| dgit(1) | dgit | dgit(1) |
dgit - 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] pbuilder|cowbuilder
[debbuildopts]
dgit [dgit-opts] push|push-built
[dgit-opts] [suite]
dgit [dgit-opts] rpush|rpush-built
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. Conversely, it allows Debian to publish the source of its packages as git branches, in a format which is directly useable by ordinary people.
This is the command line reference. Please read the tutorial(s):
| dgit-user(7) | for users: edit, build and share packages |
| dgit-nmu-simple(7) | for DDs: do a straightforward NMU |
| dgit-maint-native(7) | for maintainers of Debian-native packages |
| dgit-maint-debrebase(7) | for maintainers: a pure-git rebasish workflow |
| dgit-maint-merge(7) | for maintainers: a pure-git merging workflow |
| dgit-maint-gbp(7) | for maintainers already using git-buildpackage |
| dgit-sponsorship(7) | for sponsors and sponsored contributors |
| dgit-downstream-dsc(7) | setting up dgit push for a new distro |
See dgit(7) for detailed information about the data model, common problems likely to arise with certain kinds of package, etc.
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.
suite may be a combination of several underlying suites in the form mainsuite,subsuite...; see COMBINED SUITES in dgit(7).
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. (See also dgit update-vcs-git.)
If the branch does not exist, dgit checkout creates it, and sets it up the same way as dgit clone would. In that case, if the archive remote tracking branch does not exist, dgit checkout will do a dgit fetch first.
NB: dgit checkout will only do a fetch if it has to. If you already have the suite branch, and want to merge your branch with updates from the archive, use dgit pull.
dgit checkout will normally need to access the archive server, to canonicalise the provided suite name. The exception is if you specify the canonical name, and the branch (or tracking branch) already exists.
Tagging, signing and actually uploading should be left to dgit push.
dgit's build operations access the network, to get the -v option right. See -v, below.
Tagging, signing and actually uploading should be left to dgit push-source, or dgit push.
By default, the Vcs-Git field of the .dsc from Debian sid is used, as that is probably most up to date. Another suite may be specified, or . to indicate that the Vcs-Git of the cwd's debian/control should be used instead.
You should ensure that your dgit --build-products-dir setting matches your pbuilder --buildresult.
The debbuildopts are passed to pbuilder using its --debbuildopts option. If you want to pass other options to pbuilder, use the --pbuilder: dgit option as described below (remember that dgit options should appear between dgit and pbuilder).
You should ensure that in your pbuilderrc you do not have the setting SOURCE_ONLY_CHANGES=yes as this may cause trouble.
By default this uses --quilt=gbp, so HEAD should be a git-buildpackage style branch, not a patches-applied branch.
Tagging, signing and actually uploading should be left to dgit push.
This is the usual way to upload to Debian. It is like saying "update the source code in the archive to match my git HEAD, and let the autobuilders do the rest."
In more detail: dgit push-source builds a source package from HEAD. It then pushes the HEAD to the suite's dgit-repos branch, adjusts the .changes to include any .origs which the archive lacks and exclude .origs which the archive has (so -sa and -sd are not needed when building for dgit push), 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.
When used on a git-debrebase branch, dgit calls git-debrebase to prepare the branch for source package upload and push.
With -C, dgit push-source performs a dgit push-built, additionally ensuring that no binary packages are uploaded.
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 (e.g. 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.
dgit will check that the .dsc corresponds exactly to the current HEAD, ensuring that all users, whether of the dgit git view, or of the traditional archive, see the same source package.
| 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-source or dgit push-built 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).
The behaviour of dgit push is controlled by the dgit.default.push-subcmd git config option:
| source | runs dgit push-source | future default |
| built | and runs dgit push-built | |
| built,warn | warns, and runs dgit push-built | current default |
| reject | fails |
For dgit rpush, the behaviour is controlled by dgit.default.rpush-subcmd, falling back to dgit.default.push-subcmd if that is not set. Because dgit rpush is not typically run in a git working tree, only global git config options (and -c command line options) are relevant.
These settings can safely be passed to older dgit (via -c); the value built will be supported indefinitely. This should be used in scripts that need to work with both old versions of dgit (that don't have push-built) and new versions (where push-source is the default).
If there is an existing macro attribute line [attr]dgit-defuse-attrs in .git/info/attributes, but it is insufficient, because it was made by an earlier version of dgit and git has since introduced new transforming attributes, this modifies the macro to disable the newer transformations.
(If there is already a macro attribute line [attr]dgit-defuse-attrs in .git/info/attributes which does what dgit requires (whatever files it effects), this operation does nothing further. This fact can be used to defeat or partially defeat dgit setup-gitattributes and hence dgit setup-new-tree.)
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.
When used with a git-debrebase branch, dgit will ask git-debrebase to prepare patches. However, dgit can make patches in some situations where git-debrebase fails, so dgit quilt-fixup can be useful in its own right. To always use dgit's own patch generator instead of git-debrebase make-patches, pass --git-debrebase=true to dgit.
See FORMAT 3.0 (QUILT) in dgit(7).
This does about half the work of dgit fetch: it will convert the .dsc into a new, orphan git branch. Since dgit has no access to a corresponding source package archive or knowledge of the history it does not consider whether this version is newer than any previous import or corresponding git branches; and it therefore does not make a pseudomerge to bind the import into any existing git history.
Because a .dsc can contain a Dgit field naming a git commit (which you might not have), and specifying where to find that commit (and any history rewrite table), import-dsc might need online access. If this is a problem (or dgit's efforts to find the commit fail), consider --no-chase-dsc-distro or --force-import-dsc-with-dgit-field.
There is only one sub-option:
--require-valid-signature causes dgit to insist that the signature on the .dsc is valid (using the same criteria as dpkg-source -x). Otherwise, dgit tries to verify the signature but the outcome is reported only as messages to stderr.
If branch is prefixed with + then if it already exists, it will be simply overwritten, no matter its existing contents. If branch is prefixed with .. then if it already exists and dgit actually imports the dsc (rather than simply reading the git commit out of the Dgit field), dgit will make a pseudomerge so that the result is necessarily fast forward from the existing branch. Otherwise, if branch already exists, dgit will stop with an error message.
If branch does not start with refs/, refs/heads/ is prepended.
To make this operate off-line, the access configuration key which is used to determine the build-products-dir is the uncanonicalised version of the suite name from the changelog, or (of course) dgit.default.build-products-dir. See ACCESS CONFIGURATION, below.
This function is primarily provided for the benefit of git-debrebase.
This will delete all files which are not tracked by git. (Including any files you forgot to git add.)
--clean=... options other than dpkg-source are useful when the package's clean target is troublesome, or to avoid needing the build-dependencies.
dgit will only actually clean the tree if it needs to (because it needs to build the source package or binaries from your working tree). Otherwise it will just check that there are no untracked unignored files. See --clean=git[-ff],always, below.
With ,ignores or -wci, untracked files covered by .gitignore are tolerated, so only files which show up as ? in git status (ie, ones you maybe forgot to git add) are treated as a problem.
Without the extra d, requires the package's build dependencies.
With ...-d or -wdd, the build-dependencies are not checked (due to passing -d to dpkg-buildpackage), which violates policy, but may work in practice.
The rules clean target will only be run if it is needed: when dgit is going to build source or binary packages from your working tree, rather than from your git branch (for example because of --include-dirty or because the binary package build uses your working tree).
In all cases, dgit will check that there are (after rules clean, if applicable) no untracked un-ignored files, in case these are files you forgot to git add. (Except that this check is not done for a `3.0 (quilt)' package when dgit has to apply patches, dirtily, to the working tree.) If your package does not have a good .gitignore you will probably need --clean=dpkg-source,no-check aka -wdn.
Note that dgit may be unable to access the git history for an entirely new package which has not been accepted by the archive. So for an entirely new package you need to properly coordinate with anyone else who might upload.
Note that this does not prevent dgit from cleaning your tree, so if the changes in your working tree are in the form of untracked files, those might still be deleted, especially with --clean=git. If you want to include untracked files in the build, you can use --clean=none or --clean=dpkg-source[-d] in addition to --include-dirty. Note that this combination can fail if the untracked files are under debian/patches/.
It is safer not to specify previous-version, and usually it's not needed. Just say --overwrite, unless you know what you are doing.
This option is useful if you are the maintainer, and you have incorporated NMU changes into your own git workflow in a way that doesn't make your branch a fast forward from the NMU. It can also be useful when there was an upload made without dgit since the most recent upload made with dgit.
This option is also usually necessary the first time a package is pushed with dgit push to a particular suite. See dgit-maint-*(7).
If previous-version is not specified, dgit will check that the version in the archive is mentioned in your debian/changelog. (This will avoid losing changes, even with --overwrite, unless someone committed to git a finalised changelog entry, and then made later changes to that version.) If previous-version is specified, it ought to be the version currently in the archive.
dgit push --overwrite will, if necessary, make a pseudo-merge (that is, something that looks like the result of git merge -s ours) to stitch the archive's version into your own git history, so that your push is a fast forward from the archive.
(In quilt mode gbp, dpm, unpatched or baredebian*, implying a split between the dgit view and the maintainer view, the pseudo-merge will appear only in the dgit view.)
WARNING: If the maintainer responds by cancelling your upload from the queue, and does not make an upload of their own, this will not rewind the git branch on the dgit git server. Other dgit users will then see your push (with a warning message from dgit) even though the maintainer wanted to abolish it. Such users might unwittingly reintroduce your changes.
If this situation arises, someone should make a suitable dgit push to update the contents of dgit-repos to a version without the controversial changes.
For most operations (such as fetch and pull), disabling chasing means dgit will access only the git server for the distro you are directly working with, even if the .dsc was copied verbatim from another distro. For import-dsc, disabling chasing means dgit will work completely offline.
Disabling chasing can be hazardous: if the .dsc names a git commit which has been rewritten by those in charge of the distro, this option may prevent that rewrite from being effective. Also, it can mean that dgit fails to find necessary git commits.
This option is effective only with the following operations: quilt-fixup; push; all builds. And it is only effective when split view is actually in operation.
If ref does not start with refs/ it is taken to be a branch - i.e. refs/heads/ is prepended.
--dgit-view-save is a deprecated alias for --save-dgit-view.
When pushing to Debian, use this only 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; or, exceptionally, for the very first upload with dgit.
When split view is in operation, this also prevents the construction by dgit of a pseudomerge to make the dgit view fast forwarding. Normally only one of --overwrite (which creates a suitable pseudomerge) and --deliberately-not-fast-forward (which suppresses the pseudomerge and the fast forward checks) should be needed; --overwrite is usually better.
HEAD should be a series of plain commits (not touching debian/patches/), and pseudomerges, with as ancestor a patches-applied branch.
(If HEAD has any in-tree patches already, they must apply cleanly. This will be the case for any trees produced by dgit fetch or clone; if you do not change the upstream version nor make changes in debian/patches, it will remain true.)
Use this instead of the single-debian-patch dpkg-source format option. That dpkg-source option cannot handle certain changes to the tree that dpkg-source otherwise permits, and in some cases it can generate strange source packages that dpkg-source appears to accept but which become corrupted when people later try to modify them.
These quilt modes are known as splitting quilt modes. See --split-view, below.
--gbp (short for --quilt=gbp) is for use with git-buildpackage. Your HEAD is expected to be a patches-unapplied git branch, except that it might contain changes to upstream .gitignore files. This is the default for dgit gbp-build.
--dpm (short for --quilt=dpm) is for use with git-dpm. Your HEAD is expected to be a patches-applied git branch, except that it might contain changes to upstream .gitignore files.
--quilt=unapplied specifies that your HEAD is a patches-unapplied git branch (and that any changes to upstream .gitignore files are represented as patches in debian/patches).
--quilt=baredebian (or its alias --quilt=baredebian+git) specifies that your HEAD contains only a debian/ directory, with any changes to upstream files represented as patches in debian/patches. The upstream source must be available in git, by default, in a suitably named git tag; see --upstream-commitish. In this mode, dgit cannot check that all edited upstream files are properly represented as patches: dgit relies on debian/patches being correct.
--quilt=baredebian+tarball is like --quilt=baredebian, but is used when there is no appropriate upstream git history. To construct the dgit view, dgit will import your orig tarballs' contents into git. In this mode, dgit cannot check that the upstream parts of your upload correspond to what you intend: dgit relies on the right orig tarball(s) existing, and debian/patches being correct.
With --quilt=gbp|dpm|unapplied|baredebian*, dgit push (or precursors like quilt-fixup and build) will automatically generate a conversion of your git branch into the right form. dgit push will push the dgit-compatible form (the dgit view) to the dgit git server. The dgit view will be visible to you in the dgit remote tracking branches, but your own branch will not be modified. dgit push will create a tag debian/version for the maintainer view, and the dgit tag archive/debian/version for the dgit view. dgit quilt-fixup will merely do some checks, and cache the maintainer view.
If you have a branch like this it is essential to specify the appropriate --quilt= option! This is because it is not always possible to tell: a patches-unapplied git branch of a package with one patch, for example, looks very like a patches-applied branch where the user has used git revert to undo the patch, expecting to actually revert it. However, if you fail to specify the right --quilt option, and you aren't too lucky, dgit will notice the problem and stop, with a useful hint.
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.
When split view is in operation dgit will not make or merge any commits onto your own branch. Specifically, only the dgit view will contain dgit's pseudomerges, which bring into the git history previous uploads made with dgit push, and any commits in debian/patches required to make a correct `3.0 (quilt)' source package.
auto is the default, and splits the view only when needed: i.e., when you are working with a `3.0 (quilt)' source package and a splitting quilt mode: --[quilt=]gbp, dpm, unpatched or baredebian*.
always splits the view regardless of the source format and the quilt mode.
never will cause dgit to fail if split view is needed.
When split view is in operation, the dgit view is visible in your local git clone, but only in refs specific to dgit: notably remotes/dgit/dgit/suite and archive/distro/version.
Note that split view does not affect dgit fetch, and is not compatible with dgit pull.
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 ..).
Note that dgit push-source will always find the right .changes, regardless of this option.
By default, dgit uses the parent directory (..).
Changing this setting may necessitate moving .orig tarballs to the new directory, so it is probably best to use the dgit.default.build-products-dir configuration setting (see CONFIGURATION, below) which this command line option overrides).
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).
Options which are safe to pass include -C (and also -si -sa -sd although these should never be necessary with Debian since dgit automatically calculates whether .origs need to be uploaded.)
For other options the caveat below applies.
Use of this ability should not normally be necessary. It is provided for working around bugs, or other unusual situations. If you use these options, you may violate dgit's assumptions about the behaviour of its subprograms and cause lossage.
For dpkg-buildpackage, dpkg-genchanges, mergechanges and sbuild, the option applies only when the program is invoked directly by dgit. Usually, for passing options to dpkg-genchanges, you should use --ch:option.
Specifying --git is 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 below 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.
Any options or arguments exactly identical to option are removed. (It is not an error if there were none.)
This can only be used to delete options which are always passed by default by dgit, or to undo a previous --program:option. It cannot be used to override option(s) dynamically decided on by dgit.
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=).
gbp-build's value is used instead of gbp build or git-buildpackage. (The default is the latter unless the former exists on PATH.) gbp-pq's value is used instead of gbp pq. In both cases, unusually, the specified value is split on whitespace to produce a command and possibly some options and/or arguments.
For pbuilder and cowbuilder, the defaults are sudo -E pbuilder and sudo -E cowbuilder respectively. Like with gbp-build and gbp pq, the specified value is split on whitespace.
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.
This is not a very good simulation. It can easily go wrong in ways that a for-real push wouldn't.
Using this will make unsigned tags, and possibly other local changes, that will get in the way of a for-real push. So be prepared to burn the version number you're using.
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:
suite may be a glob pattern.
This is only used if no /usr/share/distro-info/somedistro.csv mentions the specified suite.
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.
There should be a `dgit rebase-prep' command or some such to turn a fast-forwarding branch containing pseudo-merges back into a rebasing patch stack. It might have to leave a note for a future dgit push.
If the dgit push fails halfway through, it is not necessarily restartable and idempotent. It would be good to check that the proposed signing key is available before starting work.
dgit's build functions, and dgit push, may make 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.
--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. --damp-run is likely to work much better.
dgit(7), dgit-*(7), curl(1), dput(1),
debsign(1), git-config(1), git-buildpackage(1),
dpkg-buildpackage(1),
https://browse.dgit.debian.org/
| Debian Project |