1 A tutorial introduction to git: part two
2 ========================================
4 You should work through link:tutorial.html[A tutorial introduction to
5 git] before reading this tutorial.
7 The goal of this tutorial is to introduce two fundamental pieces of
8 git's architecture--the object database and the index file--and to
9 provide the reader with everything necessary to understand the rest
10 of the git documentation.
12 The git object database
13 -----------------------
15 Let's start a new project and create a small amount of history:
17 ------------------------------------------------
21 defaulting to local storage area
22 $ echo 'hello world' > file.txt
24 $ git commit -a -m "initial commit"
25 Committing initial tree 92b8b694ffb1675e5975148e1121810081dbdffe
26 $ echo 'hello world!' >file.txt
27 $ git commit -a -m "add emphasis"
28 ------------------------------------------------
30 What are the 40 digits of hex that git responded to the first commit
33 We saw in part one of the tutorial that commits have names like this.
34 It turns out that every object in the git history is stored under
35 such a 40-digit hex name. That name is the SHA1 hash of the object's
36 contents; among other things, this ensures that git will never store
37 the same data twice (since identical data is given an identical SHA1
38 name), and that the contents of a git object will never change (since
39 that would change the object's name as well).
41 We can ask git about this particular object with the cat-file
42 command--just cut-and-paste from the reply to the initial commit, to
43 save yourself typing all 40 hex digits:
45 ------------------------------------------------
46 $ git cat-file -t 92b8b694ffb1675e5975148e1121810081dbdffe
48 ------------------------------------------------
50 A tree can refer to one or more "blob" objects, each corresponding to
51 a file. In addition, a tree can also refer to other tree objects,
52 thus creating a directory hierarchy. You can examine the contents of
53 any tree using ls-tree (remember that a long enough initial portion
54 of the SHA1 will also work):
56 ------------------------------------------------
57 $ git ls-tree 92b8b694
58 100644 blob 3b18e512dba79e4c8300dd08aeb37f8e728b8dad file.txt
59 ------------------------------------------------
61 Thus we see that this tree has one file in it. The SHA1 hash is a
62 reference to that file's data:
64 ------------------------------------------------
65 $ git cat-file -t 3b18e512
67 ------------------------------------------------
69 A "blob" is just file data, which we can also examine with cat-file:
71 ------------------------------------------------
72 $ git cat-file blob 3b18e512
74 ------------------------------------------------
76 Note that this is the old file data; so the object that git named in
77 its response to the initial tree was a tree with a snapshot of the
78 directory state that was recorded by the first commit.
80 All of these objects are stored under their SHA1 names inside the git
83 ------------------------------------------------
89 .git/objects/3b/18e512dba79e4c8300dd08aeb37f8e728b8dad
91 .git/objects/92/b8b694ffb1675e5975148e1121810081dbdffe
93 .git/objects/54/196cc2703dc165cbd373a65a4dcf22d50ae7f7
95 .git/objects/a0/423896973644771497bdc03eb99d5281615b51
97 .git/objects/d0/492b368b66bdabf2ac1fd8c92b39d3db916e59
99 .git/objects/c4/d59f390b9cfd4318117afde11d601c1085f241
100 ------------------------------------------------
102 and the contents of these files is just the compressed data plus a
103 header identifying their length and their type. The type is either a
104 blob, a tree, a commit, or a tag. We've seen a blob and a tree now,
105 so next we should look at a commit.
107 The simplest commit to find is the HEAD commit, which we can find
110 ------------------------------------------------
112 ref: refs/heads/master
113 ------------------------------------------------
115 As you can see, this tells us which branch we're currently on, and it
116 tells us this by naming a file under the .git directory, which itself
117 contains a SHA1 name referring to a commit object, which we can
118 examine with cat-file:
120 ------------------------------------------------
121 $ cat .git/refs/heads/master
122 c4d59f390b9cfd4318117afde11d601c1085f241
123 $ git cat-file -t c4d59f39
125 $ git cat-file commit c4d59f39
126 tree d0492b368b66bdabf2ac1fd8c92b39d3db916e59
127 parent 54196cc2703dc165cbd373a65a4dcf22d50ae7f7
128 author J. Bruce Fields <bfields@puzzle.fieldses.org> 1143418702 -0500
129 committer J. Bruce Fields <bfields@puzzle.fieldses.org> 1143418702 -0500
132 ------------------------------------------------
134 The "tree" object here refers to the new state of the tree:
136 ------------------------------------------------
137 $ git ls-tree d0492b36
138 100644 blob a0423896973644771497bdc03eb99d5281615b51 file.txt
139 $ git cat-file blob a0423896
141 ------------------------------------------------
143 and the "parent" object refers to the previous commit:
145 ------------------------------------------------
146 $ git-cat-file commit 54196cc2
147 tree 92b8b694ffb1675e5975148e1121810081dbdffe
148 author J. Bruce Fields <bfields@puzzle.fieldses.org> 1143414668 -0500
149 committer J. Bruce Fields <bfields@puzzle.fieldses.org> 1143414668 -0500
152 ------------------------------------------------
154 The tree object is the tree we examined first, and this commit is
155 unusual in that it lacks any parent.
157 Most commits have only one parent, but it is also common for a commit
158 to have multiple parents. In that case the commit represents a
159 merge, with the parent references pointing to the heads of the merged
162 Besides blobs, trees, and commits, the only remaining type of object
163 is a "tag", which we won't discuss here; refer to gitlink:git-tag[1]
166 So now we know how git uses the object database to represent a
169 * "commit" objects refer to "tree" objects representing the
170 snapshot of a directory tree at a particular point in the
171 history, and refer to "parent" commits to show how they're
172 connected into the project history.
173 * "tree" objects represent the state of a single directory,
174 associating directory names to "blob" objects containing file
175 data and "tree" objects containing subdirectory information.
176 * "blob" objects contain file data without any other structure.
177 * References to commit objects at the head of each branch are
178 stored in files under .git/refs/heads/.
179 * The name of the current branch is stored in .git/HEAD.
181 Note, by the way, that lots of commands take a tree as an argument.
182 But as we can see above, a tree can be referred to in many different
183 ways--by the SHA1 name for that tree, by the name of a commit that
184 refers to the tree, by the name of a branch whose head refers to that
185 tree, etc.--and most such commands can accept any of these names.
187 In command synopses, the word "tree-ish" is sometimes used to
188 designate such an argument.
193 The primary tool we've been using to create commits is "git commit
194 -a", which creates a commit including every change you've made to
195 your working tree. But what if you want to commit changes only to
196 certain files? Or only certain changes to certain files?
198 If we look at the way commits are created under the cover, we'll see
199 that there are more flexible ways creating commits.
201 Continuing with our test-project, let's modify file.txt again:
203 ------------------------------------------------
204 $ echo "hello world, again" >>file.txt
205 ------------------------------------------------
207 but this time instead of immediately making the commit, let's take an
208 intermediate step, and ask for diffs along the way to keep track of
211 ------------------------------------------------
218 $ git update-index file.txt
220 ------------------------------------------------
222 The last diff is empty, but no new commits have been made, and the
223 head still doesn't contain the new line:
225 ------------------------------------------------
227 diff --git a/file.txt b/file.txt
228 index a042389..513feba 100644
234 ------------------------------------------------
236 So "git diff" is comparing against something other than the head.
237 The thing that it's comparing against is actually the index file,
238 which is stored in .git/index in a binary format, but whose contents
239 we can examine with ls-files:
241 ------------------------------------------------
242 $ git ls-files --stage
243 100644 513feba2e53ebbd2532419ded848ba19de88ba00 0 file.txt
244 $ git cat-file -t 513feba2
246 $ git cat-file blob 513feba2
248 ------------------------------------------------
250 So what our "git update-index" did was store a new blob and then put
251 a reference to it in the index file. If we modify the file again,
252 we'll see that the new modifications are reflected in the "git-diff"
255 ------------------------------------------------
256 $ echo 'again?' >>file.txt
258 index 513feba..ba3da7b 100644
265 ------------------------------------------------
267 With the right arguments, git diff can also show us the difference
268 between the working directory and the last commit, or between the
269 index and the last commit:
271 ------------------------------------------------
273 diff --git a/file.txt b/file.txt
274 index a042389..ba3da7b 100644
282 diff --git a/file.txt b/file.txt
283 index a042389..513feba 100644
289 ------------------------------------------------
291 At any time, we can create a new commit using "git commit" (without
292 the -a option), and verify that the state committed only includes the
293 changes stored in the index file, not the additional change that is
294 still only in our working tree:
296 ------------------------------------------------
297 $ git commit -m "repeat"
299 diff --git a/file.txt b/file.txt
300 index 513feba..ba3da7b 100644
307 ------------------------------------------------
309 So by default "git commit" uses the index to create the commit, not
310 the working tree; the -a option to commit tells it to first update
311 the index with all changes in the working tree.
313 Finally, it's worth looking at the effect of "git add" on the index
316 ------------------------------------------------
317 $ echo "goodbye, world" >closing.txt
318 $ git add closing.txt
319 ------------------------------------------------
321 The effect of the "git add" was to add one entry to the index file:
323 ------------------------------------------------
324 $ git ls-files --stage
325 100644 8b9743b20d4b15be3955fc8d5cd2b09cd2336138 0 closing.txt
326 100644 513feba2e53ebbd2532419ded848ba19de88ba00 0 file.txt
327 ------------------------------------------------
329 And, as you can see with cat-file, this new entry refers to the
330 current contents of the file:
332 ------------------------------------------------
333 $ git cat-file blob a6b11f7a
335 ------------------------------------------------
337 The "status" command is a useful way to get a quick summary of the
340 ------------------------------------------------
343 # Updated but not checked in:
346 # new file: closing.txt
349 # Changed but not updated:
350 # (use git-update-index to mark for commit)
354 ------------------------------------------------
356 Since the current state of closing.txt is cached in the index file,
357 it is listed as "updated but not checked in". Since file.txt has
358 changes in the working directory that aren't reflected in the index,
359 it is marked "changed but not updated". At this point, running "git
360 commit" would create a commit that added closing.txt (with its new
361 contents), but that didn't modify file.txt.
363 Also, note that a bare "git diff" shows the changes to file.txt, but
364 not the addition of closing.txt, because the version of closing.txt
365 in the index file is identical to the one in the working directory.
367 In addition to being the staging area for new commits, the index file
368 is also populated from the object database when checking out a
369 branch, and is used to hold the trees involved in a merge operation.
370 See the link:core-tutorial.txt[core tutorial] and the relevant man
376 At this point you should know everything necessary to read the man
377 pages for any of the git commands; one good place to start would be
378 with the commands mentioned in link:everyday.html[Everyday git]. You
379 should be able to find any unknown jargon in the
380 link:glossary.html[Glossary].
382 The link:cvs-migration.html[CVS migration] document explains how to
383 import a CVS repository into git, and shows how to use git in a
386 For some interesting examples of git use, see the
387 link:howto-index.html[howtos].
389 For git developers, the link:core-tutorial.html[Core tutorial] goes
390 into detail on the lower-level git mechanisms involved in, for
391 example, creating a new commit.