dgit(1) git integration with the Debian archive


dgit [dgit-opts] clone [dgit-opts] package [suite] [./dir|/dir]
dgit [dgit-opts] fetch|pull [dgit-opts] [suite]
dgit [dgit-opts] build|sbuild|build-source [build-opts]
dgit [dgit-opts] push [dgit-opts] [suite]
dgit [dgit-opts] rpush build-host:build-dir [push args...]
dgit [dgit-opts] action ...


dgit allows you to treat the Debian archive as if it were a git repository. See dgit(7) for detailed information about the data model, common problems likely to arise with certain kinds of package, etc.

The usual workflow is:
1.     dgit clone or fetch;

2.     make, do dev tests, and commit changes in git as desired;

3.     build packages for upload, using e.g. dgit sbuild

4.     do pre-upload tests of the proposed upload;

5.     dgit push.


dgit clone package [suite] [./dir|/dir]
Consults the archive and dgit-repos to construct the git view of history for package in suite (sid by default) in a new directory (named ./package by default); also, downloads any necessary orig tarballs.

The suite's git tip is left on the local branch dgit/suite ready for work, and on the corresponding dgit remote tracking branch. The origin remote will be set up to point to the package's dgit-repos tree for the distro to which suite belongs.

For your convenience, the vcs-git remote will be set up from the package's Vcs-Git field, if there is one - but note that in the general case the history found there may be different to or even disjoint from dgit's view.

dgit fetch [suite]
Consults the archive and git-repos to update the git view of history for a specific suite (and downloads any necessary orig tarballs), and updates the remote tracking branch remotes/dgit/dgit/suite. If the current branch is dgit/suite then dgit fetch defaults to suite; otherwise it parses debian/changelog and uses the suite specified there.
dgit pull [suite]
Does dgit fetch, and then merges the new head of the remote tracking branch remotes/dgit/dgit/suite into the current branch.
dgit build ...
Runs dpkg-buildpackage with some suitable options. Options and arguments after build will be passed on to dpkg-buildpackage. It is not necessary to use dgit build when using dgit; it is OK to use any approach which ensures that the generated source package corresponds to the relevant git commit.

Tagging, signing and actually uploading should be left to dgit push.

dgit build-source ...
Builds the source package, and a changes file for a prospective source-only upload, using dpkg-source. The output is left in package_version.dsc and package_version_source.changes.

Tagging, signing and actually uploading should be left to dgit push.

dgit clean
Cleans the current working tree (according to the --clean= option in force).
dgit help
Print a usage summary.
dgit sbuild ...
Constructs the source package, uses sbuild to do a binary build, and uses mergechanges to merge the source and binary changes files. Options and arguments after sbuild will be passed on to sbuild. Changes files matching package_version_*.changes in the parent directory will be removed; the output is left in package_version_multi.changes.

Tagging, signing and actually uploading should be left to dgit push.

dgit gbp-build ...
Runs git-buildpackage with some suitable options. Options and arguments after gbp-build will be passed on to git-buildpackage.

Tagging, signing and actually uploading should be left to dgit push.

dgit push [suite]
Does an `upload', pushing the current HEAD to the archive (as a source package) and to dgit-repos (as git commits). The package must already have been built ready for upload, with the .dsc and .changes left in the parent directory. It is normally best to do the build with dgit too (eg with dgit sbuild): some existing build tools pass unhelpful options to dpkg-source et al by default, which can result in the built source package not being identical to the git tree.

In more detail: dgit push checks that the current HEAD corresponds to the .dsc. It then pushes the HEAD to the suite's dgit-repos branch, makes a signed git tag, edits the .dsc to contain the dgit metadata field, runs debsign to sign the upload (.dsc and .changes), pushes the signed tag, and finally uses dput to upload the .changes to the archive.

dgit push always uses the package, suite and version specified in the debian/changelog and the .dsc, which must agree. If the command line specifies a suite then that must match too.

If dgit push fails while uploading, it is fine to simply retry the dput on the .changes file at your leisure.

dgit rpush build-host:build-dir [push args...]
Pushes the contents of the specified directory on a remote machine. This is like running dgit push on build-host with build-dir as the current directory; however, signing operations are done on the invoking host. This allows you to do a push when the system which has the source code and the build outputs has no access to the key:

1.      Clone on build host (dgit clone)

2.     Edit code on build host (edit, git commit)

3.     Build package on build host (dgit build)

4.     Test package on build host or elsewhere (dpkg -i, test)

5.     Upload by invoking dgit rpush on host with your GPG key.

However, the build-host must be able to ssh to the dgit repos. If this is not already the case, you must organise it separately, for example by the use of ssh agent forwarding.

The remaining arguments are treated just as dgit push would handle them.

build-host and build-dir can be passed as separate arguments; this is assumed to be the case if the first argument contains no : (except perhaps one in [ ], to support IPv6 address literals).

You will need similar enough versions of dgit on the build-host and the invocation host. The build-host needs gnupg installed, with your public key in its keyring (but not your private key, obviously).

dgit setup-new-tree
Configure the current working tree the way that dgit clone would have set it up. Like running dgit setup-useremail and setup-mergechangelogs (but only does each thing if dgit is configured to do it automatically). You can use these in any git repository, not just ones used with the other dgit operations.
dgit setup-useremail
Set the working tree's user.name and user.email from the distro-specific dgit configuration (dgit-distro.distro.user-name and .user-email), or DEBFULLNAME or DEBEMAIL.
dgit setup-mergechangelogs
Configures a git merge helper for the file debian/changelog which uses dpkg-mergechangelogs.
dgit quilt-fixup
`3.0 (quilt)' format source packages need changes representing not only in-tree but also as patches in debian/patches. dgit quilt-fixup checks whether this has been done; if not, dgit will make appropriate patches in debian/patches and also commit the resulting changes to git.

This is normally done automatically by dgit build and dgit push.

dgit will try to turn each relevant commit in your git history into a new quilt patch. dgit cannot convert nontrivial merges, or certain other kinds of more exotic history. If dgit can't find a suitable linearisation of your history, by default it will fail, but you can ask it to generate a single squashed patch instead.

dgit version
Prints version information and exits.
dgit clone-dgit-repos-server destdir
Tries to fetch a copy of the source code for the dgit-repos-server, as actually being used on the dgit git server, as a git tree.


--dry-run | -n
Go through the motions, fetching all information needed, but do not actually update the output(s). For push, dgit does the required checks and leaves the new .dsc in a temporary file, but does not sign, tag, push or upload.
--damp-run | -L
Go through many more of the motions: do everything that doesn't involve either signing things, or making changes on the public servers.
Use keyid for signing the tag and the upload. The default comes from the distro's keyid config setting (see CONFIGURATION, below), or failing that, the uploader trailer line in debian/changelog.
does not sign tags or uploads (meaningful only with push).
Specifies that we should process source package package rather than looking in debian/control or debian/changelog. Valid with dgit fetch and dgit pull, only.
--clean=git | -wg
The source tree should be cleaned, before building a source package with one of the build options, using git clean -xdf. This will delete all files which are not tracked by git. Also, -wg causes dgit to pass -nc to dpkg-buildpackage, which prevents the package's own clean target from being run.

--clean=git is useful when the package's clean target is troublesome; the downside is simply that git clean may delete files you forgot to git add. --clean=git can also avoid needing the build-dependencies.

--clean=git-ff | -wgf
The source tree should be cleaned, before building a source package with one of the build options, using git clean -xdff. This is like "git clean -xdf" but it also removes any subdirectories containing different git trees (which only unusual packages are likely to create).
--clean=check | -wc
Merely check that the tree is clean (does not contain uncommitted files), before building a source package.
--clean=none | -wn
Do not clean the tree before building a source package. If there are files which are not in git, or if the build creates such files, a subsequent dgit push will fail.
--clean=dpkg-source | -wd
Use dpkg-buildpackage to do the clean, so that the source package is cleaned by dpkg-source running the package's clean target. This is the default. It requires the package's build dependencies.
--clean=dpkg-source-d | -wdd
Use dpkg-buildpackage -d to do the clean, so that the source package is cleaned by dpkg-source running the package's clean target. The build-dependencies are not checked (due to -d), which violates policy, but may work in practice.
-N | --new
The package is or may be new in this suite. Without this, dgit will refuse to push. It may (for Debian, will) be unable to access the git history for any packages which have been newly pushed and have not yet been published.
Do not complain if the working tree does not match your git HEAD. This can be useful with build, if you plan to commit later. (dgit push will still ensure that the .dsc you upload and the git tree you push are identical, so this option won't make broken pushes.)
Declare that you are deliberately doing something. This can be used to override safety catches, including safety catches which relate to distro-specific policies. The meanings of somethings understood in the context of Debian are discussed below:
Declare that you are deliberately rewinding history. When pushing to Debian, use this when you are making a renewed upload of an entirely new source package whose previous version was not accepted for release from NEW because of problems with copyright or redistributibility.
Declare that you are deliberately including, in the git history of your current push, history which contains a previously-submitted version of this package which was not approved (or has not yet been approved) by the ftpmasters. When pushing to Debian, only use this option after verifying that: none of the rejected-from-NEW (or never-accepted) versions in the git history of your current push, were rejected by ftpmaster for copyright or redistributability reasons.
Declare that you are deliberately rewinding history and want to throw away the existing repo. Not relevant when pushing to Debian, as the Debian server will do this automatically when necessary.
When fixing up source format `3.0 (quilt)' metadata, insist on generating a linear patch stack. If such a stack cannot be generated, fail. This is the default for Debian.
When fixing up source format `3.0 (quilt)' metadata, prefer to generate a linear patch stack, but if that doesn't seem possible, generate a single squashed patch for all the changes made in git. This is not a good idea for an NMU in Debian.
When fixing up source format `3.0 (quilt)' metadata, generate a single squashed patch for all the changes made in git. This is not a good idea for an NMU in Debian.
Check whether source format `3.0 (quilt)' metadata would need fixing up, but, if it does, fail. You must then fix the metadata yourself somehow before pushing. (NB that dpkg-source --commit will not work because the dgit git tree does not have a .pc directory.)
--quilt=nocheck | --no-quilt-fixup
Do not check whether up source format `3.0 (quilt)' metadata needs fixing up. If you use this option and the metadata did in fact need fixing up, dgit push will fail.
Prints debugging information to stderr. Repeating the option produces more output (currently, up to -DDDD is meaningfully different).
Specifies a git configuration option, to be used for this run. dgit itself is also controlled by git configuration options.
-vversion|_ | --since-version=version|_
Specifies the -vversion option to pass to dpkg-genchanges, during builds. Changes (from debian/changelog) since this version will be included in the built changes file, and hence in the upload. If this option is not specified, dgit will query the archive and use the latest version uploaded to the intended suite.

Specifying _ inhibits this, so that no -v option will be passed to dpkg-genchanges (and as a result, only the last stanza from debian/changelog will be used for the build and upload).

Passed to dpkg-genchanges (eventually).
Specifies a single additional option to pass, eventually, to dpkg-genchanges.
--curl=program | --dput=program |...
Specifies alternative programs to use instead of curl, dput, debsign, dpkg-source, dpkg-buildpackage, dpkg-genchanges, sbuild, gpg, ssh, dgit, git, or mergechanges.

For dpkg-buildpackage, dpkg-genchanges, mergechanges and sbuild, this applies only when the program is invoked directly by dgit.

For dgit, specifies the command to run on the remote host when dgit rpush needs to invoke a remote copy of itself. (dgit also reinvokes itself as the EDITOR for dpkg-source --commit; this is done using argv[0], and is not affected by --dgit=).

For ssh, the default value is taken from the DGIT_SSH or GIT_SSH environment variables, if set (see below). And, for ssh, when accessing the archive and dgit-repos, this command line setting is overridden by the git config variables dgit-distro.distro.ssh and .dgit.default.ssh (which can in turn be overridden with -c). Also, when dgit is using git to access dgit-repos, only git's idea of what ssh to use (eg, GIT_SSH) is relevant.

--curl:option | --dput:option |...
Specifies a single additional option to pass to curl, dput, debsign, dpkg-source, dpkg-buildpackage, dpkg-genchanges, sbuild, ssh, dgit, or mergechanges. Can be repeated as necessary.

For dpkg-buildpackage, dpkg-genchanges, mergechanges and sbuild, this applies only when the program is invoked directly by dgit. Usually, for passing options to dpkg-genchanges, you should use --ch:option.

Specifying --git not effective for some lower-level read-only git operations performed by dgit, and also not when git is invoked by another program run by dgit.

See notes above regarding ssh and dgit.

NB that --gpg:option is not supported (because debsign does not have that facility). But see -k and the keyid distro config setting.

-ddistro | --distro=distro
Specifies that the suite to be operated on is part of distro distro. This overrides the default value found from the git config option dgit-suite.suite.distro. The only effect is that other configuration variables (used for accessing the archive and dgit-repos) used are dgit-distro.distro.*.

If your suite is part of a distro that dgit already knows about, you can use this option to make dgit work even if your dgit doesn't know about the suite. For example, specifying -ddebian will work when the suite is an unknown suite in the Debian archive.

To define a new distro it is necessary to define methods and URLs for fetching (and, for dgit push, altering) a variety of information both in the archive and in dgit-repos. How to set this up is not yet documented.

Specifies the .changes file which is to be uploaded. By default dgit push looks for single .changes file in the parent directory whose filename suggests it is for the right package and version - or, if there is a _multi.changes file, dgit uses that.

If the specified changesfile pathname contains slashes, the directory part is also used as the value for --build-products-dir; otherwise, the changes file is expected in that directory (by default, in ..).

Specifies where to find the built files to be uploaded. By default, dgit looks in the parent directory (..).
dgit push needs to canonicalise the suite name. Sometimes, dgit lacks a way to ask the archive to do this without knowing the name of an existing package. Without --new we can just use the package we are trying to push. But with --new that will not work, so we guess dpkg or use the value of this option. This option is not needed with the default mechanisms for accessing the archive.
Print a usage summary.
dgit rpush uses a temporary directory on the invoking (signing) host. This option causes dgit to use directory instead. Furthermore, the specified directory will be emptied, removed and recreated before dgit starts, rather than removed after dgit finishes. The directory specified must be an absolute pathname.
Do not delete the destination directory if clone fails.


It is always possible with dgit to clone or fetch a package, make changes in git (using git-commit) on the suite branch (git checkout dgit/suite) and then dgit push. You can use whatever gitish techniques you like to construct the commits to push; the only requirement is that what you push is a descendant of the state of the archive, as provided by dgit in the remote tracking branch remotes/dgit/dgit/suite.

If you are using dgit to do an NMU (in Debian), and don't know about the maintainers' preferred packaging workflows, you should make your changes as a linear series of (logicially separated) commits on top of what's already in the archive.

If you are lucky the other uploaders have also used dgit and integrated the other relevant git history; if not you can fetch it into your tree and cherry-pick etc. as you wish.


If you are the maintainer of a package dealing with uploads made without dgit, you will probably want to merge the synthetic commits (made by dgit to represent the uploads) into your git history. Normally you can just merge the dgit branch into your own master, or indeed if you do your work on the dgit local suite branch dgit/suite you can just use dgit pull.

However the first time dgit is used it will generate a new origin commit from the archive which won't be linked into the rest of your git history. You will need to merge this.

If last upload was in fact made with git, you should usually proceed as follows: identify the commit which was actually used to build the package. (Hopefully you have a tag for this.) Check out the dgit branch (git checkout dgit/suite) and merge that other commit (git merge debian/version). Hopefully this merge will be trivial because the two trees should be very similar. The resulting branch head can be merged into your working branches (git checkout master && git merge dgit/suite).

If last upload was not made with git, a different approach is required to start using dgit. First, do dgit fetch (or clone) to obtain a git history representation of what's in the archive and record it in the remotes/dgit/dgit/suite tracking branch. Then somehow, using your other git history plus appropriate diffs and cherry picks from the dgit remote tracking branch, construct a git commit whose tree corresponds to the tree to use for the next upload. If that commit-to-be-uploaded is not a descendant of the dgit remote tracking branch, check it out and say git merge -s ours remotes/dgit/dgit/suite; that tells git that we are deliberately throwing away any differences between what's in the archive and what you intend to upload. Then run dgit push to actually upload the result.


dgit can be configured via the git config system. You may set keys with git-config (either in system-global or per-tree configuration), or provide -ckey=value on the dgit command line.

Settings likely to be useful for an end user include:

dgit-suite.suite.distro distro
Specifies the distro for a suite. dgit keys off the suite name (which appears in changelogs etc.), and uses that to determine the distro which is involved. The config used is thereafter that for the distro.
dgit.default.distro distro
The default distro for an unknown suite.
for each dgit-distro.distro.*, the default value used if there is no distro-specific setting.
One of the values for the command line --clean= option; used if --clean is not specified.
dgit-distro.distro.readonly auto|a | true|t|y|1 | false|f|n|0
Whether you have push access to the distro. For Debian, it is OK to use auto, which uses readonly mode if you are not pushing right now; but, setting this to false will avoid relying on the mirror of the dgit git repository server.
See also -k.
dgit-distro.distro.mirror url
Not relevant for Debian.
Might be useful if you have an intermediate queue server.
dgit-distro.distro.user-name dgit-distro.distro.user-email
Values to configure for user.name and user.email in new git trees. If not specified, the DEBFULLNAME and DEBEMAIL environment variables are used, respectively. Only used if .setup-usermail is not disabled.
Whether to set user.name and user.email in new git trees. True by default. Ignored for dgit setup-setup-useremail, which does it anyway.
Whether to setup a merge driver which uses dpkg-mergechangelogs for debian/changelog. True by default. Ignored for dgit setup-mergechangelogs, which does it anyway.
Program to use instead of cmd. Works like --cmd=... .
Extra options to pass to cmd. Works like --cmd:... . To pass several options, configure multiple values in git config (with git config --add). The options for dgit.default.opts-cmd dgit-distro.distro/push.opts-cmd and are all used, followed by options from dgit's command line.


There are many other settings which specify how a particular distro's services (archive and git) are provided. These should not normally be adjusted, but are documented for the benefit of distros who wish to adopt dgit.
If set, overrides corresponding non /push config when readonly=false, or when pushing and readonly=auto.
dgit-distro.distro.git-check true|false|url|ssh-cmd
dgit-distro.distro.diverts.divert new-distro|/distro-suffix
dgit-distro.distro.git-create ssh-cmd|true
dgit-distro.distro.archive-query ftpmasterapi: | madison:distro | dummycat:/path | sshpsql:user@host:dbname


specify an alternative default program (and perhaps arguments) to use instead of ssh. DGIT_SSH is consulted first and may contain arguments; if it contains any whitespace will be passed to the shell. GIT_SSH specifies just the program; no arguments can be specified, so dgit interprets it the same way as git does. See also the --ssh= and --ssh: options.
Default git user.email and user.name for new trees. See dgit setup-new-tree.
gpg, dpkg-..., debsign, git, curl, dput, LWP::UserAgent
and other subprograms and modules used by dgit are affected by various environment variables. Consult the documentaton for those programs for details.


dgit's git representation of format `3.0 (quilt)' source packages does not represent the patch stack as git commits. Currently the patch series representation cannot round trip between git and the archive. Ideally dgit would represent a quilty package with an origin commit of some kind followed by the patch stack as a series of commits followed by a pseudo-merge (to make the branch fast-forwarding). This would also mean a new `dgit rebase-prep' command or some such to turn such a fast-forwarding branch back into a rebasing patch stack, and a `force' option to dgit push (perhaps enabled automatically by a note left by rebase-prep) which will make the required pseudo-merge.

If the dgit push fails halfway through, it should be restartable and idempotent. However this is not true for the git tag operation. Also, it would be good to check that the proposed signing key is available before starting work.

dgit's handling of .orig.tar.gz is not very sophisticated. Ideally the .orig.tar.gz could be transported via the git repo as git tags. Doing this is made more complicated by the possibility of a `3.0 (quilt)' package with multiple .orig tarballs.

dgit's build functions, and dgit push, should not make any changes to your current HEAD. Sadly this is necessary for packages in the `3.0 (quilt)' source format. This is ultimately due to what I consider design problems in quilt and dpkg-source.

There should be an option which arranges for the `3.0 (quilt)' autocommit(s) to not appear on your HEAD, but instead only in the remote tracking suite branch.

--dry-run does not always work properly, as not doing some of the git fetches may result in subsequent actions being different. Doing a non-dry-run dgit fetch first will help.