core.editor

By default, Git uses whatever you’ve set as your default text editor ($VISUAL or $EDITOR) or else falls back to the vi editor to create and edit your commit and tag messages. To change that default to something else, you can use the core.editor setting:

$ git config --global core.editor emacs

Now, no matter what is set as your default shell editor, Git will fire up Emacs to edit messages.

commit.template

If you set this to the path of a file on your system, Git will use that file as the default message when you commit. For instance, suppose you create a template file at ~/.gitmessage.txt that looks like this:

subject line

what happened

[ticket: X]

To tell Git to use it as the default message that appears in your editor when you run git commit, set the commit.template configuration value:

$ git config --global commit.template ~/.gitmessage.txt
$ git commit

Then, your editor will open to something like this for your placeholder commit message when you commit:

subject line

what happened

[ticket: X]
# Please enter the commit message for your changes. Lines starting
# with '#' will be ignored, and an empty message aborts the commit.
# On branch master
# Changes to be committed:
#   (use "git reset HEAD <file>..." to unstage)
#
# modified:   lib/test.rb
#
~
~
".git/COMMIT_EDITMSG" 14L, 297C

If your team has a commit-message policy, then putting a template for that policy on your system and configuring Git to use it by default can help increase the chance of that policy being followed regularly.

core.pager

This setting determines which pager is used when Git pages output such as log and diff. You can set it to more or to your favorite pager (by default, it’s less), or you can turn it off by setting it to a blank string:

$ git config --global core.pager ''

If you run that, Git will page the entire output of all commands, no matter how long they are.

user.signingkey

If you’re making signed annotated tags (as discussed in Signing Your Work), setting your GPG signing key as a configuration setting makes things easier. Set your key ID like so:

$ git config --global user.signingkey <gpg-key-id>

Now, you can sign tags without having to specify your key every time with the git tag command:

$ git tag -s <tag-name>

core.excludesfile

You can put patterns in your project’s .gitignore file to have Git not see them as untracked files or try to stage them when you run git add on them, as discussed in Ignoring Files.

But sometimes you want to ignore certain files for all repositories that you work with. If your computer is running Mac OS X, you’re probably familiar with .DS_Store files. If your preferred editor is Emacs or Vim, you know about filenames that end with a ~ or .swp.

This setting lets you write a kind of global .gitignore file. If you create a ~/.gitignore_global file with these contents:

*~
.*.swp
.DS_Store

…and you run git config --global core.excludesfile ~/.gitignore_global, Git will never again bother you about those files.

help.autocorrect

If you mistype a command, it shows you something like this:

$ git chekcout master
git: 'chekcout' is not a git command. See 'git --help'.

Did you mean this?
    checkout

Git helpfully tries to figure out what you meant, but it still refuses to do it. If you set help.autocorrect to 1, Git will actually run this command for you:

$ git chekcout master
WARNING: You called a Git command named 'chekcout', which does not exist.
Continuing under the assumption that you meant 'checkout'
in 0.1 seconds automatically...

Note that “0.1 seconds” business. help.autocorrect is actually an integer which represents tenths of a second. So if you set it to 50, Git will give you 5 seconds to change your mind before executing the autocorrected command.

color.ui

Git automatically colors most of its output, but there’s a master switch if you don’t like this behavior. To turn off all Git’s colored terminal output, do this:

$ git config --global color.ui false

The default setting is auto, which colors output when it’s going straight to a terminal, but omits the color-control codes when the output is redirected to a pipe or a file.

You can also set it to always to ignore the difference between terminals and pipes. You’ll rarely want this; in most scenarios, if you want color codes in your redirected output, you can instead pass a --color flag to the Git command to force it to use color codes. The default setting is almost always what you’ll want.

color.*

If you want to be more specific about which commands are colored and how, Git provides verb-specific coloring settings. Each of these can be set to true, false, or always:

color.branch
color.diff
color.interactive
color.status

In addition, each of these has subsettings you can use to set specific colors for parts of the output, if you want to override each color. For example, to set the meta information in your diff output to blue foreground, black background, and bold text, you can run

$ git config --global color.diff.meta "blue black bold"

You can set the color to any of the following values: normal, black, red, green, yellow, blue, magenta, cyan, or white. If you want an attribute like bold in the previous example, you can choose from bold, dim, ul (underline), blink, and reverse (swap foreground and background).

core.autocrlf

If you’re programming on Windows and working with people who are not (or vice-versa), you’ll probably run into line-ending issues at some point. This is because Windows uses both a carriage-return character and a linefeed character for newlines in its files, whereas Mac and Linux systems use only the linefeed character. This is a subtle but incredibly annoying fact of cross-platform work; many editors on Windows silently replace existing LF-style line endings with CRLF, or insert both line-ending characters when the user hits the enter key.

Git can handle this by auto-converting CRLF line endings into LF when you add a file to the index, and vice versa when it checks out code onto your filesystem. You can turn on this functionality with the core.autocrlf setting. If you’re on a Windows machine, set it to true – this converts LF endings into CRLF when you check out code:

$ git config --global core.autocrlf true

If you’re on a Linux or Mac system that uses LF line endings, then you don’t want Git to automatically convert them when you check out files; however, if a file with CRLF endings accidentally gets introduced, then you may want Git to fix it. You can tell Git to convert CRLF to LF on commit but not the other way around by setting core.autocrlf to input:

$ git config --global core.autocrlf input

This setup should leave you with CRLF endings in Windows checkouts, but LF endings on Mac and Linux systems and in the repository.

If you’re a Windows programmer doing a Windows-only project, then you can turn off this functionality, recording the carriage returns in the repository by setting the config value to false:

$ git config --global core.autocrlf false

core.whitespace

Git comes preset to detect and fix some whitespace issues. It can look for six primary whitespace issues – three are enabled by default and can be turned off, and three are disabled by default but can be activated.

The three that are turned on by default are blank-at-eol, which looks for spaces at the end of a line; blank-at-eof, which notices blank lines at the end of a file; and space-before-tab, which looks for spaces before tabs at the beginning of a line.

The three that are disabled by default but can be turned on are indent-with-non-tab, which looks for lines that begin with spaces instead of tabs (and is controlled by the tabwidth option); tab-in-indent, which watches for tabs in the indentation portion of a line; and cr-at-eol, which tells Git that carriage returns at the end of lines are OK.

You can tell Git which of these you want enabled by setting core.whitespace to the values you want on or off, separated by commas. You can disable settings by either leaving them out of the setting string or prepending a - in front of the value. For example, if you want all but cr-at-eol to be set, you can do this:

$ git config --global core.whitespace \
    trailing-space,space-before-tab,indent-with-non-tab

Git will detect these issues when you run a git diff command and try to color them so you can possibly fix them before you commit. It will also use these values to help you when you apply patches with git apply. When you’re applying patches, you can ask Git to warn you if it’s applying patches with the specified whitespace issues:

$ git apply --whitespace=warn <patch>

Or you can have Git try to automatically fix the issue before applying the patch:

$ git apply --whitespace=fix <patch>

These options apply to the git rebase command as well. If you’ve committed whitespace issues but haven’t yet pushed upstream, you can run git rebase --whitespace=fix to have Git automatically fix whitespace issues as it’s rewriting the patches.

receive.fsckObjects

Git is capable of making sure every object received during a push still matches its SHA-1 checksum and points to valid objects. However, it doesn’t do this by default; it’s a fairly expensive operation, and might slow down the operation, especially on large repositories or pushes. If you want Git to check object consistency on every push, you can force it to do so by setting receive.fsckObjects to true:

$ git config --system receive.fsckObjects true

Now, Git will check the integrity of your repository before each push is accepted to make sure faulty (or malicious) clients aren’t introducing corrupt data.

receive.denyNonFastForwards

If you rebase commits that you’ve already pushed and then try to push again, or otherwise try to push a commit to a remote branch that doesn’t contain the commit that the remote branch currently points to, you’ll be denied. This is generally good policy; but in the case of the rebase, you may determine that you know what you’re doing and can force-update the remote branch with a -f flag to your push command.

To tell Git to refuse force-pushes, set receive.denyNonFastForwards:

$ git config --system receive.denyNonFastForwards true

The other way you can do this is via server-side receive hooks, which we’ll cover in a bit. That approach lets you do more complex things like deny non-fast-forwards to a certain subset of users.

receive.denyDeletes

One of the workarounds to the denyNonFastForwards policy is for the user to delete the branch and then push it back up with the new reference. To avoid this, set receive.denyDeletes to true:

$ git config --system receive.denyDeletes true

This denies any deletion of branches or tags – no user can do it. To remove remote branches, you must remove the ref files from the server manually. There are also more interesting ways to do this on a per-user basis via ACLs, as you’ll learn in An Example Git-Enforced Policy.

Identifying Binary Files

Some files look like text files but for all intents and purposes are to be treated as binary data. For instance, Xcode projects on the Mac contain a file that ends in .pbxproj, which is basically a JSON (plain-text JavaScript data format) dataset written out to disk by the IDE, which records your build settings and so on. Although it’s technically a text file (because it’s all UTF-8), you don’t want to treat it as such because it’s really a lightweight database – you can’t merge the contents if two people change it, and diffs generally aren’t helpful. The file is meant to be consumed by a machine. In essence, you want to treat it like a binary file.

To tell Git to treat all pbxproj files as binary data, add the following line to your .gitattributes file:

*.pbxproj binary

Now, Git won’t try to convert or fix CRLF issues; nor will it try to compute or print a diff for changes in this file when you run git show or git diff on your project.

Diffing Binary Files

You can also use the Git attributes functionality to effectively diff binary files. You do this by telling Git how to convert your binary data to a text format that can be compared via the normal diff.

First, you’ll use this technique to solve one of the most annoying problems known to humanity: version-controlling Microsoft Word documents. Everyone knows that Word is the most horrific editor around, but oddly, everyone still uses it. If you want to version-control Word documents, you can stick them in a Git repository and commit every once in a while; but what good does that do? If you run git diff normally, you only see something like this:

$ git diff
diff --git a/chapter1.docx b/chapter1.docx
index 88839c4..4afcb7c 100644
Binary files a/chapter1.docx and b/chapter1.docx differ

You can’t directly compare two versions unless you check them out and scan them manually, right? It turns out you can do this fairly well using Git attributes. Put the following line in your .gitattributes file:

*.docx diff=word

This tells Git that any file that matches this pattern (.docx) should use the “word” filter when you try to view a diff that contains changes. What is the “word” filter? You have to set it up. Here you’ll configure Git to use the docx2txt program to convert Word documents into readable text files, which it will then diff properly.

First, you’ll need to install docx2txt; you can download it from http://docx2txt.sourceforge.net. Follow the instructions in the INSTALL file to put it somewhere your shell can find it. Next, you’ll write a wrapper script to convert output to the format Git expects. Create a file that’s somewhere in your path called docx2txt, and add these contents:

#!/bin/bash
docx2txt.pl $1 -

Don’t forget to chmod a+x that file. Finally, you can configure Git to use this script:

$ git config diff.word.textconv docx2txt

Now Git knows that if it tries to do a diff between two snapshots, and any of the files end in .docx, it should run those files through the “word” filter, which is defined as the docx2txt program. This effectively makes nice text-based versions of your Word files before attempting to diff them.

Here’s an example: Chapter 1 of this book was converted to Word format and committed in a Git repository. Then a new paragraph was added. Here’s what git diff shows:

$ git diff
diff --git a/chapter1.docx b/chapter1.docx
index 0b013ca..ba25db5 100644
--- a/chapter1.docx
+++ b/chapter1.docx
@@ -2,6 +2,7 @@
 This chapter will be about getting started with Git. We will begin at the beginning by explaining some background on version control tools, then move on to how to get Git running on your system and finally how to get it setup to start working with. At the end of this chapter you should understand why Git is around, why you should use it and you should be all setup to do so.
 1.1. About Version Control
 What is "version control", and why should you care? Version control is a system that records changes to a file or set of files over time so that you can recall specific versions later. For the examples in this book you will use software source code as the files being version controlled, though in reality you can do this with nearly any type of file on a computer.
+Testing: 1, 2, 3.
 If you are a graphic or web designer and want to keep every version of an image or layout (which you would most certainly want to), a Version Control System (VCS) is a very wise thing to use. It allows you to revert files back to a previous state, revert the entire project back to a previous state, compare changes over time, see who last modified something that might be causing a problem, who introduced an issue and when, and more. Using a VCS also generally means that if you screw things up or lose files, you can easily recover. In addition, you get all this for very little overhead.
 1.1.1. Local Version Control Systems
 Many people's version-control method of choice is to copy files into another directory (perhaps a time-stamped directory, if they're clever). This approach is very common because it is so simple, but it is also incredibly error prone. It is easy to forget which directory you're in and accidentally write to the wrong file or copy over files you don't mean to.

Git successfully and succinctly tells us that we added the string “Testing: 1, 2, 3.”, which is correct. It’s not perfect – formatting changes wouldn’t show up here – but it certainly works.

Another interesting problem you can solve this way involves diffing image files. One way to do this is to run image files through a filter that extracts their EXIF information – metadata that is recorded with most image formats. If you download and install the exiftool program, you can use it to convert your images into text about the metadata, so at least the diff will show you a textual representation of any changes that happened. Put the following line in your .gitattributes file:

*.png diff=exif

Configure Git to use this tool:

$ git config diff.exif.textconv exiftool

If you replace an image in your project and run git diff, you see something like this:

diff --git a/image.png b/image.png
index 88839c4..4afcb7c 100644
--- a/image.png
+++ b/image.png
@@ -1,12 +1,12 @@
 ExifTool Version Number         : 7.74
-File Size                       : 70 kB
-File Modification Date/Time     : 2009:04:21 07:02:45-07:00
+File Size                       : 94 kB
+File Modification Date/Time     : 2009:04:21 07:02:43-07:00
 File Type                       : PNG
 MIME Type                       : image/png
-Image Width                     : 1058
-Image Height                    : 889
+Image Width                     : 1056
+Image Height                    : 827
 Bit Depth                       : 8
 Color Type                      : RGB with Alpha

You can easily see that the file size and image dimensions have both changed.

export-ignore

You can tell Git not to export certain files or directories when generating an archive. If there is a subdirectory or file that you don’t want to include in your archive file but that you do want checked into your project, you can determine those files via the export-ignore attribute.

For example, say you have some test files in a test/ subdirectory, and it doesn’t make sense to include them in the tarball export of your project. You can add the following line to your Git attributes file:

test/ export-ignore

Now, when you run git archive to create a tarball of your project, that directory won’t be included in the archive.

export-subst

When exporting files for deployment you can apply git log's formatting and keyword-expansion processing to selected portions of files marked with the export-subst attribute.

For instance, if you want to include a file named LAST_COMMIT in your project, and have metadata about the last commit automatically injected into it when git archive runs, you can for example set up your .gitattributes and LAST_COMMIT files like this:

LAST_COMMIT export-subst

$ echo 'Last commit date: $Format:%cd by %aN$' > LAST_COMMIT
$ git add LAST_COMMIT .gitattributes
$ git commit -am 'adding LAST_COMMIT file for archives'

When you run git archive, the contents of the archived file will look like this:

$ git archive HEAD | tar xCf ../deployment-testing -
$ cat ../deployment-testing/LAST_COMMIT
Last commit date: Tue Apr 21 08:38:48 2009 -0700 by Scott Chacon

The substitutions can include for example the commit message and any git notes, and git log can do simple word wrapping:

$ echo '$Format:Last commit: %h by %aN at %cd%n%+w(76,6,9)%B$' > LAST_COMMIT
$ git commit -am 'export-subst uses git log's custom formatter

git archive uses git log's `pretty=format:` processor
directly, and strips the surrounding `$Format:` and `$`
markup from the output.
'
$ git archive @ | tar xfO - LAST_COMMIT
Last commit: 312ccc8 by Jim Hill at Fri May 8 09:14:04 2015 -0700
       export-subst uses git log's custom formatter

         git archive uses git log's `pretty=format:` processor directly, and
         strips the surrounding `$Format:` and `$` markup from the output.

The resulting archive is suitable for deployment work, but like any exported archive it isn’t suitable for further development work.

Committing-Workflow Hooks

The first four hooks have to do with the committing process.

The pre-commit hook is run first, before you even type in a commit message. It’s used to inspect the snapshot that’s about to be committed, to see if you’ve forgotten something, to make sure tests run, or to examine whatever you need to inspect in the code. Exiting non-zero from this hook aborts the commit, although you can bypass it with git commit --no-verify. You can do things like check for code style (run lint or something equivalent), check for trailing whitespace (the default hook does exactly this), or check for appropriate documentation on new methods.

The prepare-commit-msg hook is run before the commit message editor is fired up but after the default message is created. It lets you edit the default message before the commit author sees it. This hook takes a few parameters: the path to the file that holds the commit message so far, the type of commit, and the commit SHA-1 if this is an amended commit. This hook generally isn’t useful for normal commits; rather, it’s good for commits where the default message is auto-generated, such as templated commit messages, merge commits, squashed commits, and amended commits. You may use it in conjunction with a commit template to programmatically insert information.

The commit-msg hook takes one parameter, which again is the path to a temporary file that contains the commit message written by the developer. If this script exits non-zero, Git aborts the commit process, so you can use it to validate your project state or commit message before allowing a commit to go through. In the last section of this chapter, We’ll demonstrate using this hook to check that your commit message is conformant to a required pattern.

After the entire commit process is completed, the post-commit hook runs. It doesn’t take any parameters, but you can easily get the last commit by running git log -1 HEAD. Generally, this script is used for notification or something similar.

Email Workflow Hooks

You can set up three client-side hooks for an email-based workflow. They’re all invoked by the git am command, so if you aren’t using that command in your workflow, you can safely skip to the next section. If you’re taking patches over email prepared by git format-patch, then some of these may be helpful to you.

The first hook that is run is applypatch-msg. It takes a single argument: the name of the temporary file that contains the proposed commit message. Git aborts the patch if this script exits non-zero. You can use this to make sure a commit message is properly formatted, or to normalize the message by having the script edit it in place.

The next hook to run when applying patches via git am is pre-applypatch. Somewhat confusingly, it is run after the patch is applied but before a commit is made, so you can use it to inspect the snapshot before making the commit. You can run tests or otherwise inspect the working tree with this script. If something is missing or the tests don’t pass, exiting non-zero aborts the git am script without committing the patch.

The last hook to run during a git am operation is post-applypatch, which runs after the commit is made. You can use it to notify a group or the author of the patch you pulled in that you’ve done so. You can’t stop the patching process with this script.

Other Client Hooks

The pre-rebase hook runs before you rebase anything and can halt the process by exiting non-zero. You can use this hook to disallow rebasing any commits that have already been pushed. The example pre-rebase hook that Git installs does this, although it makes some assumptions that may not match with your workflow.

The post-rewrite hook is run by commands that replace commits, such as git commit --amend and git rebase (though not by git filter-branch). Its single argument is which command triggered the rewrite, and it receives a list of rewrites on stdin. This hook has many of the same uses as the post-checkout and post-merge hooks.

After you run a successful git checkout, the post-checkout hook runs; you can use it to set up your working directory properly for your project environment. This may mean moving in large binary files that you don’t want source controlled, auto-generating documentation, or something along those lines.

The post-merge hook runs after a successful merge command. You can use it to restore data in the working tree that Git can’t track, such as permissions data. This hook can likewise validate the presence of files external to Git control that you may want copied in when the working tree changes.

The pre-push hook runs during git push, after the remote refs have been updated but before any objects have been transferred. It receives the name and location of the remote as parameters, and a list of to-be-updated refs through stdin. You can use it to validate a set of ref updates before a push occurs (a non-zero exit code will abort the push).

Git occasionally does garbage collection as part of its normal operation, by invoking git gc --auto. The pre-auto-gc hook is invoked just before the garbage collection takes place, and can be used to notify you that this is happening, or to abort the collection if now isn’t a good time.

pre-receive

The first script to run when handling a push from a client is pre-receive. It takes a list of references that are being pushed from stdin; if it exits non-zero, none of them are accepted. You can use this hook to do things like make sure none of the updated references are non-fast-forwards, or to do access control for all the refs and files they’re modifying with the push.

update

The update script is very similar to the pre-receive script, except that it’s run once for each branch the pusher is trying to update. If the pusher is trying to push to multiple branches, pre-receive runs only once, whereas update runs once per branch they’re pushing to. Instead of reading from stdin, this script takes three arguments: the name of the reference (branch), the SHA-1 that reference pointed to before the push, and the SHA-1 the user is trying to push. If the update script exits non-zero, only that reference is rejected; other references can still be updated.

post-receive

The post-receive hook runs after the entire process is completed and can be used to update other services or notify users. It takes the same stdin data as the pre-receive hook. Examples include emailing a list, notifying a continuous integration server, or updating a ticket-tracking system – you can even parse the commit messages to see if any tickets need to be opened, modified, or closed. This script can’t stop the push process, but the client doesn’t disconnect until it has completed, so be careful if you try to do anything that may take a long time.

Enforcing a Specific Commit-Message Format

Your first challenge is to enforce that each commit message adheres to a particular format. Just to have a target, assume that each message has to include a string that looks like “ref: 1234” because you want each commit to link to a work item in your ticketing system. You must look at each commit being pushed up, see if that string is in the commit message, and, if the string is absent from any of the commits, exit non-zero so the push is rejected.

You can get a list of the SHA-1 values of all the commits that are being pushed by taking the $newrev and $oldrev values and passing them to a Git plumbing command called git rev-list. This is basically the git log command, but by default it prints out only the SHA-1 values and no other information. So, to get a list of all the commit SHA-1s introduced between one commit SHA-1 and another, you can run something like this:

$ git rev-list 538c33..d14fc7
d14fc7c847ab946ec39590d87783c69b031bdfb7
9f585da4401b0a3999e84113824d15245c13f0be
234071a1be950e2a8d078e6141f5cd20c1e61ad3
dfa04c9ef3d5197182f13fb5b9b1fb7717d2222a
17716ec0f1ff5c77eff40b7fe912f9f6cfd0e475

You can take that output, loop through each of those commit SHA-1s, grab the message for it, and test that message against a regular expression that looks for a pattern.

You have to figure out how to get the commit message from each of these commits to test. To get the raw commit data, you can use another plumbing command called git cat-file. We’ll go over all these plumbing commands in detail in Git Internals; but for now, here’s what that command gives you:

$ git cat-file commit ca82a6
tree cfda3bf379e4f8dba8717dee55aab78aef7f4daf
parent 085bb3bcb608e1e8451d4b2432f8ecbe6306e7e7
author Scott Chacon <schacon@gmail.com> 1205815931 -0700
committer Scott Chacon <schacon@gmail.com> 1240030591 -0700

changed the version number

A simple way to get the commit message from a commit when you have the SHA-1 value is to go to the first blank line and take everything after that. You can do so with the sed command on Unix systems:

$ git cat-file commit ca82a6 | sed '1,/^$/d'
changed the version number

You can use that incantation to grab the commit message from each commit that is trying to be pushed and exit if you see anything that doesn’t match. To exit the script and reject the push, exit non-zero. The whole method looks like this:

$regex = /\[ref: (\d+)\]/

# enforced custom commit message format
def check_message_format
  missed_revs = `git rev-list #{$oldrev}..#{$newrev}`.split("\n")
  missed_revs.each do |rev|
    message = `git cat-file commit #{rev} | sed '1,/^$/d'`
    if !$regex.match(message)
      puts "[POLICY] Your message is not formatted correctly"
      exit 1
    end
  end
end
check_message_format

Putting that in your update script will reject updates that contain commits that have messages that don’t adhere to your rule.

Enforcing a User-Based ACL System

Suppose you want to add a mechanism that uses an access control list (ACL) that specifies which users are allowed to push changes to which parts of your projects. Some people have full access, and others can only push changes to certain subdirectories or specific files. To enforce this, you’ll write those rules to a file named acl that lives in your bare Git repository on the server. You’ll have the update hook look at those rules, see what files are being introduced for all the commits being pushed, and determine whether the user doing the push has access to update all those files.

The first thing you’ll do is write your ACL. Here you’ll use a format very much like the CVS ACL mechanism: it uses a series of lines, where the first field is avail or unavail, the next field is a comma-delimited list of the users to which the rule applies, and the last field is the path to which the rule applies (blank meaning open access). All of these fields are delimited by a pipe (|) character.

In this case, you have a couple of administrators, some documentation writers with access to the doc directory, and one developer who only has access to the lib and tests directories, and your ACL file looks like this:

avail|nickh,pjhyett,defunkt,tpw
avail|usinclair,cdickens,ebronte|doc
avail|schacon|lib
avail|schacon|tests

You begin by reading this data into a structure that you can use. In this case, to keep the example simple, you’ll only enforce the avail directives. Here is a method that gives you an associative array where the key is the user name and the value is an array of paths to which the user has write access:

def get_acl_access_data(acl_file)
  # read in ACL data
  acl_file = File.read(acl_file).split("\n").reject { |line| line == '' }
  access = {}
  acl_file.each do |line|
    avail, users, path = line.split('|')
    next unless avail == 'avail'
    users.split(',').each do |user|
      access[user] ||= []
      access[user] << path
    end
  end
  access
end

On the ACL file you looked at earlier, this get_acl_access_data method returns a data structure that looks like this:

{"defunkt"=>[nil],
 "tpw"=>[nil],
 "nickh"=>[nil],
 "pjhyett"=>[nil],
 "schacon"=>["lib", "tests"],
 "cdickens"=>["doc"],
 "usinclair"=>["doc"],
 "ebronte"=>["doc"]}

Now that you have the permissions sorted out, you need to determine what paths the commits being pushed have modified, so you can make sure the user who’s pushing has access to all of them.

You can pretty easily see what files have been modified in a single commit with the --name-only option to the git log command (mentioned briefly in Git Basics):

$ git log -1 --name-only --pretty=format:'' 9f585d

README
lib/test.rb

If you use the ACL structure returned from the get_acl_access_data method and check it against the listed files in each of the commits, you can determine whether the user has access to push all of their commits:

# only allows certain users to modify certain subdirectories in a project
def check_directory_perms
  access = get_acl_access_data('acl')

  # see if anyone is trying to push something they can't
  new_commits = `git rev-list #{$oldrev}..#{$newrev}`.split("\n")
  new_commits.each do |rev|
    files_modified = `git log -1 --name-only --pretty=format:'' #{rev}`.split("\n")
    files_modified.each do |path|
      next if path.size == 0
      has_file_access = false
      access[$user].each do |access_path|
        if !access_path  # user has access to everything
           || (path.start_with? access_path) # access to this path
          has_file_access = true
        end
      end
      if !has_file_access
        puts "[POLICY] You do not have access to push to #{path}"
        exit 1
      end
    end
  end
end

check_directory_perms

You get a list of new commits being pushed to your server with git rev-list. Then, for each of those commits, you find which files are modified and make sure the user who’s pushing has access to all the paths being modified.

Now your users can’t push any commits with badly formed messages or with modified files outside of their designated paths.

Testing It Out

If you run chmod u+x .git/hooks/update, which is the file into which you should have put all this code, and then try to push a commit with a non-compliant message, you get something like this:

$ git push -f origin master
Counting objects: 5, done.
Compressing objects: 100% (3/3), done.
Writing objects: 100% (3/3), 323 bytes, done.
Total 3 (delta 1), reused 0 (delta 0)
Unpacking objects: 100% (3/3), done.
Enforcing Policies...
(refs/heads/master) (8338c5) (c5b616)
[POLICY] Your message is not formatted correctly
error: hooks/update exited with error code 1
error: hook declined to update refs/heads/master
To git@gitserver:project.git
 ! [remote rejected] master -> master (hook declined)
error: failed to push some refs to 'git@gitserver:project.git'

There are a couple of interesting things here. First, you see this where the hook starts running.

Enforcing Policies...
(refs/heads/master) (fb8c72) (c56860)

Remember that you printed that out at the very beginning of your update script. Anything your script echoes to stdout will be transferred to the client.

The next thing you’ll notice is the error message.

[POLICY] Your message is not formatted correctly
error: hooks/update exited with error code 1
error: hook declined to update refs/heads/master

The first line was printed out by you, the other two were Git telling you that the update script exited non-zero and that is what is declining your push. Lastly, you have this:

To git@gitserver:project.git
 ! [remote rejected] master -> master (hook declined)
error: failed to push some refs to 'git@gitserver:project.git'

You’ll see a remote rejected message for each reference that your hook declined, and it tells you that it was declined specifically because of a hook failure.

Furthermore, if someone tries to edit a file they don’t have access to and push a commit containing it, they will see something similar. For instance, if a documentation author tries to push a commit modifying something in the lib directory, they see

[POLICY] You do not have access to push to lib/test.rb

From now on, as long as that update script is there and executable, your repository will never have a commit message without your pattern in it, and your users will be sandboxed.