Split "git-pull-script" into two parts
[git/git.git] / README
e83c5163 1
4 GIT - the stupid content tracker
6ad6d3d3 6
7"git" can mean anything, depending on your mood.
9 - random three-letter combination that is pronounceable, and not
10 actually used by any common UNIX command. The fact that it is a
11 mispronounciation of "get" may or may not be relevant.
12 - stupid. contemptible and despicable. simple. Take your pick from the
13 dictionary of slang.
14 - "global information tracker": you're in a good mood, and it actually
15 works for you. Angels sing, and a light suddenly fills the room.
16 - "goddamn idiotic truckload of sh*t": when it breaks
18This is a stupid (but extremely fast) directory content manager. It
19doesn't do a whole lot, but what it _does_ do is track directory
20contents efficiently.
22There are two object abstractions: the "object database", and the
23"current directory cache" aka "index".
27 The Object Database (SHA1_FILE_DIRECTORY)
6ad6d3d3 29
30The object database is literally just a content-addressable collection
31of objects. All objects are named by their content, which is
32approximated by the SHA1 hash of the object itself. Objects may refer
33to other objects (by referencing their SHA1 hash), and so you can build
34up a hierarchy of objects.
36All objects have a statically determined "type" aka "tag", which is
37determined at object creation time, and which identifies the format of
38the object (ie how it is used, and how it can refer to other objects).
39There are currently three different object types: "blob", "tree" and
42A "blob" object cannot refer to any other object, and is, like the tag
43implies, a pure storage object containing some user data. It is used to
44actually store the file data, ie a blob object is associated with some
45particular version of some file.
47A "tree" object is an object that ties one or more "blob" objects into a
48directory structure. In addition, a tree object can refer to other tree
49objects, thus creating a directory hierarchy.
51Finally, a "commit" object ties such directory hierarchies together into
52a DAG of revisions - each "commit" is associated with exactly one tree
53(the directory hierarchy at the time of the commit). In addition, a
54"commit" refers to one or more "parent" commit objects that describe the
55history of how we arrived at that directory hierarchy.
57As a special case, a commit object with no parents is called the "root"
58object, and is the point of an initial project commit. Each project
59must have at least one root, and while you can tie several different
60root objects together into one project by creating a commit object which
61has two or more separate roots as its ultimate parents, that's probably
62just going to confuse people. So aim for the notion of "one root object
63per project", even if git itself does not enforce that.
65Regardless of object type, all objects are share the following
66characteristics: they are all in deflated with zlib, and have a header
67that not only specifies their tag, but also size information about the
68data in the object. It's worth noting that the SHA1 hash that is used
69to name the object is always the hash of this _compressed_ object, not
70the original data.
72As a result, the general consistency of an object can always be tested
73independently of the contents or the type of the object: all objects can
74be validated by verifying that (a) their hashes match the content of the
75file and (b) the object successfully inflates to a stream of bytes that
76forms a sequence of <ascii tag without space> + <space> + <ascii decimal
77size> + <byte\0> + <binary object data>.
79The structured objects can further have their structure and connectivity
80to other objects verified. This is generally done with the "fsck-cache"
81program, which generates a full dependency graph of all objects, and
82verifies their internal consistency (in addition to just verifying their
83superficial consistency through the hash).
85The object types in some more detail:
87 BLOB: A "blob" object is nothing but a binary blob of data, and
88 doesn't refer to anything else. There is no signature or any
89 other verification of the data, so while the object is
90 consistent (it _is_ indexed by its sha1 hash, so the data itself
91 is certainly correct), it has absolutely no other attributes.
92 No name associations, no permissions. It is purely a blob of
93 data (ie normally "file contents").
95 In particular, since the blob is entirely defined by its data,
96 if two files in a directory tree (or in multiple different
97 versions of the repository) have the same contents, they will
98 share the same blob object. The object is toally independent
99 of it's location in the directory tree, and renaming a file does
100 not change the object that file is associated with in any way.
102 TREE: The next hierarchical object type is the "tree" object. A tree
103 object is a list of mode/name/blob data, sorted by name.
104 Alternatively, the mode data may specify a directory mode, in
105 which case instead of naming a blob, that name is associated
106 with another TREE object.
108 Like the "blob" object, a tree object is uniquely determined by
109 the set contents, and so two separate but identical trees will
110 always share the exact same object. This is true at all levels,
111 ie it's true for a "leaf" tree (which does not refer to any
112 other trees, only blobs) as well as for a whole subdirectory.
114 For that reason a "tree" object is just a pure data abstraction:
115 it has no history, no signatures, no verification of validity,
116 except that since the contents are again protected by the hash
117 itself, we can trust that the tree is immutable and its contents
118 never change.
120 So you can trust the contents of a tree to be valid, the same
121 way you can trust the contents of a blob, but you don't know
122 where those contents _came_ from.
124 Side note on trees: since a "tree" object is a sorted list of
125 "filename+content", you can create a diff between two trees
126 without actually having to unpack two trees. Just ignore all
127 common parts, and your diff will look right. In other words,
128 you can effectively (and efficiently) tell the difference
129 between any two random trees by O(n) where "n" is the size of
130 the difference, rather than the size of the tree.
132 Side note 2 on trees: since the name of a "blob" depends
133 entirely and exclusively on its contents (ie there are no names
134 or permissions involved), you can see trivial renames or
135 permission changes by noticing that the blob stayed the same.
136 However, renames with data changes need a smarter "diff" implementation.
138CHANGESET: The "changeset" object is an object that introduces the
139 notion of history into the picture. In contrast to the other
140 objects, it doesn't just describe the physical state of a tree,
141 it describes how we got there, and why.
143 A "changeset" is defined by the tree-object that it results in,
144 the parent changesets (zero, one or more) that led up to that
145 point, and a comment on what happened. Again, a changeset is
146 not trusted per se: the contents are well-defined and "safe" due
147 to the cryptographically strong signatures at all levels, but
148 there is no reason to believe that the tree is "good" or that
149 the merge information makes sense. The parents do not have to
150 actually have any relationship with the result, for example.
152 Note on changesets: unlike real SCM's, changesets do not contain
153 rename information or file mode chane information. All of that
154 is implicit in the trees involved (the result tree, and the
155 result trees of the parents), and describing that makes no sense
156 in this idiotic file manager.
158TRUST: The notion of "trust" is really outside the scope of "git", but
159 it's worth noting a few things. First off, since everything is
160 hashed with SHA1, you _can_ trust that an object is intact and
161 has not been messed with by external sources. So the name of an
162 object uniquely identifies a known state - just not a state that
163 you may want to trust.
165 Furthermore, since the SHA1 signature of a changeset refers to
166 the SHA1 signatures of the tree it is associated with and the
167 signatures of the parent, a single named changeset specifies
168 uniquely a whole set of history, with full contents. You can't
169 later fake any step of the way once you have the name of a
170 changeset.
172 So to introduce some real trust in the system, the only thing
173 you need to do is to digitally sign just _one_ special note,
174 which includes the name of a top-level changeset. Your digital
175 signature shows others that you trust that changeset, and the
176 immutability of the history of changesets tells others that they
177 can trust the whole history.
179 In other words, you can easily validate a whole archive by just
180 sending out a single email that tells the people the name (SHA1
181 hash) of the top changeset, and digitally sign that email using
182 something like GPG/PGP.
184 In particular, you can also have a separate archive of "trust
185 points" or tags, which document your (and other peoples) trust.
186 You may, of course, archive these "certificates of trust" using
187 "git" itself, but it's not something "git" does for you.
189Another way of saying the last point: "git" itself only handles content
190integrity, the trust has to come from outside.
e83c5163 192
e83c5163 193
194 The "index" aka "Current Directory Cache" (".git/index")
197The index is a simple binary file, which contains an efficient
198representation of a virtual directory content at some random time. It
199does so by a simple array that associates a set of names, dates,
200permissions and content (aka "blob") objects together. The cache is
201always kept ordered by name, and names are unique (with a few very
202specific rules) at any point in time, but the cache has no long-term
203meaning, and can be partially updated at any time.
205In particular, the index certainly does not need to be consistent with
206the current directory contents (in fact, most operations will depend on
207different ways to make the index _not_ be consistent with the directory
208hierarchy), but it has three very important attributes:
210 (a) it can re-generate the full state it caches (not just the directory
211 structure: it contains pointers to the "blob" objects so that it
212 can regenerate the data too)
214 As a special case, there is a clear and unambiguous one-way mapping
215 from a current directory cache to a "tree object", which can be
216 efficiently created from just the current directory cache without
217 actually looking at any other data. So a directory cache at any
218 one time uniquely specifies one and only one "tree" object (but
219 has additional data to make it easy to match up that tree object
220 with what has happened in the directory)
222 (b) it has efficient methods for finding inconsistencies between that
223 cached state ("tree object waiting to be instantiated") and the
224 current state.
226 (c) it can additionally efficiently represent information about merge
227 conflicts between different tree objects, allowing each pathname to
228 be associated with sufficient information about the trees involved
229 that you can create a three-way merge between them.
231Those are the three ONLY things that the directory cache does. It's a
232cache, and the normal operation is to re-generate it completely from a
233known tree object, or update/compare it with a live tree that is being
234developed. If you blow the directory cache away entirely, you generally
235haven't lost any information as long as you have the name of the tree
236that it described.
238At the same time, the directory index is at the same time also the
239staging area for creating new trees, and creating a new tree always
240involves a controlled modification of the index file. In particular,
241the index file can have the representation of an intermediate tree that
242has not yet been instantiated. So the index can be thought of as a
243write-back cache, which can contain dirty information that has not yet
244been written back to the backing store.
248 The Workflow
251Generally, all "git" operations work on the index file. Some operations
252work _purely_ on the index file (showing the current state of the
253index), but most operations move data to and from the index file. Either
254from the database or from the working directory. Thus there are four
255main combinations:
257 1) working directory -> index
259 You update the index with information from the working directory
260 with the "update-cache" command. You generally update the index
261 information by just specifying the filename you want to update,
262 like so:
264 update-cache filename
266 but to avoid common mistakes with filename globbing etc, the
267 command will not normally add totally new entries or remove old
268 entries, ie it will normally just update existing cache entryes.
270 To tell git that yes, you really do realize that certain files
271 no longer exist in the archive, or that new files should be
272 added, you should use the "--remove" and "--add" flags
273 respectively.
275 NOTE! A "--remove" flag does _not_ mean that subsequent
276 filenames will necessarily be removed: if the files still exist
277 in your directory structure, the index will be updated with
278 their new status, not removed. The only thing "--remove" means
279 is that update-cache will be considering a removed file to be a
280 valid thing, and if the file really does not exist any more, it
281 will update the index accordingly.
283 As a special case, you can also do "update-cache --refresh",
284 which will refresh the "stat" information of each index to match
285 the current stat information. It will _not_ update the object
286 status itself, and it wil only update the fields that are used
287 to quickly test whether an object still matches its old backing
288 store object.
290 2) index -> object database
292 You write your current index file to a "tree" object with the
293 program
295 write-tree
297 that doesn't come with any options - it will just write out the
298 current index into the set of tree objects that describe that
299 state, and it will return the name of the resulting top-level
300 tree. You can use that tree to re-generate the index at any time
301 by going in the other direction:
303 3) object database -> index
305 You read a "tree" file from the object database, and use that to
306 populate (and overwrite - don't do this if your index contains
307 any unsaved state that you might want to restore later!) your
308 current index. Normal operation is just
310 read-tree <sha1 of tree>
312 and your index file will now be equivalent to the tree that you
313 saved earlier. However, that is only your _index_ file: your
314 working directory contents have not been modified.
316 4) index -> working directory
318 You update your working directory from the index by "checking
319 out" files. This is not a very common operation, since normally
320 you'd just keep your files updated, and rather than write to
321 your working directory, you'd tell the index files about the
322 changes in your working directory (ie "update-cache").
324 However, if you decide to jump to a new version, or check out
325 somebody elses version, or just restore a previous tree, you'd
326 populate your index file with read-tree, and then you need to
327 check out the result with
329 checkout-cache filename
331 or, if you want to check out all of the index, use "-a".
333 NOTE! checkout-cache normally refuses to overwrite old files, so
334 if you have an old version of the tree already checked out, you
335 will need to use the "-f" flag (_before_ the "-a" flag or the
336 filename) to _force_ the checkout.
339Finally, there are a few odds and ends which are not purely moving from
340one representation to the other:
342 5) Tying it all together
344 To commit a tree you have instantiated with "write-tree", you'd
345 create a "commit" object that refers to that tree and the
346 history behind it - most notably the "parent" commits that
347 preceded it in history.
349 Normally a "commit" has one parent: the previous state of the
350 tree before a certain change was made. However, sometimes it can
351 have two or more parent commits, in which case we call it a
352 "merge", due to the fact that such a commit brings together
353 ("merges") two or more previous states represented by other
354 commits.
356 In other words, while a "tree" represents a particular directory
357 state of a working directory, a "commit" represents that state
358 in "time", and explains how we got there.
360 You create a commit object by giving it the tree that describes
361 the state at the time of the commit, and a list of parents:
363 commit-tree <tree> -p <parent> [-p <parent2> ..]
365 and then giving the reason for the commit on stdin (either
366 through redirection from a pipe or file, or by just typing it at
367 the tty).
369 commit-tree will return the name of the object that represents
370 that commit, and you should save it away for later use.
371 Normally, you'd commit a new "HEAD" state, and while git doesn't
372 care where you save the note about that state, in practice we
373 tend to just write the result to the file ".git/HEAD", so that
374 we can always see what the last committed state was.
376 6) Examining the data
378 You can examine the data represented in the object database and
379 the index with various helper tools. For every object, you can
380 use "cat-file" to examine details about the object:
382 cat-file -t <objectname>
384 shows the type of the object, and once you have the type (which
385 is usually implicit in where you find the object), you can use
387 cat-file blob|tree|commit <objectname>
389 to show its contents. NOTE! Trees have binary content, and as a
390 result there is a special helper for showing that content,
391 called "ls-tree", which turns the binary content into a more
392 easily readable form.
394 It's especially instructive to look at "commit" objects, since
395 those tend to be small and fairly self-explanatory. In
396 particular, if you follow the convention of having the top
397 commit name in ".git/HEAD", you can do
399 cat-file commit $(cat .git/HEAD)
401 to see what the top commit was.
403 7) Merging multiple trees
405 Git helps you do a three-way merge, which you can expand to
406 n-way by repeating the merge procedure arbitrary times until you
407 finally "commit" the state. The normal situation is that you'd
408 only do one three-way merge (two parents), and commit it, but if
409 you like to, you can do multiple parents in one go.
411 To do a three-way merge, you need the two sets of "commit"
412 objects that you want to merge, use those to find the closest
413 common parent (a third "commit" object), and then use those
414 commit objects to find the state of the directory ("tree"
415 object) at these points.
417 To get the "base" for the merge, you first look up the common
418 parent of two commits with
420 merge-base <commit1> <commit2>
422 which will return you the commit they are both based on. You
423 should now look up the "tree" objects of those commits, which
424 you can easily do with (for example)
426 cat-file commit <commitname> | head -1
428 since the tree object information is always the first line in a
429 commit object.
431 Once you know the three trees you are going to merge (the one
432 "original" tree, aka the common case, and the two "result" trees,
433 aka the branches you want to merge), you do a "merge" read into
434 the index. This will throw away your old index contents, so you
435 should make sure that you've committed those - in fact you would
436 normally always do a merge against your last commit (which
437 should thus match what you have in your current index anyway).
438 To do the merge, do
440 read-tree -m <origtree> <target1tree> <target2tree>
442 which will do all trivial merge operations for you directly in
443 the index file, and you can just write the result out with
444 "write-tree".
446 NOTE! Because the merge is done in the index file, and not in
447 your working directory, your working directory will no longer
448 match your index. You can use "checkout-cache -f -a" to make the
449 effect of the merge be seen in your working directory.
451 NOTE2! Sadly, many merges aren't trivial. If there are files
452 that have been added.moved or removed, or if both branches have
453 modified the same file, you will be left with an index tree that
454 contains "merge entries" in it. Such an index tree can _NOT_ be
455 written out to a tree object, and you will have to resolve any
456 such merge clashes using other tools before you can write out
457 the result.
459 [ fixme: talk about resolving merges here ]