. The second argument tells git
2553to fetch the branch named "todo" from the remote repository, and to
2554store it locally under the name refs/heads/my-todo-work.
2555
2556You can also fetch branches from other repositories; so
2557
2558-------------------------------------------------
2559$ git fetch git://example.com/proj.git master:example-master
2560-------------------------------------------------
2561
2562will create a new branch named "example-master" and store in it the
2563branch named "master" from the repository at the given URL. If you
2564already have a branch named example-master, it will attempt to
59723040
BF
2565<<fast-forwards,fast-forward>> to the commit given by example.com's
2566master branch. In more detail:
b181d57f 2567
59723040
BF
2568[[fetch-fast-forwards]]
2569git fetch and fast-forwards
2570---------------------------
b181d57f
BF
2571
2572In the previous example, when updating an existing branch, "git
2573fetch" checks to make sure that the most recent commit on the remote
2574branch is a descendant of the most recent commit on your copy of the
2575branch before updating your copy of the branch to point at the new
59723040 2576commit. Git calls this process a <<fast-forwards,fast forward>>.
b181d57f
BF
2577
2578A fast forward looks something like this:
2579
1dc71a91 2580................................................
b181d57f
BF
2581 o--o--o--o <-- old head of the branch
2582 \
2583 o--o--o <-- new head of the branch
1dc71a91 2584................................................
b181d57f
BF
2585
2586
2587In some cases it is possible that the new head will *not* actually be
2588a descendant of the old head. For example, the developer may have
2589realized she made a serious mistake, and decided to backtrack,
2590resulting in a situation like:
2591
1dc71a91 2592................................................
b181d57f
BF
2593 o--o--o--o--a--b <-- old head of the branch
2594 \
2595 o--o--o <-- new head of the branch
1dc71a91 2596................................................
b181d57f
BF
2597
2598In this case, "git fetch" will fail, and print out a warning.
2599
2600In that case, you can still force git to update to the new head, as
2601described in the following section. However, note that in the
2602situation above this may mean losing the commits labeled "a" and "b",
2603unless you've already created a reference of your own pointing to
2604them.
2605
e34caace 2606[[forcing-fetch]]
b181d57f
BF
2607Forcing git fetch to do non-fast-forward updates
2608------------------------------------------------
2609
2610If git fetch fails because the new head of a branch is not a
2611descendant of the old head, you may force the update with:
2612
2613-------------------------------------------------
2614$ git fetch git://example.com/proj.git +master:refs/remotes/example/master
2615-------------------------------------------------
2616
c64415e2
BF
2617Note the addition of the "+" sign. Alternatively, you can use the "-f"
2618flag to force updates of all the fetched branches, as in:
2619
2620-------------------------------------------------
2621$ git fetch -f origin
2622-------------------------------------------------
2623
2624Be aware that commits that the old version of example/master pointed at
2625may be lost, as we saw in the previous section.
b181d57f 2626
e34caace 2627[[remote-branch-configuration]]
b181d57f
BF
2628Configuring remote branches
2629---------------------------
2630
2631We saw above that "origin" is just a shortcut to refer to the
79c96c57 2632repository that you originally cloned from. This information is
b181d57f 2633stored in git configuration variables, which you can see using
9d13bda3 2634gitlink:git-config[1]:
b181d57f
BF
2635
2636-------------------------------------------------
9d13bda3 2637$ git config -l
b181d57f
BF
2638core.repositoryformatversion=0
2639core.filemode=true
2640core.logallrefupdates=true
2641remote.origin.url=git://git.kernel.org/pub/scm/git/git.git
2642remote.origin.fetch=+refs/heads/*:refs/remotes/origin/*
2643branch.master.remote=origin
2644branch.master.merge=refs/heads/master
2645-------------------------------------------------
2646
2647If there are other repositories that you also use frequently, you can
2648create similar configuration options to save typing; for example,
2649after
2650
2651-------------------------------------------------
9d13bda3 2652$ git config remote.example.url git://example.com/proj.git
b181d57f
BF
2653-------------------------------------------------
2654
2655then the following two commands will do the same thing:
2656
2657-------------------------------------------------
2658$ git fetch git://example.com/proj.git master:refs/remotes/example/master
2659$ git fetch example master:refs/remotes/example/master
2660-------------------------------------------------
2661
2662Even better, if you add one more option:
2663
2664-------------------------------------------------
9d13bda3 2665$ git config remote.example.fetch master:refs/remotes/example/master
b181d57f
BF
2666-------------------------------------------------
2667
2668then the following commands will all do the same thing:
2669
2670-------------------------------------------------
52c80037
BF
2671$ git fetch git://example.com/proj.git master:refs/remotes/example/master
2672$ git fetch example master:refs/remotes/example/master
b181d57f
BF
2673$ git fetch example
2674-------------------------------------------------
2675
2676You can also add a "+" to force the update each time:
2677
2678-------------------------------------------------
9d13bda3 2679$ git config remote.example.fetch +master:ref/remotes/example/master
b181d57f
BF
2680-------------------------------------------------
2681
2682Don't do this unless you're sure you won't mind "git fetch" possibly
2683throwing away commits on mybranch.
2684
2685Also note that all of the above configuration can be performed by
2686directly editing the file .git/config instead of using
9d13bda3 2687gitlink:git-config[1].
b181d57f 2688
9d13bda3 2689See gitlink:git-config[1] for more details on the configuration
b181d57f 2690options mentioned above.
d19fbc3c 2691
d19fbc3c 2692
35121930 2693[[git-internals]]
d19fbc3c
BF
2694Git internals
2695=============
2696
a536b08b
BF
2697Git depends on two fundamental abstractions: the "object database", and
2698the "current directory cache" aka "index".
b181d57f 2699
e34caace 2700[[the-object-database]]
b181d57f
BF
2701The Object Database
2702-------------------
2703
2704The object database is literally just a content-addressable collection
2705of objects. All objects are named by their content, which is
2706approximated by the SHA1 hash of the object itself. Objects may refer
2707to other objects (by referencing their SHA1 hash), and so you can
2708build up a hierarchy of objects.
2709
c64415e2 2710All objects have a statically determined "type" which is
b181d57f
BF
2711determined at object creation time, and which identifies the format of
2712the object (i.e. how it is used, and how it can refer to other
2713objects). There are currently four different object types: "blob",
a536b08b 2714"tree", "commit", and "tag".
b181d57f 2715
a536b08b
BF
2716A <<def_blob_object,"blob" object>> cannot refer to any other object,
2717and is, as the name implies, a pure storage object containing some
2718user data. It is used to actually store the file data, i.e. a blob
2719object is associated with some particular version of some file.
b181d57f 2720
a536b08b
BF
2721A <<def_tree_object,"tree" object>> is an object that ties one or more
2722"blob" objects into a directory structure. In addition, a tree object
2723can refer to other tree objects, thus creating a directory hierarchy.
b181d57f 2724
a536b08b
BF
2725A <<def_commit_object,"commit" object>> ties such directory hierarchies
2726together into a <<def_DAG,directed acyclic graph>> of revisions - each
2727"commit" is associated with exactly one tree (the directory hierarchy at
2728the time of the commit). In addition, a "commit" refers to one or more
2729"parent" commit objects that describe the history of how we arrived at
2730that directory hierarchy.
b181d57f
BF
2731
2732As a special case, a commit object with no parents is called the "root"
c64415e2 2733commit, and is the point of an initial project commit. Each project
b181d57f
BF
2734must have at least one root, and while you can tie several different
2735root objects together into one project by creating a commit object which
2736has two or more separate roots as its ultimate parents, that's probably
2737just going to confuse people. So aim for the notion of "one root object
2738per project", even if git itself does not enforce that.
2739
a536b08b
BF
2740A <<def_tag_object,"tag" object>> symbolically identifies and can be
2741used to sign other objects. It contains the identifier and type of
2742another object, a symbolic name (of course!) and, optionally, a
2743signature.
b181d57f
BF
2744
2745Regardless of object type, all objects share the following
2746characteristics: they are all deflated with zlib, and have a header
2747that not only specifies their type, but also provides size information
2748about the data in the object. It's worth noting that the SHA1 hash
2749that is used to name the object is the hash of the original data
2750plus this header, so `sha1sum` 'file' does not match the object name
2751for 'file'.
2752(Historical note: in the dawn of the age of git the hash
2753was the sha1 of the 'compressed' object.)
2754
2755As a result, the general consistency of an object can always be tested
2756independently of the contents or the type of the object: all objects can
2757be validated by verifying that (a) their hashes match the content of the
2758file and (b) the object successfully inflates to a stream of bytes that
2759forms a sequence of <ascii type without space> + <space> + <ascii decimal
2760size> + <byte\0> + <binary object data>.
2761
2762The structured objects can further have their structure and
2763connectivity to other objects verified. This is generally done with
04e50e94 2764the `git-fsck` program, which generates a full dependency graph
b181d57f
BF
2765of all objects, and verifies their internal consistency (in addition
2766to just verifying their superficial consistency through the hash).
2767
2768The object types in some more detail:
2769
e34caace 2770[[blob-object]]
b181d57f
BF
2771Blob Object
2772-----------
2773
2774A "blob" object is nothing but a binary blob of data, and doesn't
2775refer to anything else. There is no signature or any other
2776verification of the data, so while the object is consistent (it 'is'
2777indexed by its sha1 hash, so the data itself is certainly correct), it
2778has absolutely no other attributes. No name associations, no
2779permissions. It is purely a blob of data (i.e. normally "file
2780contents").
2781
2782In particular, since the blob is entirely defined by its data, if two
2783files in a directory tree (or in multiple different versions of the
2784repository) have the same contents, they will share the same blob
2785object. The object is totally independent of its location in the
2786directory tree, and renaming a file does not change the object that
2787file is associated with in any way.
2788
2789A blob is typically created when gitlink:git-update-index[1]
2790is run, and its data can be accessed by gitlink:git-cat-file[1].
2791
e34caace 2792[[tree-object]]
b181d57f
BF
2793Tree Object
2794-----------
2795
2796The next hierarchical object type is the "tree" object. A tree object
2797is a list of mode/name/blob data, sorted by name. Alternatively, the
2798mode data may specify a directory mode, in which case instead of
2799naming a blob, that name is associated with another TREE object.
2800
2801Like the "blob" object, a tree object is uniquely determined by the
2802set contents, and so two separate but identical trees will always
2803share the exact same object. This is true at all levels, i.e. it's
2804true for a "leaf" tree (which does not refer to any other trees, only
2805blobs) as well as for a whole subdirectory.
2806
2807For that reason a "tree" object is just a pure data abstraction: it
2808has no history, no signatures, no verification of validity, except
2809that since the contents are again protected by the hash itself, we can
2810trust that the tree is immutable and its contents never change.
2811
2812So you can trust the contents of a tree to be valid, the same way you
2813can trust the contents of a blob, but you don't know where those
2814contents 'came' from.
2815
2816Side note on trees: since a "tree" object is a sorted list of
2817"filename+content", you can create a diff between two trees without
2818actually having to unpack two trees. Just ignore all common parts,
2819and your diff will look right. In other words, you can effectively
2820(and efficiently) tell the difference between any two random trees by
2821O(n) where "n" is the size of the difference, rather than the size of
2822the tree.
2823
2824Side note 2 on trees: since the name of a "blob" depends entirely and
2825exclusively on its contents (i.e. there are no names or permissions
2826involved), you can see trivial renames or permission changes by
2827noticing that the blob stayed the same. However, renames with data
2828changes need a smarter "diff" implementation.
2829
2830A tree is created with gitlink:git-write-tree[1] and
2831its data can be accessed by gitlink:git-ls-tree[1].
2832Two trees can be compared with gitlink:git-diff-tree[1].
2833
e34caace 2834[[commit-object]]
b181d57f
BF
2835Commit Object
2836-------------
2837
2838The "commit" object is an object that introduces the notion of
2839history into the picture. In contrast to the other objects, it
2840doesn't just describe the physical state of a tree, it describes how
2841we got there, and why.
2842
2843A "commit" is defined by the tree-object that it results in, the
2844parent commits (zero, one or more) that led up to that point, and a
2845comment on what happened. Again, a commit is not trusted per se:
2846the contents are well-defined and "safe" due to the cryptographically
2847strong signatures at all levels, but there is no reason to believe
2848that the tree is "good" or that the merge information makes sense.
2849The parents do not have to actually have any relationship with the
2850result, for example.
2851
c64415e2 2852Note on commits: unlike some SCM's, commits do not contain
b181d57f
BF
2853rename information or file mode change information. All of that is
2854implicit in the trees involved (the result tree, and the result trees
2855of the parents), and describing that makes no sense in this idiotic
2856file manager.
2857
2858A commit is created with gitlink:git-commit-tree[1] and
2859its data can be accessed by gitlink:git-cat-file[1].
2860
e34caace 2861[[trust]]
b181d57f
BF
2862Trust
2863-----
2864
2865An aside on the notion of "trust". Trust is really outside the scope
2866of "git", but it's worth noting a few things. First off, since
2867everything is hashed with SHA1, you 'can' trust that an object is
2868intact and has not been messed with by external sources. So the name
2869of an object uniquely identifies a known state - just not a state that
2870you may want to trust.
2871
2872Furthermore, since the SHA1 signature of a commit refers to the
2873SHA1 signatures of the tree it is associated with and the signatures
2874of the parent, a single named commit specifies uniquely a whole set
2875of history, with full contents. You can't later fake any step of the
2876way once you have the name of a commit.
2877
2878So to introduce some real trust in the system, the only thing you need
2879to do is to digitally sign just 'one' special note, which includes the
2880name of a top-level commit. Your digital signature shows others
2881that you trust that commit, and the immutability of the history of
2882commits tells others that they can trust the whole history.
2883
2884In other words, you can easily validate a whole archive by just
2885sending out a single email that tells the people the name (SHA1 hash)
2886of the top commit, and digitally sign that email using something
2887like GPG/PGP.
2888
2889To assist in this, git also provides the tag object...
2890
e34caace 2891[[tag-object]]
b181d57f
BF
2892Tag Object
2893----------
2894
2895Git provides the "tag" object to simplify creating, managing and
2896exchanging symbolic and signed tokens. The "tag" object at its
2897simplest simply symbolically identifies another object by containing
2898the sha1, type and symbolic name.
2899
2900However it can optionally contain additional signature information
2901(which git doesn't care about as long as there's less than 8k of
2902it). This can then be verified externally to git.
2903
2904Note that despite the tag features, "git" itself only handles content
2905integrity; the trust framework (and signature provision and
2906verification) has to come from outside.
2907
2908A tag is created with gitlink:git-mktag[1],
2909its data can be accessed by gitlink:git-cat-file[1],
2910and the signature can be verified by
2911gitlink:git-verify-tag[1].
2912
2913
e34caace 2914[[the-index]]
b181d57f
BF
2915The "index" aka "Current Directory Cache"
2916-----------------------------------------
2917
2918The index is a simple binary file, which contains an efficient
c64415e2 2919representation of the contents of a virtual directory. It
b181d57f
BF
2920does so by a simple array that associates a set of names, dates,
2921permissions and content (aka "blob") objects together. The cache is
2922always kept ordered by name, and names are unique (with a few very
2923specific rules) at any point in time, but the cache has no long-term
2924meaning, and can be partially updated at any time.
2925
2926In particular, the index certainly does not need to be consistent with
2927the current directory contents (in fact, most operations will depend on
2928different ways to make the index 'not' be consistent with the directory
2929hierarchy), but it has three very important attributes:
2930
2931'(a) it can re-generate the full state it caches (not just the
2932directory structure: it contains pointers to the "blob" objects so
2933that it can regenerate the data too)'
2934
2935As a special case, there is a clear and unambiguous one-way mapping
2936from a current directory cache to a "tree object", which can be
2937efficiently created from just the current directory cache without
2938actually looking at any other data. So a directory cache at any one
2939time uniquely specifies one and only one "tree" object (but has
2940additional data to make it easy to match up that tree object with what
2941has happened in the directory)
2942
2943'(b) it has efficient methods for finding inconsistencies between that
2944cached state ("tree object waiting to be instantiated") and the
2945current state.'
2946
2947'(c) it can additionally efficiently represent information about merge
2948conflicts between different tree objects, allowing each pathname to be
2949associated with sufficient information about the trees involved that
2950you can create a three-way merge between them.'
2951
79c96c57 2952Those are the ONLY three things that the directory cache does. It's a
b181d57f
BF
2953cache, and the normal operation is to re-generate it completely from a
2954known tree object, or update/compare it with a live tree that is being
2955developed. If you blow the directory cache away entirely, you generally
2956haven't lost any information as long as you have the name of the tree
2957that it described.
2958
2959At the same time, the index is at the same time also the
2960staging area for creating new trees, and creating a new tree always
2961involves a controlled modification of the index file. In particular,
2962the index file can have the representation of an intermediate tree that
2963has not yet been instantiated. So the index can be thought of as a
2964write-back cache, which can contain dirty information that has not yet
2965been written back to the backing store.
2966
2967
2968
e34caace 2969[[the-workflow]]
b181d57f
BF
2970The Workflow
2971------------
2972
2973Generally, all "git" operations work on the index file. Some operations
2974work *purely* on the index file (showing the current state of the
2975index), but most operations move data to and from the index file. Either
2976from the database or from the working directory. Thus there are four
2977main combinations:
2978
e34caace 2979[[working-directory-to-index]]
b181d57f
BF
2980working directory -> index
2981~~~~~~~~~~~~~~~~~~~~~~~~~~
2982
2983You update the index with information from the working directory with
2984the gitlink:git-update-index[1] command. You
2985generally update the index information by just specifying the filename
2986you want to update, like so:
2987
2988-------------------------------------------------
2989$ git-update-index filename
2990-------------------------------------------------
2991
2992but to avoid common mistakes with filename globbing etc, the command
2993will not normally add totally new entries or remove old entries,
2994i.e. it will normally just update existing cache entries.
2995
2996To tell git that yes, you really do realize that certain files no
2997longer exist, or that new files should be added, you
2998should use the `--remove` and `--add` flags respectively.
2999
3000NOTE! A `--remove` flag does 'not' mean that subsequent filenames will
3001necessarily be removed: if the files still exist in your directory
3002structure, the index will be updated with their new status, not
3003removed. The only thing `--remove` means is that update-cache will be
3004considering a removed file to be a valid thing, and if the file really
3005does not exist any more, it will update the index accordingly.
3006
3007As a special case, you can also do `git-update-index --refresh`, which
3008will refresh the "stat" information of each index to match the current
3009stat information. It will 'not' update the object status itself, and
3010it will only update the fields that are used to quickly test whether
3011an object still matches its old backing store object.
3012
e34caace 3013[[index-to-object-database]]
b181d57f
BF
3014index -> object database
3015~~~~~~~~~~~~~~~~~~~~~~~~
3016
3017You write your current index file to a "tree" object with the program
3018
3019-------------------------------------------------
3020$ git-write-tree
3021-------------------------------------------------
3022
3023that doesn't come with any options - it will just write out the
3024current index into the set of tree objects that describe that state,
3025and it will return the name of the resulting top-level tree. You can
3026use that tree to re-generate the index at any time by going in the
3027other direction:
3028
e34caace 3029[[object-database-to-index]]
b181d57f
BF
3030object database -> index
3031~~~~~~~~~~~~~~~~~~~~~~~~
3032
3033You read a "tree" file from the object database, and use that to
3034populate (and overwrite - don't do this if your index contains any
3035unsaved state that you might want to restore later!) your current
3036index. Normal operation is just
3037
3038-------------------------------------------------
3039$ git-read-tree <sha1 of tree>
3040-------------------------------------------------
3041
3042and your index file will now be equivalent to the tree that you saved
3043earlier. However, that is only your 'index' file: your working
3044directory contents have not been modified.
3045
e34caace 3046[[index-to-working-directory]]
b181d57f
BF
3047index -> working directory
3048~~~~~~~~~~~~~~~~~~~~~~~~~~
3049
3050You update your working directory from the index by "checking out"
3051files. This is not a very common operation, since normally you'd just
3052keep your files updated, and rather than write to your working
3053directory, you'd tell the index files about the changes in your
3054working directory (i.e. `git-update-index`).
3055
3056However, if you decide to jump to a new version, or check out somebody
3057else's version, or just restore a previous tree, you'd populate your
3058index file with read-tree, and then you need to check out the result
3059with
3060
3061-------------------------------------------------
3062$ git-checkout-index filename
3063-------------------------------------------------
3064
3065or, if you want to check out all of the index, use `-a`.
3066
3067NOTE! git-checkout-index normally refuses to overwrite old files, so
3068if you have an old version of the tree already checked out, you will
3069need to use the "-f" flag ('before' the "-a" flag or the filename) to
3070'force' the checkout.
3071
3072
3073Finally, there are a few odds and ends which are not purely moving
3074from one representation to the other:
3075
e34caace 3076[[tying-it-all-together]]
b181d57f
BF
3077Tying it all together
3078~~~~~~~~~~~~~~~~~~~~~
3079
3080To commit a tree you have instantiated with "git-write-tree", you'd
3081create a "commit" object that refers to that tree and the history
3082behind it - most notably the "parent" commits that preceded it in
3083history.
3084
3085Normally a "commit" has one parent: the previous state of the tree
3086before a certain change was made. However, sometimes it can have two
3087or more parent commits, in which case we call it a "merge", due to the
3088fact that such a commit brings together ("merges") two or more
3089previous states represented by other commits.
3090
3091In other words, while a "tree" represents a particular directory state
3092of a working directory, a "commit" represents that state in "time",
3093and explains how we got there.
3094
3095You create a commit object by giving it the tree that describes the
3096state at the time of the commit, and a list of parents:
3097
3098-------------------------------------------------
3099$ git-commit-tree <tree> -p <parent> [-p <parent2> ..]
3100-------------------------------------------------
3101
3102and then giving the reason for the commit on stdin (either through
3103redirection from a pipe or file, or by just typing it at the tty).
3104
3105git-commit-tree will return the name of the object that represents
3106that commit, and you should save it away for later use. Normally,
3107you'd commit a new `HEAD` state, and while git doesn't care where you
3108save the note about that state, in practice we tend to just write the
3109result to the file pointed at by `.git/HEAD`, so that we can always see
3110what the last committed state was.
3111
3112Here is an ASCII art by Jon Loeliger that illustrates how
3113various pieces fit together.
3114
3115------------
3116
3117 commit-tree
3118 commit obj
3119 +----+
3120 | |
3121 | |
3122 V V
3123 +-----------+
3124 | Object DB |
3125 | Backing |
3126 | Store |
3127 +-----------+
3128 ^
3129 write-tree | |
3130 tree obj | |
3131 | | read-tree
3132 | | tree obj
3133 V
3134 +-----------+
3135 | Index |
3136 | "cache" |
3137 +-----------+
3138 update-index ^
3139 blob obj | |
3140 | |
3141 checkout-index -u | | checkout-index
3142 stat | | blob obj
3143 V
3144 +-----------+
3145 | Working |
3146 | Directory |
3147 +-----------+
3148
3149------------
3150
3151
e34caace 3152[[examining-the-data]]
b181d57f
BF
3153Examining the data
3154------------------
3155
3156You can examine the data represented in the object database and the
3157index with various helper tools. For every object, you can use
3158gitlink:git-cat-file[1] to examine details about the
3159object:
3160
3161-------------------------------------------------
3162$ git-cat-file -t <objectname>
3163-------------------------------------------------
3164
3165shows the type of the object, and once you have the type (which is
3166usually implicit in where you find the object), you can use
3167
3168-------------------------------------------------
3169$ git-cat-file blob|tree|commit|tag <objectname>
3170-------------------------------------------------
3171
3172to show its contents. NOTE! Trees have binary content, and as a result
3173there is a special helper for showing that content, called
3174`git-ls-tree`, which turns the binary content into a more easily
3175readable form.
3176
3177It's especially instructive to look at "commit" objects, since those
3178tend to be small and fairly self-explanatory. In particular, if you
3179follow the convention of having the top commit name in `.git/HEAD`,
3180you can do
3181
3182-------------------------------------------------
3183$ git-cat-file commit HEAD
3184-------------------------------------------------
3185
3186to see what the top commit was.
3187
e34caace 3188[[merging-multiple-trees]]
b181d57f 3189Merging multiple trees
d19fbc3c
BF
3190----------------------
3191
b181d57f
BF
3192Git helps you do a three-way merge, which you can expand to n-way by
3193repeating the merge procedure arbitrary times until you finally
3194"commit" the state. The normal situation is that you'd only do one
3195three-way merge (two parents), and commit it, but if you like to, you
3196can do multiple parents in one go.
3197
3198To do a three-way merge, you need the two sets of "commit" objects
3199that you want to merge, use those to find the closest common parent (a
3200third "commit" object), and then use those commit objects to find the
3201state of the directory ("tree" object) at these points.
3202
3203To get the "base" for the merge, you first look up the common parent
3204of two commits with
3205
3206-------------------------------------------------
3207$ git-merge-base <commit1> <commit2>
3208-------------------------------------------------
3209
3210which will return you the commit they are both based on. You should
3211now look up the "tree" objects of those commits, which you can easily
3212do with (for example)
3213
3214-------------------------------------------------
3215$ git-cat-file commit <commitname> | head -1
3216-------------------------------------------------
3217
3218since the tree object information is always the first line in a commit
3219object.
3220
1191ee18 3221Once you know the three trees you are going to merge (the one "original"
c64415e2 3222tree, aka the common tree, and the two "result" trees, aka the branches
1191ee18
BF
3223you want to merge), you do a "merge" read into the index. This will
3224complain if it has to throw away your old index contents, so you should
b181d57f 3225make sure that you've committed those - in fact you would normally
1191ee18
BF
3226always do a merge against your last commit (which should thus match what
3227you have in your current index anyway).
b181d57f
BF
3228
3229To do the merge, do
3230
3231-------------------------------------------------
3232$ git-read-tree -m -u <origtree> <yourtree> <targettree>
3233-------------------------------------------------
3234
3235which will do all trivial merge operations for you directly in the
3236index file, and you can just write the result out with
3237`git-write-tree`.
3238
3239
e34caace 3240[[merging-multiple-trees-2]]
b181d57f
BF
3241Merging multiple trees, continued
3242---------------------------------
3243
3244Sadly, many merges aren't trivial. If there are files that have
3245been added.moved or removed, or if both branches have modified the
3246same file, you will be left with an index tree that contains "merge
3247entries" in it. Such an index tree can 'NOT' be written out to a tree
3248object, and you will have to resolve any such merge clashes using
3249other tools before you can write out the result.
3250
3251You can examine such index state with `git-ls-files --unmerged`
3252command. An example:
3253
3254------------------------------------------------
3255$ git-read-tree -m $orig HEAD $target
3256$ git-ls-files --unmerged
3257100644 263414f423d0e4d70dae8fe53fa34614ff3e2860 1 hello.c
3258100644 06fa6a24256dc7e560efa5687fa84b51f0263c3a 2 hello.c
3259100644 cc44c73eb783565da5831b4d820c962954019b69 3 hello.c
3260------------------------------------------------
3261
3262Each line of the `git-ls-files --unmerged` output begins with
3263the blob mode bits, blob SHA1, 'stage number', and the
3264filename. The 'stage number' is git's way to say which tree it
3265came from: stage 1 corresponds to `$orig` tree, stage 2 `HEAD`
3266tree, and stage3 `$target` tree.
3267
3268Earlier we said that trivial merges are done inside
3269`git-read-tree -m`. For example, if the file did not change
3270from `$orig` to `HEAD` nor `$target`, or if the file changed
3271from `$orig` to `HEAD` and `$orig` to `$target` the same way,
3272obviously the final outcome is what is in `HEAD`. What the
3273above example shows is that file `hello.c` was changed from
3274`$orig` to `HEAD` and `$orig` to `$target` in a different way.
3275You could resolve this by running your favorite 3-way merge
c64415e2
BF
3276program, e.g. `diff3`, `merge`, or git's own merge-file, on
3277the blob objects from these three stages yourself, like this:
b181d57f
BF
3278
3279------------------------------------------------
3280$ git-cat-file blob 263414f... >hello.c~1
3281$ git-cat-file blob 06fa6a2... >hello.c~2
3282$ git-cat-file blob cc44c73... >hello.c~3
c64415e2 3283$ git merge-file hello.c~2 hello.c~1 hello.c~3
b181d57f
BF
3284------------------------------------------------
3285
3286This would leave the merge result in `hello.c~2` file, along
3287with conflict markers if there are conflicts. After verifying
3288the merge result makes sense, you can tell git what the final
3289merge result for this file is by:
3290
3291-------------------------------------------------
3292$ mv -f hello.c~2 hello.c
3293$ git-update-index hello.c
3294-------------------------------------------------
3295
3296When a path is in unmerged state, running `git-update-index` for
3297that path tells git to mark the path resolved.
3298
3299The above is the description of a git merge at the lowest level,
3300to help you understand what conceptually happens under the hood.
3301In practice, nobody, not even git itself, uses three `git-cat-file`
3302for this. There is `git-merge-index` program that extracts the
3303stages to temporary files and calls a "merge" script on it:
3304
3305-------------------------------------------------
3306$ git-merge-index git-merge-one-file hello.c
3307-------------------------------------------------
3308
207dfa07 3309and that is what higher level `git merge -s resolve` is implemented with.
b181d57f 3310
e34caace 3311[[pack-files]]
b181d57f
BF
3312How git stores objects efficiently: pack files
3313----------------------------------------------
3314
3315We've seen how git stores each object in a file named after the
3316object's SHA1 hash.
3317
3318Unfortunately this system becomes inefficient once a project has a
3319lot of objects. Try this on an old project:
3320
3321------------------------------------------------
3322$ git count-objects
33236930 objects, 47620 kilobytes
3324------------------------------------------------
3325
3326The first number is the number of objects which are kept in
3327individual files. The second is the amount of space taken up by
3328those "loose" objects.
3329
3330You can save space and make git faster by moving these loose objects in
3331to a "pack file", which stores a group of objects in an efficient
3332compressed format; the details of how pack files are formatted can be
3333found in link:technical/pack-format.txt[technical/pack-format.txt].
3334
3335To put the loose objects into a pack, just run git repack:
3336
3337------------------------------------------------
3338$ git repack
3339Generating pack...
3340Done counting 6020 objects.
3341Deltifying 6020 objects.
3342 100% (6020/6020) done
3343Writing 6020 objects.
3344 100% (6020/6020) done
3345Total 6020, written 6020 (delta 4070), reused 0 (delta 0)
3346Pack pack-3e54ad29d5b2e05838c75df582c65257b8d08e1c created.
3347------------------------------------------------
3348
3349You can then run
3350
3351------------------------------------------------
3352$ git prune
3353------------------------------------------------
3354
3355to remove any of the "loose" objects that are now contained in the
3356pack. This will also remove any unreferenced objects (which may be
3357created when, for example, you use "git reset" to remove a commit).
3358You can verify that the loose objects are gone by looking at the
3359.git/objects directory or by running
3360
3361------------------------------------------------
3362$ git count-objects
33630 objects, 0 kilobytes
3364------------------------------------------------
3365
3366Although the object files are gone, any commands that refer to those
3367objects will work exactly as they did before.
3368
3369The gitlink:git-gc[1] command performs packing, pruning, and more for
3370you, so is normally the only high-level command you need.
d19fbc3c 3371
59723040 3372[[dangling-objects]]
21dcb3b7 3373Dangling objects
61b41790 3374----------------
21dcb3b7 3375
04e50e94 3376The gitlink:git-fsck[1] command will sometimes complain about dangling
21dcb3b7
BF
3377objects. They are not a problem.
3378
1191ee18
BF
3379The most common cause of dangling objects is that you've rebased a
3380branch, or you have pulled from somebody else who rebased a branch--see
3381<<cleaning-up-history>>. In that case, the old head of the original
59723040
BF
3382branch still exists, as does everything it pointed to. The branch
3383pointer itself just doesn't, since you replaced it with another one.
1191ee18 3384
59723040 3385There are also other situations that cause dangling objects. For
1191ee18
BF
3386example, a "dangling blob" may arise because you did a "git add" of a
3387file, but then, before you actually committed it and made it part of the
3388bigger picture, you changed something else in that file and committed
3389that *updated* thing - the old state that you added originally ends up
3390not being pointed to by any commit or tree, so it's now a dangling blob
3391object.
3392
3393Similarly, when the "recursive" merge strategy runs, and finds that
3394there are criss-cross merges and thus more than one merge base (which is
3395fairly unusual, but it does happen), it will generate one temporary
3396midway tree (or possibly even more, if you had lots of criss-crossing
3397merges and more than two merge bases) as a temporary internal merge
3398base, and again, those are real objects, but the end result will not end
3399up pointing to them, so they end up "dangling" in your repository.
3400
3401Generally, dangling objects aren't anything to worry about. They can
3402even be very useful: if you screw something up, the dangling objects can
3403be how you recover your old tree (say, you did a rebase, and realized
3404that you really didn't want to - you can look at what dangling objects
3405you have, and decide to reset your head to some old dangling state).
21dcb3b7 3406
59723040 3407For commits, you can just use:
21dcb3b7
BF
3408
3409------------------------------------------------
3410$ gitk <dangling-commit-sha-goes-here> --not --all
3411------------------------------------------------
3412
59723040
BF
3413This asks for all the history reachable from the given commit but not
3414from any branch, tag, or other reference. If you decide it's something
3415you want, you can always create a new reference to it, e.g.,
3416
3417------------------------------------------------
3418$ git branch recovered-branch <dangling-commit-sha-goes-here>
3419------------------------------------------------
3420
3421For blobs and trees, you can't do the same, but you can still examine
3422them. You can just do
21dcb3b7
BF
3423
3424------------------------------------------------
3425$ git show <dangling-blob/tree-sha-goes-here>
3426------------------------------------------------
3427
1191ee18
BF
3428to show what the contents of the blob were (or, for a tree, basically
3429what the "ls" for that directory was), and that may give you some idea
3430of what the operation was that left that dangling object.
21dcb3b7 3431
1191ee18
BF
3432Usually, dangling blobs and trees aren't very interesting. They're
3433almost always the result of either being a half-way mergebase (the blob
3434will often even have the conflict markers from a merge in it, if you
3435have had conflicting merges that you fixed up by hand), or simply
3436because you interrupted a "git fetch" with ^C or something like that,
3437leaving _some_ of the new objects in the object database, but just
3438dangling and useless.
21dcb3b7
BF
3439
3440Anyway, once you are sure that you're not interested in any dangling
3441state, you can just prune all unreachable objects:
3442
3443------------------------------------------------
3444$ git prune
3445------------------------------------------------
3446
1191ee18
BF
3447and they'll be gone. But you should only run "git prune" on a quiescent
3448repository - it's kind of like doing a filesystem fsck recovery: you
3449don't want to do that while the filesystem is mounted.
21dcb3b7 3450
04e50e94
BF
3451(The same is true of "git-fsck" itself, btw - but since
3452git-fsck never actually *changes* the repository, it just reports
3453on what it found, git-fsck itself is never "dangerous" to run.
21dcb3b7
BF
3454Running it while somebody is actually changing the repository can cause
3455confusing and scary messages, but it won't actually do anything bad. In
3456contrast, running "git prune" while somebody is actively changing the
3457repository is a *BAD* idea).
3458
126640af 3459[[birdview-on-the-source-code]]
a5fc33b4
BF
3460A birds-eye view of Git's source code
3461-------------------------------------
126640af 3462
a5fc33b4
BF
3463It is not always easy for new developers to find their way through Git's
3464source code. This section gives you a little guidance to show where to
3465start.
126640af 3466
a5fc33b4 3467A good place to start is with the contents of the initial commit, with:
126640af
JS
3468
3469----------------------------------------------------
a5fc33b4 3470$ git checkout e83c5163
126640af
JS
3471----------------------------------------------------
3472
a5fc33b4
BF
3473The initial revision lays the foundation for almost everything git has
3474today, but is small enough to read in one sitting.
126640af 3475
a5fc33b4
BF
3476Note that terminology has changed since that revision. For example, the
3477README in that revision uses the word "changeset" to describe what we
3478now call a <<def_commit_object,commit>>.
126640af 3479
a5fc33b4 3480Also, we do not call it "cache" any more, but "index", however, the
126640af
JS
3481file is still called `cache.h`. Remark: Not much reason to change it now,
3482especially since there is no good single name for it anyway, because it is
3483basically _the_ header file which is included by _all_ of Git's C sources.
3484
a5fc33b4
BF
3485If you grasp the ideas in that initial commit, you should check out a
3486more recent version and skim `cache.h`, `object.h` and `commit.h`.
126640af
JS
3487
3488In the early days, Git (in the tradition of UNIX) was a bunch of programs
3489which were extremely simple, and which you used in scripts, piping the
3490output of one into another. This turned out to be good for initial
3491development, since it was easier to test new things. However, recently
3492many of these parts have become builtins, and some of the core has been
3493"libified", i.e. put into libgit.a for performance, portability reasons,
3494and to avoid code duplication.
3495
3496By now, you know what the index is (and find the corresponding data
3497structures in `cache.h`), and that there are just a couple of object types
3498(blobs, trees, commits and tags) which inherit their common structure from
3499`struct object`, which is their first member (and thus, you can cast e.g.
3500`(struct object *)commit` to achieve the _same_ as `&commit->object`, i.e.
3501get at the object name and flags).
3502
3503Now is a good point to take a break to let this information sink in.
3504
3505Next step: get familiar with the object naming. Read <<naming-commits>>.
3506There are quite a few ways to name an object (and not only revisions!).
3507All of these are handled in `sha1_name.c`. Just have a quick look at
3508the function `get_sha1()`. A lot of the special handling is done by
3509functions like `get_sha1_basic()` or the likes.
3510
3511This is just to get you into the groove for the most libified part of Git:
3512the revision walker.
3513
3514Basically, the initial version of `git log` was a shell script:
3515
3516----------------------------------------------------------------
3517$ git-rev-list --pretty $(git-rev-parse --default HEAD "$@") | \
3518 LESS=-S ${PAGER:-less}
3519----------------------------------------------------------------
3520
3521What does this mean?
3522
3523`git-rev-list` is the original version of the revision walker, which
3524_always_ printed a list of revisions to stdout. It is still functional,
3525and needs to, since most new Git programs start out as scripts using
3526`git-rev-list`.
3527
3528`git-rev-parse` is not as important any more; it was only used to filter out
3529options that were relevant for the different plumbing commands that were
3530called by the script.
3531
3532Most of what `git-rev-list` did is contained in `revision.c` and
3533`revision.h`. It wraps the options in a struct named `rev_info`, which
3534controls how and what revisions are walked, and more.
3535
3536The original job of `git-rev-parse` is now taken by the function
3537`setup_revisions()`, which parses the revisions and the common command line
3538options for the revision walker. This information is stored in the struct
3539`rev_info` for later consumption. You can do your own command line option
3540parsing after calling `setup_revisions()`. After that, you have to call
3541`prepare_revision_walk()` for initialization, and then you can get the
3542commits one by one with the function `get_revision()`.
3543
3544If you are interested in more details of the revision walking process,
3545just have a look at the first implementation of `cmd_log()`; call
3546`git-show v1.3.0~155^2~4` and scroll down to that function (note that you
3547no longer need to call `setup_pager()` directly).
3548
3549Nowadays, `git log` is a builtin, which means that it is _contained_ in the
3550command `git`. The source side of a builtin is
3551
3552- a function called `cmd_<bla>`, typically defined in `builtin-<bla>.c`,
3553 and declared in `builtin.h`,
3554
3555- an entry in the `commands[]` array in `git.c`, and
3556
3557- an entry in `BUILTIN_OBJECTS` in the `Makefile`.
3558
3559Sometimes, more than one builtin is contained in one source file. For
3560example, `cmd_whatchanged()` and `cmd_log()` both reside in `builtin-log.c`,
3561since they share quite a bit of code. In that case, the commands which are
3562_not_ named like the `.c` file in which they live have to be listed in
3563`BUILT_INS` in the `Makefile`.
3564
3565`git log` looks more complicated in C than it does in the original script,
3566but that allows for a much greater flexibility and performance.
3567
3568