Initial revision of "git", the information manager from hell
[git/git.git] / README
CommitLineData
e83c5163
LT
1
2 GIT - the stupid content tracker
3
4"git" can mean anything, depending on your mood.
5
6 - random three-letter combination that is pronounceable, and not
7 actually used by any common UNIX command. The fact that it is a
8 mispronounciation of "get" may or may not be relevant.
9 - stupid. contemptible and despicable. simple. Take your pick from the
10 dictionary of slang.
11 - "global information tracker": you're in a good mood, and it actually
12 works for you. Angels sing, and a light suddenly fills the room.
13 - "goddamn idiotic truckload of sh*t": when it breaks
14
15This is a stupid (but extremely fast) directory content manager. It
16doesn't do a whole lot, but what it _does_ do is track directory
17contents efficiently.
18
19There are two object abstractions: the "object database", and the
20"current directory cache".
21
22 The Object Database (SHA1_FILE_DIRECTORY)
23
24The object database is literally just a content-addressable collection
25of objects. All objects are named by their content, which is
26approximated by the SHA1 hash of the object itself. Objects may refer
27to other objects (by referencing their SHA1 hash), and so you can build
28up a hierarchy of objects.
29
30There are several kinds of objects in the content-addressable collection
31database. They are all in deflated with zlib, and start off with a tag
32of their type, and size information about the data. The SHA1 hash is
33always the hash of the _compressed_ object, not the original one.
34
35In particular, the consistency of an object can always be tested
36independently of the contents or the type of the object: all objects can
37be validated by verifying that (a) their hashes match the content of the
38file and (b) the object successfully inflates to a stream of bytes that
39forms a sequence of <ascii tag without space> + <space> + <ascii decimal
40size> + <byte\0> + <binary object data>.
41
42BLOB: A "blob" object is nothing but a binary blob of data, and doesn't
43refer to anything else. There is no signature or any other verification
44of the data, so while the object is consistent (it _is_ indexed by its
45sha1 hash, so the data itself is certainly correct), it has absolutely
46no other attributes. No name associations, no permissions. It is
47purely a blob of data (ie normally "file contents").
48
49TREE: The next hierarchical object type is the "tree" object. A tree
50object is a list of permission/name/blob data, sorted by name. In other
51words the tree object is uniquely determined by the set contents, and so
52two separate but identical trees will always share the exact same
53object.
54
55Again, a "tree" object is just a pure data abstraction: it has no
56history, no signatures, no verification of validity, except that the
57contents are again protected by the hash itself. So you can trust the
58contents of a tree, the same way you can trust the contents of a blob,
59but you don't know where those contents _came_ from.
60
61Side note on trees: since a "tree" object is a sorted list of
62"filename+content", you can create a diff between two trees without
63actually having to unpack two trees. Just ignore all common parts, and
64your diff will look right. In other words, you can effectively (and
65efficiently) tell the difference between any two random trees by O(n)
66where "n" is the size of the difference, rather than the size of the
67tree.
68
69Side note 2 on trees: since the name of a "blob" depends entirely and
70exclusively on its contents (ie there are no names or permissions
71involved), you can see trivial renames or permission changes by noticing
72that the blob stayed the same. However, renames with data changes need
73a smarter "diff" implementation.
74
75CHANGESET: The "changeset" object is an object that introduces the
76notion of history into the picture. In contrast to the other objects,
77it doesn't just describe the physical state of a tree, it describes how
78we got there, and why.
79
80A "changeset" is defined by the tree-object that it results in, the
81parent changesets (zero, one or more) that led up to that point, and a
82comment on what happened. Again, a changeset is not trusted per se:
83the contents are well-defined and "safe" due to the cryptographically
84strong signatures at all levels, but there is no reason to believe that
85the tree is "good" or that the merge information makes sense. The
86parents do not have to actually have any relationship with the result,
87for example.
88
89Note on changesets: unlike real SCM's, changesets do not contain rename
90information or file mode chane information. All of that is implicit in
91the trees involved (the result tree, and the result trees of the
92parents), and describing that makes no sense in this idiotic file
93manager.
94
95TRUST: The notion of "trust" is really outside the scope of "git", but
96it's worth noting a few things. First off, since everything is hashed
97with SHA1, you _can_ trust that an object is intact and has not been
98messed with by external sources. So the name of an object uniquely
99identifies a known state - just not a state that you may want to trust.
100
101Furthermore, since the SHA1 signature of a changeset refers to the
102SHA1 signatures of the tree it is associated with and the signatures
103of the parent, a single named changeset specifies uniquely a whole
104set of history, with full contents. You can't later fake any step of
105the way once you have the name of a changeset.
106
107So to introduce some real trust in the system, the only thing you need
108to do is to digitally sign just _one_ special note, which includes the
109name of a top-level changeset. Your digital signature shows others that
110you trust that changeset, and the immutability of the history of
111changesets tells others that they can trust the whole history.
112
113In other words, you can easily validate a whole archive by just sending
114out a single email that tells the people the name (SHA1 hash) of the top
115changeset, and digitally sign that email using something like GPG/PGP.
116
117In particular, you can also have a separate archive of "trust points" or
118tags, which document your (and other peoples) trust. You may, of
119course, archive these "certificates of trust" using "git" itself, but
120it's not something "git" does for you.
121
122Another way of saying the same thing: "git" itself only handles content
123integrity, the trust has to come from outside.
124
125 Current Directory Cache (".dircache/index")
126
127The "current directory cache" is a simple binary file, which contains an
128efficient representation of a virtual directory content at some random
129time. It does so by a simple array that associates a set of names,
130dates, permissions and content (aka "blob") objects together. The cache
131is always kept ordered by name, and names are unique at any point in
132time, but the cache has no long-term meaning, and can be partially
133updated at any time.
134
135In particular, the "current directory cache" certainly does not need to
136be consistent with the current directory contents, but it has two very
137important attributes:
138
139 (a) it can re-generate the full state it caches (not just the directory
140 structure: through the "blob" object it can regenerate the data too)
141
142 As a special case, there is a clear and unambiguous one-way mapping
143 from a current directory cache to a "tree object", which can be
144 efficiently created from just the current directory cache without
145 actually looking at any other data. So a directory cache at any
146 one time uniquely specifies one and only one "tree" object (but
147 has additional data to make it easy to match up that tree object
148 with what has happened in the directory)
149
150
151and
152
153 (b) it has efficient methods for finding inconsistencies between that
154 cached state ("tree object waiting to be instantiated") and the
155 current state.
156
157Those are the two ONLY things that the directory cache does. It's a
158cache, and the normal operation is to re-generate it completely from a
159known tree object, or update/compare it with a live tree that is being
160developed. If you blow the directory cache away entirely, you haven't
161lost any information as long as you have the name of the tree that it
162described.
163
164(But directory caches can also have real information in them: in
165particular, they can have the representation of an intermediate tree
166that has not yet been instantiated. So they do have meaning and usage
167outside of caching - in one sense you can think of the current directory
168cache as being the "work in progress" towards a tree commit).