1A short git tutorial 2==================== 3v0.99.5, Aug 2005 4 5Introduction 6------------ 7 8This is trying to be a short tutorial on setting up and using a git 9repository, mainly because being hands-on and using explicit examples is 10often the best way of explaining what is going on. 11 12In normal life, most people wouldn't use the "core" git programs 13directly, but rather script around them to make them more palatable. 14Understanding the core git stuff may help some people get those scripts 15done, though, and it may also be instructive in helping people 16understand what it is that the higher-level helper scripts are actually 17doing. 18 19The core git is often called "plumbing", with the prettier user 20interfaces on top of it called "porcelain". You may not want to use the 21plumbing directly very often, but it can be good to know what the 22plumbing does for when the porcelain isn't flushing... 23 24 25Creating a git repository 26------------------------- 27 28Creating a new git repository couldn't be easier: all git repositories start 29out empty, and the only thing you need to do is find yourself a 30subdirectory that you want to use as a working tree - either an empty 31one for a totally new project, or an existing working tree that you want 32to import into git. 33 34For our first example, we're going to start a totally new repository from 35scratch, with no pre-existing files, and we'll call it `git-tutorial`. 36To start up, create a subdirectory for it, change into that 37subdirectory, and initialize the git infrastructure with `git-init-db`: 38 39------------------------------------------------ 40mkdir git-tutorial 41cd git-tutorial 42git-init-db 43------------------------------------------------ 44 45to which git will reply 46 47 defaulting to local storage area 48 49which is just git's way of saying that you haven't been doing anything 50strange, and that it will have created a local `.git` directory setup for 51your new project. You will now have a `.git` directory, and you can 52inspect that with `ls`. For your new empty project, it should show you 53three entries, among other things: 54 55 - a symlink called `HEAD`, pointing to `refs/heads/master` (if your 56 platform does not have native symlinks, it is a file containing the 57 line "ref: refs/heads/master") 58+ 59Don't worry about the fact that the file that the `HEAD` link points to 60doesn't even exist yet -- you haven't created the commit that will 61start your `HEAD` development branch yet. 62 63 - a subdirectory called `objects`, which will contain all the 64 objects of your project. You should never have any real reason to 65 look at the objects directly, but you might want to know that these 66 objects are what contains all the real 'data' in your repository. 67 68 - a subdirectory called `refs`, which contains references to objects. 69 70In particular, the `refs` subdirectory will contain two other 71subdirectories, named `heads` and `tags` respectively. They do 72exactly what their names imply: they contain references to any number 73of different 'heads' of development (aka 'branches'), and to any 74'tags' that you have created to name specific versions in your 75repository. 76 77One note: the special `master` head is the default branch, which is 78why the `.git/HEAD` file was created as a symlink to it even if it 79doesn't yet exist. Basically, the `HEAD` link is supposed to always 80point to the branch you are working on right now, and you always 81start out expecting to work on the `master` branch. 82 83However, this is only a convention, and you can name your branches 84anything you want, and don't have to ever even 'have' a `master` 85branch. A number of the git tools will assume that `.git/HEAD` is 86valid, though. 87 88[NOTE] 89An 'object' is identified by its 160-bit SHA1 hash, aka 'object name', 90and a reference to an object is always the 40-byte hex 91representation of that SHA1 name. The files in the `refs` 92subdirectory are expected to contain these hex references 93(usually with a final `\'\n\'` at the end), and you should thus 94expect to see a number of 41-byte files containing these 95references in these `refs` subdirectories when you actually start 96populating your tree. 97 98[NOTE] 99An advanced user may want to take a look at the 100link:repository-layout.html[repository layout] document 101after finishing this tutorial. 102 103You have now created your first git repository. Of course, since it's 104empty, that's not very useful, so let's start populating it with data. 105 106 107Populating a git repository 108--------------------------- 109 110We'll keep this simple and stupid, so we'll start off with populating a 111few trivial files just to get a feel for it. 112 113Start off with just creating any random files that you want to maintain 114in your git repository. We'll start off with a few bad examples, just to 115get a feel for how this works: 116 117------------------------------------------------ 118echo "Hello World" >hello 119echo "Silly example" >example 120------------------------------------------------ 121 122you have now created two files in your working tree (aka 'working directory'), but to 123actually check in your hard work, you will have to go through two steps: 124 125 - fill in the 'index' file (aka 'cache') with the information about your 126 working tree state. 127 128 - commit that index file as an object. 129 130The first step is trivial: when you want to tell git about any changes 131to your working tree, you use the `git-update-index` program. That 132program normally just takes a list of filenames you want to update, but 133to avoid trivial mistakes, it refuses to add new entries to the cache 134(or remove existing ones) unless you explicitly tell it that you're 135adding a new entry with the `\--add` flag (or removing an entry with the 136`\--remove`) flag. 137 138So to populate the index with the two files you just created, you can do 139 140------------------------------------------------ 141git-update-index --add hello example 142------------------------------------------------ 143 144and you have now told git to track those two files. 145 146In fact, as you did that, if you now look into your object directory, 147you'll notice that git will have added two new objects to the object 148database. If you did exactly the steps above, you should now be able to do 149 150 ls .git/objects/??/* 151 152and see two files: 153 154 .git/objects/55/7db03de997c86a4a028e1ebd3a1ceb225be238 155 .git/objects/f2/4c74a2e500f5ee1332c86b94199f52b1d1d962 156 157which correspond with the objects with names of 557db... and f24c7.. 158respectively. 159 160If you want to, you can use `git-cat-file` to look at those objects, but 161you'll have to use the object name, not the filename of the object: 162 163 git-cat-file -t 557db03de997c86a4a028e1ebd3a1ceb225be238 164 165where the `-t` tells `git-cat-file` to tell you what the "type" of the 166object is. Git will tell you that you have a "blob" object (ie just a 167regular file), and you can see the contents with 168 169 git-cat-file "blob" 557db03 170 171which will print out "Hello World". The object 557db03 is nothing 172more than the contents of your file `hello`. 173 174[NOTE] 175Don't confuse that object with the file `hello` itself. The 176object is literally just those specific *contents* of the file, and 177however much you later change the contents in file `hello`, the object 178we just looked at will never change. Objects are immutable. 179 180[NOTE] 181The second example demonstrates that you can 182abbreviate the object name to only the first several 183hexadecimal digits in most places. 184 185Anyway, as we mentioned previously, you normally never actually take a 186look at the objects themselves, and typing long 40-character hex 187names is not something you'd normally want to do. The above digression 188was just to show that `git-update-index` did something magical, and 189actually saved away the contents of your files into the git object 190database. 191 192Updating the cache did something else too: it created a `.git/index` 193file. This is the index that describes your current working tree, and 194something you should be very aware of. Again, you normally never worry 195about the index file itself, but you should be aware of the fact that 196you have not actually really "checked in" your files into git so far, 197you've only *told* git about them. 198 199However, since git knows about them, you can now start using some of the 200most basic git commands to manipulate the files or look at their status. 201 202In particular, let's not even check in the two files into git yet, we'll 203start off by adding another line to `hello` first: 204 205------------------------------------------------ 206echo "It's a new day for git" >>hello 207------------------------------------------------ 208 209and you can now, since you told git about the previous state of `hello`, ask 210git what has changed in the tree compared to your old index, using the 211`git-diff-files` command: 212 213------------ 214git-diff-files 215------------ 216 217Oops. That wasn't very readable. It just spit out its own internal 218version of a `diff`, but that internal version really just tells you 219that it has noticed that "hello" has been modified, and that the old object 220contents it had have been replaced with something else. 221 222To make it readable, we can tell git-diff-files to output the 223differences as a patch, using the `-p` flag: 224 225------------ 226git-diff-files -p 227------------ 228 229which will spit out 230 231------------ 232diff --git a/hello b/hello 233index 557db03..263414f 100644 234--- a/hello 235+++ b/hello 236@@ -1 +1,2 @@ 237 Hello World 238+It's a new day for git 239---- 240 241i.e. the diff of the change we caused by adding another line to `hello`. 242 243In other words, `git-diff-files` always shows us the difference between 244what is recorded in the index, and what is currently in the working 245tree. That's very useful. 246 247A common shorthand for `git-diff-files -p` is to just write `git 248diff`, which will do the same thing. 249 250 251Committing git state 252-------------------- 253 254Now, we want to go to the next stage in git, which is to take the files 255that git knows about in the index, and commit them as a real tree. We do 256that in two phases: creating a 'tree' object, and committing that 'tree' 257object as a 'commit' object together with an explanation of what the 258tree was all about, along with information of how we came to that state. 259 260Creating a tree object is trivial, and is done with `git-write-tree`. 261There are no options or other input: git-write-tree will take the 262current index state, and write an object that describes that whole 263index. In other words, we're now tying together all the different 264filenames with their contents (and their permissions), and we're 265creating the equivalent of a git "directory" object: 266 267------------------------------------------------ 268git-write-tree 269------------------------------------------------ 270 271and this will just output the name of the resulting tree, in this case 272(if you have done exactly as I've described) it should be 273 274 8988da15d077d4829fc51d8544c097def6644dbb 275 276which is another incomprehensible object name. Again, if you want to, 277you can use `git-cat-file -t 8988d\...` to see that this time the object 278is not a "blob" object, but a "tree" object (you can also use 279`git-cat-file` to actually output the raw object contents, but you'll see 280mainly a binary mess, so that's less interesting). 281 282However -- normally you'd never use `git-write-tree` on its own, because 283normally you always commit a tree into a commit object using the 284`git-commit-tree` command. In fact, it's easier to not actually use 285`git-write-tree` on its own at all, but to just pass its result in as an 286argument to `git-commit-tree`. 287 288`git-commit-tree` normally takes several arguments -- it wants to know 289what the 'parent' of a commit was, but since this is the first commit 290ever in this new repository, and it has no parents, we only need to pass in 291the object name of the tree. However, `git-commit-tree` 292also wants to get a commit message 293on its standard input, and it will write out the resulting object name for the 294commit to its standard output. 295 296And this is where we create the `.git/refs/heads/master` file. This file is 297supposed to contain the reference to the top-of-tree, and since that's 298exactly what `git-commit-tree` spits out, we can do this all with a simple 299shell pipeline: 300 301------------------------------------------------ 302echo "Initial commit" | \ 303 git-commit-tree $(git-write-tree) > .git/refs/heads/master 304------------------------------------------------ 305 306which will say: 307 308 Committing initial tree 8988da15d077d4829fc51d8544c097def6644dbb 309 310just to warn you about the fact that it created a totally new commit 311that is not related to anything else. Normally you do this only *once* 312for a project ever, and all later commits will be parented on top of an 313earlier commit, and you'll never see this "Committing initial tree" 314message ever again. 315 316Again, normally you'd never actually do this by hand. There is a 317helpful script called `git commit` that will do all of this for you. So 318you could have just written `git commit` 319instead, and it would have done the above magic scripting for you. 320 321 322Making a change 323--------------- 324 325Remember how we did the `git-update-index` on file `hello` and then we 326changed `hello` afterward, and could compare the new state of `hello` with the 327state we saved in the index file? 328 329Further, remember how I said that `git-write-tree` writes the contents 330of the *index* file to the tree, and thus what we just committed was in 331fact the *original* contents of the file `hello`, not the new ones. We did 332that on purpose, to show the difference between the index state, and the 333state in the working tree, and how they don't have to match, even 334when we commit things. 335 336As before, if we do `git-diff-files -p` in our git-tutorial project, 337we'll still see the same difference we saw last time: the index file 338hasn't changed by the act of committing anything. However, now that we 339have committed something, we can also learn to use a new command: 340`git-diff-index`. 341 342Unlike `git-diff-files`, which showed the difference between the index 343file and the working tree, `git-diff-index` shows the differences 344between a committed *tree* and either the index file or the working 345tree. In other words, `git-diff-index` wants a tree to be diffed 346against, and before we did the commit, we couldn't do that, because we 347didn't have anything to diff against. 348 349But now we can do 350 351 git-diff-index -p HEAD 352 353(where `-p` has the same meaning as it did in `git-diff-files`), and it 354will show us the same difference, but for a totally different reason. 355Now we're comparing the working tree not against the index file, 356but against the tree we just wrote. It just so happens that those two 357are obviously the same, so we get the same result. 358 359Again, because this is a common operation, you can also just shorthand 360it with 361 362 git diff HEAD 363 364which ends up doing the above for you. 365 366In other words, `git-diff-index` normally compares a tree against the 367working tree, but when given the `\--cached` flag, it is told to 368instead compare against just the index cache contents, and ignore the 369current working tree state entirely. Since we just wrote the index 370file to HEAD, doing `git-diff-index \--cached -p HEAD` should thus return 371an empty set of differences, and that's exactly what it does. 372 373[NOTE] 374================ 375`git-diff-index` really always uses the index for its 376comparisons, and saying that it compares a tree against the working 377tree is thus not strictly accurate. In particular, the list of 378files to compare (the "meta-data") *always* comes from the index file, 379regardless of whether the `\--cached` flag is used or not. The `\--cached` 380flag really only determines whether the file *contents* to be compared 381come from the working tree or not. 382 383This is not hard to understand, as soon as you realize that git simply 384never knows (or cares) about files that it is not told about 385explicitly. Git will never go *looking* for files to compare, it 386expects you to tell it what the files are, and that's what the index 387is there for. 388================ 389 390However, our next step is to commit the *change* we did, and again, to 391understand what's going on, keep in mind the difference between "working 392tree contents", "index file" and "committed tree". We have changes 393in the working tree that we want to commit, and we always have to 394work through the index file, so the first thing we need to do is to 395update the index cache: 396 397------------------------------------------------ 398git-update-index hello 399------------------------------------------------ 400 401(note how we didn't need the `\--add` flag this time, since git knew 402about the file already). 403 404Note what happens to the different `git-diff-\*` versions here. After 405we've updated `hello` in the index, `git-diff-files -p` now shows no 406differences, but `git-diff-index -p HEAD` still *does* show that the 407current state is different from the state we committed. In fact, now 408`git-diff-index` shows the same difference whether we use the `--cached` 409flag or not, since now the index is coherent with the working tree. 410 411Now, since we've updated `hello` in the index, we can commit the new 412version. We could do it by writing the tree by hand again, and 413committing the tree (this time we'd have to use the `-p HEAD` flag to 414tell commit that the HEAD was the *parent* of the new commit, and that 415this wasn't an initial commit any more), but you've done that once 416already, so let's just use the helpful script this time: 417 418------------------------------------------------ 419git commit 420------------------------------------------------ 421 422which starts an editor for you to write the commit message and tells you 423a bit about what you have done. 424 425Write whatever message you want, and all the lines that start with '#' 426will be pruned out, and the rest will be used as the commit message for 427the change. If you decide you don't want to commit anything after all at 428this point (you can continue to edit things and update the cache), you 429can just leave an empty message. Otherwise `git commit` will commit 430the change for you. 431 432You've now made your first real git commit. And if you're interested in 433looking at what `git commit` really does, feel free to investigate: 434it's a few very simple shell scripts to generate the helpful (?) commit 435message headers, and a few one-liners that actually do the 436commit itself (`git-commit`). 437 438 439Inspecting Changes 440------------------ 441 442While creating changes is useful, it's even more useful if you can tell 443later what changed. The most useful command for this is another of the 444`diff` family, namely `git-diff-tree`. 445 446`git-diff-tree` can be given two arbitrary trees, and it will tell you the 447differences between them. Perhaps even more commonly, though, you can 448give it just a single commit object, and it will figure out the parent 449of that commit itself, and show the difference directly. Thus, to get 450the same diff that we've already seen several times, we can now do 451 452 git-diff-tree -p HEAD 453 454(again, `-p` means to show the difference as a human-readable patch), 455and it will show what the last commit (in `HEAD`) actually changed. 456 457More interestingly, you can also give `git-diff-tree` the `-v` flag, which 458tells it to also show the commit message and author and date of the 459commit, and you can tell it to show a whole series of diffs. 460Alternatively, you can tell it to be "silent", and not show the diffs at 461all, but just show the actual commit message. 462 463In fact, together with the `git-rev-list` program (which generates a 464list of revisions), `git-diff-tree` ends up being a veritable fount of 465changes. A trivial (but very useful) script called `git-whatchanged` is 466included with git which does exactly this, and shows a log of recent 467activities. 468 469To see the whole history of our pitiful little git-tutorial project, you 470can do 471 472 git log 473 474which shows just the log messages, or if we want to see the log together 475with the associated patches use the more complex (and much more 476powerful) 477 478 git-whatchanged -p --root 479 480and you will see exactly what has changed in the repository over its 481short history. 482 483[NOTE] 484The `\--root` flag is a flag to `git-diff-tree` to tell it to 485show the initial aka 'root' commit too. Normally you'd probably not 486want to see the initial import diff, but since the tutorial project 487was started from scratch and is so small, we use it to make the result 488a bit more interesting. 489 490With that, you should now be having some inkling of what git does, and 491can explore on your own. 492 493[NOTE] 494Most likely, you are not directly using the core 495git Plumbing commands, but using Porcelain like Cogito on top 496of it. Cogito works a bit differently and you usually do not 497have to run `git-update-index` yourself for changed files (you 498do tell underlying git about additions and removals via 499`cg-add` and `cg-rm` commands). Just before you make a commit 500with `cg-commit`, Cogito figures out which files you modified, 501and runs `git-update-index` on them for you. 502 503 504Tagging a version 505----------------- 506 507In git, there are two kinds of tags, a "light" one, and an "annotated tag". 508 509A "light" tag is technically nothing more than a branch, except we put 510it in the `.git/refs/tags/` subdirectory instead of calling it a `head`. 511So the simplest form of tag involves nothing more than 512 513------------------------------------------------ 514git tag my-first-tag 515------------------------------------------------ 516 517which just writes the current `HEAD` into the `.git/refs/tags/my-first-tag` 518file, after which point you can then use this symbolic name for that 519particular state. You can, for example, do 520 521 git diff my-first-tag 522 523to diff your current state against that tag (which at this point will 524obviously be an empty diff, but if you continue to develop and commit 525stuff, you can use your tag as an "anchor-point" to see what has changed 526since you tagged it. 527 528An "annotated tag" is actually a real git object, and contains not only a 529pointer to the state you want to tag, but also a small tag name and 530message, along with optionally a PGP signature that says that yes, 531you really did 532that tag. You create these annotated tags with either the `-a` or 533`-s` flag to `git tag`: 534 535 git tag -s <tagname> 536 537which will sign the current `HEAD` (but you can also give it another 538argument that specifies the thing to tag, ie you could have tagged the 539current `mybranch` point by using `git tag <tagname> mybranch`). 540 541You normally only do signed tags for major releases or things 542like that, while the light-weight tags are useful for any marking you 543want to do -- any time you decide that you want to remember a certain 544point, just create a private tag for it, and you have a nice symbolic 545name for the state at that point. 546 547 548Copying repositories 549-------------------- 550 551Git repositories are normally totally self-sufficient, and it's worth noting 552that unlike CVS, for example, there is no separate notion of 553"repository" and "working tree". A git repository normally *is* the 554working tree, with the local git information hidden in the `.git` 555subdirectory. There is nothing else. What you see is what you got. 556 557[NOTE] 558You can tell git to split the git internal information from 559the directory that it tracks, but we'll ignore that for now: it's not 560how normal projects work, and it's really only meant for special uses. 561So the mental model of "the git information is always tied directly to 562the working tree that it describes" may not be technically 100% 563accurate, but it's a good model for all normal use. 564 565This has two implications: 566 567 - if you grow bored with the tutorial repository you created (or you've 568 made a mistake and want to start all over), you can just do simple 569 570 rm -rf git-tutorial 571+ 572and it will be gone. There's no external repository, and there's no 573history outside the project you created. 574 575 - if you want to move or duplicate a git repository, you can do so. There 576 is `git clone` command, but if all you want to do is just to 577 create a copy of your repository (with all the full history that 578 went along with it), you can do so with a regular 579 `cp -a git-tutorial new-git-tutorial`. 580+ 581Note that when you've moved or copied a git repository, your git index 582file (which caches various information, notably some of the "stat" 583information for the files involved) will likely need to be refreshed. 584So after you do a `cp -a` to create a new copy, you'll want to do 585 586 git-update-index --refresh 587+ 588in the new repository to make sure that the index file is up-to-date. 589 590Note that the second point is true even across machines. You can 591duplicate a remote git repository with *any* regular copy mechanism, be it 592`scp`, `rsync` or `wget`. 593 594When copying a remote repository, you'll want to at a minimum update the 595index cache when you do this, and especially with other peoples' 596repositories you often want to make sure that the index cache is in some 597known state (you don't know *what* they've done and not yet checked in), 598so usually you'll precede the `git-update-index` with a 599 600 git-read-tree --reset HEAD 601 git-update-index --refresh 602 603which will force a total index re-build from the tree pointed to by `HEAD`. 604It resets the index contents to `HEAD`, and then the `git-update-index` 605makes sure to match up all index entries with the checked-out files. 606If the original repository had uncommitted changes in its 607working tree, `git-update-index --refresh` notices them and 608tells you they need to be updated. 609 610The above can also be written as simply 611 612 git reset 613 614and in fact a lot of the common git command combinations can be scripted 615with the `git xyz` interfaces. You can learn things by just looking 616at what the various git scripts do. For example, `git reset` is the 617above two lines implemented in `git-reset`, but some things like 618`git status` and `git commit` are slightly more complex scripts around 619the basic git commands. 620 621Many (most?) public remote repositories will not contain any of 622the checked out files or even an index file, and will *only* contain the 623actual core git files. Such a repository usually doesn't even have the 624`.git` subdirectory, but has all the git files directly in the 625repository. 626 627To create your own local live copy of such a "raw" git repository, you'd 628first create your own subdirectory for the project, and then copy the 629raw repository contents into the `.git` directory. For example, to 630create your own copy of the git repository, you'd do the following 631 632 mkdir my-git 633 cd my-git 634 rsync -rL rsync://rsync.kernel.org/pub/scm/git/git.git/ .git 635 636followed by 637 638 git-read-tree HEAD 639 640to populate the index. However, now you have populated the index, and 641you have all the git internal files, but you will notice that you don't 642actually have any of the working tree files to work on. To get 643those, you'd check them out with 644 645 git-checkout-index -u -a 646 647where the `-u` flag means that you want the checkout to keep the index 648up-to-date (so that you don't have to refresh it afterward), and the 649`-a` flag means "check out all files" (if you have a stale copy or an 650older version of a checked out tree you may also need to add the `-f` 651flag first, to tell git-checkout-index to *force* overwriting of any old 652files). 653 654Again, this can all be simplified with 655 656 git clone rsync://rsync.kernel.org/pub/scm/git/git.git/ my-git 657 cd my-git 658 git checkout 659 660which will end up doing all of the above for you. 661 662You have now successfully copied somebody else's (mine) remote 663repository, and checked it out. 664 665 666Creating a new branch 667--------------------- 668 669Branches in git are really nothing more than pointers into the git 670object database from within the `.git/refs/` subdirectory, and as we 671already discussed, the `HEAD` branch is nothing but a symlink to one of 672these object pointers. 673 674You can at any time create a new branch by just picking an arbitrary 675point in the project history, and just writing the SHA1 name of that 676object into a file under `.git/refs/heads/`. You can use any filename you 677want (and indeed, subdirectories), but the convention is that the 678"normal" branch is called `master`. That's just a convention, though, 679and nothing enforces it. 680 681To show that as an example, let's go back to the git-tutorial repository we 682used earlier, and create a branch in it. You do that by simply just 683saying that you want to check out a new branch: 684 685------------ 686git checkout -b mybranch 687------------ 688 689will create a new branch based at the current `HEAD` position, and switch 690to it. 691 692[NOTE] 693================================================ 694If you make the decision to start your new branch at some 695other point in the history than the current `HEAD`, you can do so by 696just telling `git checkout` what the base of the checkout would be. 697In other words, if you have an earlier tag or branch, you'd just do 698 699------------ 700git checkout -b mybranch earlier-commit 701------------ 702 703and it would create the new branch `mybranch` at the earlier commit, 704and check out the state at that time. 705================================================ 706 707You can always just jump back to your original `master` branch by doing 708 709------------ 710git checkout master 711------------ 712 713(or any other branch-name, for that matter) and if you forget which 714branch you happen to be on, a simple 715 716------------ 717ls -l .git/HEAD 718------------ 719 720will tell you where it's pointing (Note that on platforms with bad or no 721symlink support, you have to execute 722 723------------ 724cat .git/HEAD 725------------ 726 727instead). To get the list of branches you have, you can say 728 729------------ 730git branch 731------------ 732 733which is nothing more than a simple script around `ls .git/refs/heads`. 734There will be asterisk in front of the branch you are currently on. 735 736Sometimes you may wish to create a new branch _without_ actually 737checking it out and switching to it. If so, just use the command 738 739------------ 740git branch <branchname> [startingpoint] 741------------ 742 743which will simply _create_ the branch, but will not do anything further. 744You can then later -- once you decide that you want to actually develop 745on that branch -- switch to that branch with a regular `git checkout` 746with the branchname as the argument. 747 748 749Merging two branches 750-------------------- 751 752One of the ideas of having a branch is that you do some (possibly 753experimental) work in it, and eventually merge it back to the main 754branch. So assuming you created the above `mybranch` that started out 755being the same as the original `master` branch, let's make sure we're in 756that branch, and do some work there. 757 758------------------------------------------------ 759git checkout mybranch 760echo "Work, work, work" >>hello 761git commit -m 'Some work.' hello 762------------------------------------------------ 763 764Here, we just added another line to `hello`, and we used a shorthand for 765doing both `git-update-index hello` and `git commit` by just giving the 766filename directly to `git commit`. The `-m` flag is to give the 767commit log message from the command line. 768 769Now, to make it a bit more interesting, let's assume that somebody else 770does some work in the original branch, and simulate that by going back 771to the master branch, and editing the same file differently there: 772 773------------ 774git checkout master 775------------ 776 777Here, take a moment to look at the contents of `hello`, and notice how they 778don't contain the work we just did in `mybranch` -- because that work 779hasn't happened in the `master` branch at all. Then do 780 781------------ 782echo "Play, play, play" >>hello 783echo "Lots of fun" >>example 784git commit -m 'Some fun.' hello example 785------------ 786 787since the master branch is obviously in a much better mood. 788 789Now, you've got two branches, and you decide that you want to merge the 790work done. Before we do that, let's introduce a cool graphical tool that 791helps you view what's going on: 792 793 gitk --all 794 795will show you graphically both of your branches (that's what the `\--all` 796means: normally it will just show you your current `HEAD`) and their 797histories. You can also see exactly how they came to be from a common 798source. 799 800Anyway, let's exit `gitk` (`^Q` or the File menu), and decide that we want 801to merge the work we did on the `mybranch` branch into the `master` 802branch (which is currently our `HEAD` too). To do that, there's a nice 803script called `git resolve`, which wants to know which branches you want 804to resolve and what the merge is all about: 805 806------------ 807git resolve HEAD mybranch "Merge work in mybranch" 808------------ 809 810where the third argument is going to be used as the commit message if 811the merge can be resolved automatically. 812 813Now, in this case we've intentionally created a situation where the 814merge will need to be fixed up by hand, though, so git will do as much 815of it as it can automatically (which in this case is just merge the `example` 816file, which had no differences in the `mybranch` branch), and say: 817 818 Simple merge failed, trying Automatic merge 819 Auto-merging hello. 820 merge: warning: conflicts during merge 821 ERROR: Merge conflict in hello. 822 fatal: merge program failed 823 Automatic merge failed, fix up by hand 824 825which is way too verbose, but it basically tells you that it failed the 826really trivial merge ("Simple merge") and did an "Automatic merge" 827instead, but that too failed due to conflicts in `hello`. 828 829Not to worry. It left the (trivial) conflict in `hello` in the same form you 830should already be well used to if you've ever used CVS, so let's just 831open `hello` in our editor (whatever that may be), and fix it up somehow. 832I'd suggest just making it so that `hello` contains all four lines: 833 834------------ 835Hello World 836It's a new day for git 837Play, play, play 838Work, work, work 839------------ 840 841and once you're happy with your manual merge, just do a 842 843------------ 844git commit hello 845------------ 846 847which will very loudly warn you that you're now committing a merge 848(which is correct, so never mind), and you can write a small merge 849message about your adventures in git-merge-land. 850 851After you're done, start up `gitk \--all` to see graphically what the 852history looks like. Notice that `mybranch` still exists, and you can 853switch to it, and continue to work with it if you want to. The 854`mybranch` branch will not contain the merge, but next time you merge it 855from the `master` branch, git will know how you merged it, so you'll not 856have to do _that_ merge again. 857 858Another useful tool, especially if you do not always work in X-Window 859environment, is `git show-branch`. 860 861------------------------------------------------ 862$ git show-branch master mybranch 863* [master] Merged "mybranch" changes. 864 ! [mybranch] Some work. 865-- 866+ [master] Merged "mybranch" changes. 867++ [mybranch] Some work. 868------------------------------------------------ 869 870The first two lines indicate that it is showing the two branches 871and the first line of the commit log message from their 872top-of-the-tree commits, you are currently on `master` branch 873(notice the asterisk `*` character), and the first column for 874the later output lines is used to show commits contained in the 875`master` branch, and the second column for the `mybranch` 876branch. Three commits are shown along with their log messages. 877All of them have plus `+` characters in the first column, which 878means they are now part of the `master` branch. Only the "Some 879work" commit has the plus `+` character in the second column, 880because `mybranch` has not been merged to incorporate these 881commits from the master branch. The string inside brackets 882before the commit log message is a short name you can use to 883name the commit. In the above example, 'master' and 'mybranch' 884are branch heads. 'master~1' is the first parent of 'master' 885branch head. Please see 'git-rev-parse' documentation if you 886see more complex cases. 887 888Now, let's pretend you are the one who did all the work in 889`mybranch`, and the fruit of your hard work has finally been merged 890to the `master` branch. Let's go back to `mybranch`, and run 891resolve to get the "upstream changes" back to your branch. 892 893------------ 894git checkout mybranch 895git resolve HEAD master "Merge upstream changes." 896------------ 897 898This outputs something like this (the actual commit object names 899would be different) 900 901 Updating from ae3a2da... to a80b4aa.... 902 example | 1 + 903 hello | 1 + 904 2 files changed, 2 insertions(+), 0 deletions(-) 905 906Because your branch did not contain anything more than what are 907already merged into the `master` branch, the resolve operation did 908not actually do a merge. Instead, it just updated the top of 909the tree of your branch to that of the `master` branch. This is 910often called 'fast forward' merge. 911 912You can run `gitk \--all` again to see how the commit ancestry 913looks like, or run `show-branch`, which tells you this. 914 915------------------------------------------------ 916$ git show-branch master mybranch 917! [master] Merged "mybranch" changes. 918 * [mybranch] Merged "mybranch" changes. 919-- 920++ [master] Merged "mybranch" changes. 921------------------------------------------------ 922 923 924Merging external work 925--------------------- 926 927It's usually much more common that you merge with somebody else than 928merging with your own branches, so it's worth pointing out that git 929makes that very easy too, and in fact, it's not that different from 930doing a `git resolve`. In fact, a remote merge ends up being nothing 931more than "fetch the work from a remote repository into a temporary tag" 932followed by a `git resolve`. 933 934Fetching from a remote repository is done by, unsurprisingly, 935`git fetch`: 936 937 git fetch <remote-repository> 938 939One of the following transports can be used to name the 940repository to download from: 941 942Rsync:: 943 `rsync://remote.machine/path/to/repo.git/` 944+ 945Rsync transport is usable for both uploading and downloading, 946but is completely unaware of what git does, and can produce 947unexpected results when you download from the public repository 948while the repository owner is uploading into it via `rsync` 949transport. Most notably, it could update the files under 950`refs/` which holds the object name of the topmost commits 951before uploading the files in `objects/` -- the downloader would 952obtain head commit object name while that object itself is still 953not available in the repository. For this reason, it is 954considered deprecated. 955 956SSH:: 957 `remote.machine:/path/to/repo.git/` or 958+ 959`ssh://remote.machine/path/to/repo.git/` 960+ 961This transport can be used for both uploading and downloading, 962and requires you to have a log-in privilege over `ssh` to the 963remote machine. It finds out the set of objects the other side 964lacks by exchanging the head commits both ends have and 965transfers (close to) minimum set of objects. It is by far the 966most efficient way to exchange git objects between repositories. 967 968Local directory:: 969 `/path/to/repo.git/` 970+ 971This transport is the same as SSH transport but uses `sh` to run 972both ends on the local machine instead of running other end on 973the remote machine via `ssh`. 974 975GIT Native:: 976 `git://remote.machine/path/to/repo.git/` 977+ 978This transport was designed for anonymous downloading. Like SSH 979transport, it finds out the set of objects the downstream side 980lacks and transfers (close to) minimum set of objects. 981 982HTTP(s):: 983 `http://remote.machine/path/to/repo.git/` 984+ 985HTTP and HTTPS transport are used only for downloading. They 986first obtain the topmost commit object name from the remote site 987by looking at `repo.git/info/refs` file, tries to obtain the 988commit object by downloading from `repo.git/objects/xx/xxx\...` 989using the object name of that commit object. Then it reads the 990commit object to find out its parent commits and the associate 991tree object; it repeats this process until it gets all the 992necessary objects. Because of this behaviour, they are 993sometimes also called 'commit walkers'. 994+ 995The 'commit walkers' are sometimes also called 'dumb 996transports', because they do not require any GIT aware smart 997server like GIT Native transport does. Any stock HTTP server 998would suffice. 999+1000There are (confusingly enough) `git-ssh-fetch` and `git-ssh-upload`1001programs, which are 'commit walkers'; they outlived their1002usefulness when GIT Native and SSH transports were introduced,1003and not used by `git pull` or `git push` scripts.10041005Once you fetch from the remote repository, you `resolve` that1006with your current branch.10071008However -- it's such a common thing to `fetch` and then1009immediately `resolve`, that it's called `git pull`, and you can1010simply do10111012 git pull <remote-repository>10131014and optionally give a branch-name for the remote end as a second1015argument.10161017[NOTE]1018You could do without using any branches at all, by1019keeping as many local repositories as you would like to have1020branches, and merging between them with `git pull`, just like1021you merge between branches. The advantage of this approach is1022that it lets you keep set of files for each `branch` checked1023out and you may find it easier to switch back and forth if you1024juggle multiple lines of development simultaneously. Of1025course, you will pay the price of more disk usage to hold1026multiple working trees, but disk space is cheap these days.10271028[NOTE]1029You could even pull from your own repository by1030giving '.' as <remote-repository> parameter to `git pull`.10311032It is likely that you will be pulling from the same remote1033repository from time to time. As a short hand, you can store1034the remote repository URL in a file under .git/remotes/1035directory, like this:10361037------------------------------------------------1038mkdir -p .git/remotes/1039cat >.git/remotes/linus <<\EOF1040URL: http://www.kernel.org/pub/scm/git/git.git/1041EOF1042------------------------------------------------10431044and use the filename to `git pull` instead of the full URL.1045The URL specified in such file can even be a prefix1046of a full URL, like this:10471048------------------------------------------------1049cat >.git/remotes/jgarzik <<\EOF1050URL: http://www.kernel.org/pub/scm/linux/git/jgarzik/1051EOF1052------------------------------------------------105310541055Examples.10561057. `git pull linus`1058. `git pull linus tag v0.99.1`1059. `git pull jgarzik/netdev-2.6.git/ e100`10601061the above are equivalent to:10621063. `git pull http://www.kernel.org/pub/scm/git/git.git/ HEAD`1064. `git pull http://www.kernel.org/pub/scm/git/git.git/ tag v0.99.1`1065. `git pull http://www.kernel.org/pub/.../jgarzik/netdev-2.6.git e100`106610671068Publishing your work1069--------------------10701071So we can use somebody else's work from a remote repository; but1072how can *you* prepare a repository to let other people pull from1073it?10741075Your do your real work in your working tree that has your1076primary repository hanging under it as its `.git` subdirectory.1077You *could* make that repository accessible remotely and ask1078people to pull from it, but in practice that is not the way1079things are usually done. A recommended way is to have a public1080repository, make it reachable by other people, and when the1081changes you made in your primary working tree are in good shape,1082update the public repository from it. This is often called1083'pushing'.10841085[NOTE]1086This public repository could further be mirrored, and that is1087how git repositories at `kernel.org` are managed.10881089Publishing the changes from your local (private) repository to1090your remote (public) repository requires a write privilege on1091the remote machine. You need to have an SSH account there to1092run a single command, `git-receive-pack`.10931094First, you need to create an empty repository on the remote1095machine that will house your public repository. This empty1096repository will be populated and be kept up-to-date by pushing1097into it later. Obviously, this repository creation needs to be1098done only once.10991100[NOTE]1101`git push` uses a pair of programs,1102`git-send-pack` on your local machine, and `git-receive-pack`1103on the remote machine. The communication between the two over1104the network internally uses an SSH connection.11051106Your private repository's GIT directory is usually `.git`, but1107your public repository is often named after the project name,1108i.e. `<project>.git`. Let's create such a public repository for1109project `my-git`. After logging into the remote machine, create1110an empty directory:11111112------------1113mkdir my-git.git1114------------11151116Then, make that directory into a GIT repository by running1117`git init-db`, but this time, since its name is not the usual1118`.git`, we do things slightly differently:11191120------------1121GIT_DIR=my-git.git git-init-db1122------------11231124Make sure this directory is available for others you want your1125changes to be pulled by via the transport of your choice. Also1126you need to make sure that you have the `git-receive-pack`1127program on the `$PATH`.11281129[NOTE]1130Many installations of sshd do not invoke your shell as the login1131shell when you directly run programs; what this means is that if1132your login shell is `bash`, only `.bashrc` is read and not1133`.bash_profile`. As a workaround, make sure `.bashrc` sets up1134`$PATH` so that you can run `git-receive-pack` program.11351136[NOTE]1137If you plan to publish this repository to be accessed over http,1138you should do `chmod +x my-git.git/hooks/post-update` at this1139point. This makes sure that every time you push into this1140repository, `git-update-server-info` is run.11411142Your "public repository" is now ready to accept your changes.1143Come back to the machine you have your private repository. From1144there, run this command:11451146------------1147git push <public-host>:/path/to/my-git.git master1148------------11491150This synchronizes your public repository to match the named1151branch head (i.e. `master` in this case) and objects reachable1152from them in your current repository.11531154As a real example, this is how I update my public git1155repository. Kernel.org mirror network takes care of the1156propagation to other publicly visible machines:11571158------------1159git push master.kernel.org:/pub/scm/git/git.git/ 1160------------116111621163Packing your repository1164-----------------------11651166Earlier, we saw that one file under `.git/objects/??/` directory1167is stored for each git object you create. This representation1168is efficient to create atomically and safely, but1169not so convenient to transport over the network. Since git objects are1170immutable once they are created, there is a way to optimize the1171storage by "packing them together". The command11721173------------1174git repack1175------------11761177will do it for you. If you followed the tutorial examples, you1178would have accumulated about 17 objects in `.git/objects/??/`1179directories by now. `git repack` tells you how many objects it1180packed, and stores the packed file in `.git/objects/pack`1181directory.11821183[NOTE]1184You will see two files, `pack-\*.pack` and `pack-\*.idx`,1185in `.git/objects/pack` directory. They are closely related to1186each other, and if you ever copy them by hand to a different1187repository for whatever reason, you should make sure you copy1188them together. The former holds all the data from the objects1189in the pack, and the latter holds the index for random1190access.11911192If you are paranoid, running `git-verify-pack` command would1193detect if you have a corrupt pack, but do not worry too much.1194Our programs are always perfect ;-).11951196Once you have packed objects, you do not need to leave the1197unpacked objects that are contained in the pack file anymore.11981199------------1200git prune-packed1201------------12021203would remove them for you.12041205You can try running `find .git/objects -type f` before and after1206you run `git prune-packed` if you are curious. Also `git1207count-objects` would tell you how many unpacked objects are in1208your repository and how much space they are consuming.12091210[NOTE]1211`git pull` is slightly cumbersome for HTTP transport, as a1212packed repository may contain relatively few objects in a1213relatively large pack. If you expect many HTTP pulls from your1214public repository you might want to repack & prune often, or1215never.12161217If you run `git repack` again at this point, it will say1218"Nothing to pack". Once you continue your development and1219accumulate the changes, running `git repack` again will create a1220new pack, that contains objects created since you packed your1221repository the last time. We recommend that you pack your project1222soon after the initial import (unless you are starting your1223project from scratch), and then run `git repack` every once in a1224while, depending on how active your project is.12251226When a repository is synchronized via `git push` and `git pull`1227objects packed in the source repository are usually stored1228unpacked in the destination, unless rsync transport is used.1229While this allows you to use different packing strategies on1230both ends, it also means you may need to repack both1231repositories every once in a while.123212331234Working with Others1235-------------------12361237Although git is a truly distributed system, it is often1238convenient to organize your project with an informal hierarchy1239of developers. Linux kernel development is run this way. There1240is a nice illustration (page 17, "Merges to Mainline") in Randy1241Dunlap's presentation (`http://tinyurl.com/a2jdg`).12421243It should be stressed that this hierarchy is purely *informal*.1244There is nothing fundamental in git that enforces the "chain of1245patch flow" this hierarchy implies. You do not have to pull1246from only one remote repository.12471248A recommended workflow for a "project lead" goes like this:124912501. Prepare your primary repository on your local machine. Your1251 work is done there.125212532. Prepare a public repository accessible to others.1254+1255If other people are pulling from your repository over dumb1256transport protocols, you need to keep this repository 'dumb1257transport friendly'. After `git init-db`,1258`$GIT_DIR/hooks/post-update` copied from the standard templates1259would contain a call to `git-update-server-info` but the1260`post-update` hook itself is disabled by default -- enable it1261with `chmod +x post-update`.126212633. Push into the public repository from your primary1264 repository.126512664. `git repack` the public repository. This establishes a big1267 pack that contains the initial set of objects as the1268 baseline, and possibly `git prune` if the transport1269 used for pulling from your repository supports packed1270 repositories.127112725. Keep working in your primary repository. Your changes1273 include modifications of your own, patches you receive via1274 e-mails, and merges resulting from pulling the "public"1275 repositories of your "subsystem maintainers".1276+1277You can repack this private repository whenever you feel like.127812796. Push your changes to the public repository, and announce it1280 to the public.128112827. Every once in a while, "git repack" the public repository.1283 Go back to step 5. and continue working.128412851286A recommended work cycle for a "subsystem maintainer" who works1287on that project and has an own "public repository" goes like this:128812891. Prepare your work repository, by `git clone` the public1290 repository of the "project lead". The URL used for the1291 initial cloning is stored in `.git/remotes/origin`.129212932. Prepare a public repository accessible to others, just like1294 the "project lead" person does.129512963. Copy over the packed files from "project lead" public1297 repository to your public repository.129812994. Push into the public repository from your primary1300 repository. Run `git repack`, and possibly `git prune` if the1301 transport used for pulling from your repository supports1302 packed repositories.130313045. Keep working in your primary repository. Your changes1305 include modifications of your own, patches you receive via1306 e-mails, and merges resulting from pulling the "public"1307 repositories of your "project lead" and possibly your1308 "sub-subsystem maintainers".1309+1310You can repack this private repository whenever you feel1311like.131213136. Push your changes to your public repository, and ask your1314 "project lead" and possibly your "sub-subsystem1315 maintainers" to pull from it.131613177. Every once in a while, `git repack` the public repository.1318 Go back to step 5. and continue working.131913201321A recommended work cycle for an "individual developer" who does1322not have a "public" repository is somewhat different. It goes1323like this:132413251. Prepare your work repository, by `git clone` the public1326 repository of the "project lead" (or a "subsystem1327 maintainer", if you work on a subsystem). The URL used for1328 the initial cloning is stored in `.git/remotes/origin`.132913302. Do your work in your repository on 'master' branch.133113323. Run `git fetch origin` from the public repository of your1333 upstream every once in a while. This does only the first1334 half of `git pull` but does not merge. The head of the1335 public repository is stored in `.git/refs/heads/origin`.133613374. Use `git cherry origin` to see which ones of your patches1338 were accepted, and/or use `git rebase origin` to port your1339 unmerged changes forward to the updated upstream.134013415. Use `git format-patch origin` to prepare patches for e-mail1342 submission to your upstream and send it out. Go back to1343 step 2. and continue.134413451346Working with Others, Shared Repository Style1347--------------------------------------------13481349If you are coming from CVS background, the style of cooperation1350suggested in the previous section may be new to you. You do not1351have to worry. git supports "shared public repository" style of1352cooperation you are probably more familiar with as well.13531354For this, set up a public repository on a machine that is1355reachable via SSH by people with "commit privileges". Put the1356committers in the same user group and make the repository1357writable by that group.13581359You, as an individual committer, then:13601361- First clone the shared repository to a local repository:1362------------------------------------------------1363$ git clone repo.shared.xz:/pub/scm/project.git/ my-project1364$ cd my-project1365$ hack away1366------------------------------------------------13671368- Merge the work others might have done while you were hacking1369 away:1370------------------------------------------------1371$ git pull origin1372$ test the merge result1373------------------------------------------------1374[NOTE]1375================================1376The first `git clone` would have placed the following in1377`my-project/.git/remotes/origin` file, and that's why this and1378the next step work.1379------------1380URL: repo.shared.xz:/pub/scm/project.git/ my-project1381Pull: master:origin1382------------1383================================13841385- push your work as the new head of the shared1386 repository.1387------------------------------------------------1388$ git push origin master1389------------------------------------------------1390If somebody else pushed into the same shared repository while1391you were working locally, `git push` in the last step would1392complain, telling you that the remote `master` head does not1393fast forward. You need to pull and merge those other changes1394back before you push your work when it happens.139513961397Bundling your work together1398---------------------------13991400It is likely that you will be working on more than one thing at1401a time. It is easy to use those more-or-less independent tasks1402using branches with git.14031404We have already seen how branches work in a previous example,1405with "fun and work" example using two branches. The idea is the1406same if there are more than two branches. Let's say you started1407out from "master" head, and have some new code in the "master"1408branch, and two independent fixes in the "commit-fix" and1409"diff-fix" branches:14101411------------1412$ git show-branch1413! [commit-fix] Fix commit message normalization.1414 ! [diff-fix] Fix rename detection.1415 * [master] Release candidate #11416---1417 + [diff-fix] Fix rename detection.1418 + [diff-fix~1] Better common substring algorithm.1419+ [commit-fix] Fix commit message normalization.1420 + [master] Release candidate #11421+++ [diff-fix~2] Pretty-print messages.1422------------14231424Both fixes are tested well, and at this point, you want to merge1425in both of them. You could merge in 'diff-fix' first and then1426'commit-fix' next, like this:14271428------------1429$ git resolve master diff-fix 'Merge fix in diff-fix'1430$ git resolve master commit-fix 'Merge fix in commit-fix'1431------------14321433Which would result in:14341435------------1436$ git show-branch1437! [commit-fix] Fix commit message normalization.1438 ! [diff-fix] Fix rename detection.1439 * [master] Merge fix in commit-fix1440---1441 + [master] Merge fix in commit-fix1442+ + [commit-fix] Fix commit message normalization.1443 + [master~1] Merge fix in diff-fix1444 ++ [diff-fix] Fix rename detection.1445 ++ [diff-fix~1] Better common substring algorithm.1446 + [master~2] Release candidate #11447+++ [master~3] Pretty-print messages.1448------------14491450However, there is no particular reason to merge in one branch1451first and the other next, when what you have are a set of truly1452independent changes (if the order mattered, then they are not1453independent by definition). You could instead merge those two1454branches into the current branch at once. First let's undo what1455we just did and start over. We would want to get the master1456branch before these two merges by resetting it to 'master~2':14571458------------1459$ git reset --hard master~21460------------14611462You can make sure 'git show-branch' matches the state before1463those two 'git resolve' you just did. Then, instead of running1464two 'git resolve' commands in a row, you would pull these two1465branch heads (this is known as 'making an Octopus'):14661467------------1468$ git pull . commit-fix diff-fix1469$ git show-branch1470! [commit-fix] Fix commit message normalization.1471 ! [diff-fix] Fix rename detection.1472 * [master] Octopus merge of branches 'diff-fix' and 'commit-fix'1473---1474 + [master] Octopus merge of branches 'diff-fix' and 'commit-fix'1475+ + [commit-fix] Fix commit message normalization.1476 ++ [diff-fix] Fix rename detection.1477 ++ [diff-fix~1] Better common substring algorithm.1478 + [master~1] Release candidate #11479+++ [master~2] Pretty-print messages.1480------------14811482Note that you should not do Octopus because you can. An octopus1483is a valid thing to do and often makes it easier to view the1484commit history if you are pulling more than two independent1485changes at the same time. However, if you have merge conflicts1486with any of the branches you are merging in and need to hand1487resolve, that is an indication that the development happened in1488those branches were not independent after all, and you should1489merge two at a time, documenting how you resolved the conflicts,1490and the reason why you preferred changes made in one side over1491the other. Otherwise it would make the project history harder1492to follow, not easier.14931494[ to be continued.. cvsimports ]