Table of Contents
This chapter contains information related to creating, uploading, maintaining, and porting packages.
If you want to create a new package for the Debian distribution, you should first check the Work-Needing and Prospective Packages (WNPP) list. Checking the WNPP list ensures that no one is already working on packaging that software, and that effort is not duplicated. Read the WNPP web pages for more information.
Assuming no one else is already working on your prospective package, you must
then submit a bug report (Section 7.1, “Bug reporting”) against the
pseudo-package wnpp
describing your
plan to create a new package, including, but not limiting yourself to, a
description of the package, the license of the prospective package, and the
current URL where it can be downloaded from.
You should set the subject of the bug to ITP:
, substituting the name of the new
package for foo
-- short
description
foo
.
The severity of the bug report must be set to wishlist
.
Please send a copy to <debian-devel@lists.debian.org>
by using the X-Debbugs-CC
header (don't use CC:, because that way the message's subject won't
indicate the bug number). If you are packaging so many new packages (>10)
that notifying the mailing list in separate messages is too disruptive,
send a summary after filing the bugs to the debian-devel list instead.
This will inform the other developers about upcoming packages and will
allow a review of your description and package name.
Please include a Closes: #
entry in the changelog of the new package in order for the bug report to
be automatically closed once the new package is installed in the archive
(see Section 5.8.4, “When bugs are closed by new uploads”).
nnnnn
If you think your package needs some explanations for the administrators of the
NEW package queue, include them in your changelog, send to <ftpmaster@debian.org>
a reply to the email you receive as a maintainer after your upload, or reply to
the rejection email in case you are already re-uploading.
When closing security bugs include CVE numbers as well as the
Closes: #
.
This is useful for the security team to track vulnerabilities. If an upload is
made to fix the bug before the advisory ID is known, it is encouraged to modify
the historical changelog entry with the next upload. Even in this case, please
include all available pointers to background information in the original
changelog entry.
nnnnn
There are a number of reasons why we ask maintainers to announce their intentions:
It helps the (potentially new) maintainer to tap into the experience of people on the list, and lets them know if anyone else is working on it already.
It lets other people thinking about working on the package know that there already is a volunteer, so efforts may be shared.
It lets the rest of the maintainers know more about the package than the one
line description and the usual changelog entry ``Initial release'' that gets
posted to <debian-devel-changes@lists.debian.org>
.
It is helpful to the people who live off unstable
(and form
our first line of testers). We should encourage these people.
The announcements give maintainers and other interested parties a better feel of what is going on, and what is new, in the project.
Please see http://ftp-master.debian.org/REJECT-FAQ.html for common rejection reasons for a new package.
Changes that you make to the package need to be recorded in the
debian/changelog
. These changes should provide a concise
description of what was changed, why (if it's in doubt), and note if any bugs
were closed. They also record when the package was completed. This file will
be installed in
/usr/share/doc/
,
or
package
/changelog.Debian.gz/usr/share/doc/
for native packages.
package
/changelog.gz
The debian/changelog
file conforms to a certain structure,
with a number of different fields. One field of note, the
distribution
, is described in Section 5.5, “Picking a distribution”. More information about the structure of this file
can be found in the Debian Policy section titled
debian/changelog
.
Changelog entries can be used to automatically close Debian bugs when the package is installed into the archive. See Section 5.8.4, “When bugs are closed by new uploads”.
It is conventional that the changelog entry of a package that contains a new upstream version of the software looks like this:
* New upstream release.
There are tools to help you create entries and finalize the
changelog
for release — see Section A.6.1, “devscripts
”
and Section A.6.6, “dpkg-dev-el
”.
See also Section 6.3, “Best practices for debian/changelog
”.
Before you upload your package, you should do basic testing on it. At a minimum, you should try the following activities (you'll need to have an older version of the same Debian package around):
Install the package and make sure the software works, or upgrade the package from an older version to your new version if a Debian package for it already exists.
Run lintian over the package. You can run
lintian as follows: lintian -v
. This will check
the source package as well as the binary package. If you don't understand the
output that lintian generates, try adding the
package-version
.changes-i
switch, which will cause lintian to
output a very verbose description of the problem.
Normally, a package should not be uploaded if it causes
lintian to emit errors (they will start with E
).
For more information on lintian, see Section A.2.1, “lintian
”.
Optionally run debdiff (see Section A.2.2, “debdiff”) to analyze changes from an older version, if one exists.
Downgrade the package to the previous version (if one exists) — this tests
the postrm
and prerm
scripts.
Remove the package, then reinstall it.
Copy the source package in a different directory and try unpacking it and
rebuilding it. This tests if the package relies on existing files outside of
it, or if it relies on permissions being preserved on the files shipped inside
the .diff.gz
file.
There are two types of Debian source packages:
the so-called native
packages, where there is no
distinction between the original sources and the patches applied for Debian
the (more common) packages where there's an original source tarball file accompanied by another file that contains the changes made by Debian
For the native packages, the source package includes a Debian source control
file (.dsc
) and the source tarball
(.tar.{gz,bz2,xz}
). A source package of a non-native package
includes a Debian source control file, the original source tarball
(.orig.tar.{gz,bz2,xz}
) and the Debian changes
(.diff.gz
for the source format “1.0” or
.debian.tar.{gz,bz2,xz}
for the source format “3.0 (quilt)”).
With source format “1.0”, whether a package is native or not was determined
by dpkg-source at build time. Nowadays it is recommended
to be explicit about the desired source format by putting either “3.0 (quilt)”
or “3.0 (native)” in debian/source/format
.
The rest of this section relates only to non-native packages.
The first time a version is uploaded which corresponds to a particular upstream
version, the original source tar file should be uploaded and included in the
.changes
file. Subsequently, this very same tar file
should be used to build the new diffs and .dsc
files, and
will not need to be re-uploaded.
By default, dpkg-genchanges and
dpkg-buildpackage will include the original source tar file
if and only if the current changelog entry has a different upstream version
from the preceding entry. This behavior may be modified by using
-sa
to always include it or -sd
to always
leave it out.
If no original source is included in the upload, the original source tar-file
used by dpkg-source when constructing the
.dsc
file and diff to be uploaded
must be byte-for-byte identical with the one already in
the archive.
Please notice that, in non-native packages, permissions on files that are not
present in the *.orig.tar.{gz,bz2,xz}
will not be preserved, as diff does not store file
permissions in the patch. However when using source format “3.0 (quilt)”,
permissions of files inside the debian
directory are
preserved since they are stored in a tar archive.
Each upload needs to specify which distribution the package is intended for.
The package build process extracts this information from the first line of the
debian/changelog
file and places it in the
Distribution
field of the .changes
file.
There are several possible values for this field: stable
,
unstable
, testing-proposed-updates
and
experimental
. Normally, packages are uploaded into
unstable
.
Actually, there are two other possible distributions: stable-security
and testing-security
, but read
Section 5.8.5, “Handling security-related bugs” for more information on those.
It is not possible to upload a package into several distributions at the same time.
Uploading to stable
means that the package will transferred
to the proposed-updates-new
queue for review by the stable
release managers, and if approved will be installed in
stable-proposed-updates
directory of the Debian archive.
From there, it will be included in stable
with the next
point release.
To ensure that your upload will be accepted, you should discuss the changes
with the stable release team before you upload. For that, file a bug against
the release.debian.org
pseudo-package
using reportbug, including the patch you want to
apply to the package version currently in stable
. Always
be verbose and detailed in your changelog entries for uploads to the
stable
distribution.
Extra care should be taken when uploading to stable
.
Basically, a package should only be uploaded to stable
if
one of the following happens:
a truly critical functionality problem
the package becomes uninstallable
a released architecture lacks the package
In the past, uploads to stable
were used to address
security problems as well. However, this practice is deprecated, as uploads
used for Debian security advisories are automatically copied to the appropriate
proposed-updates
archive when the advisory is released.
See Section 5.8.5, “Handling security-related bugs” for detailed information on handling
security problems. If the security teams deems the problem to be too
benign to be fixed through a DSA
, the stable release
managers are usually willing to include your fix nonetheless in a regular
upload to stable
.
Changing anything else in the package that isn't important is discouraged, because even trivial fixes can cause bugs later on.
Packages uploaded to stable
need to be compiled on systems
running stable
, so that their dependencies are limited to
the libraries (and other packages) available in stable
;
for example, a package uploaded to stable
that depends on
a library package that only exists in unstable
will be
rejected. Making changes to dependencies of other packages (by messing with
Provides
or shlibs
files), possibly
making those other packages uninstallable, is strongly discouraged.
Uploads to the oldstable
distributions are possible as
long as it hasn't been archived. The same rules as for stable
apply.
Please see the information in the testing section for details.
To upload a package, you should upload the files (including the signed changes
and dsc-file) with anonymous ftp to ftp.upload.debian.org
in
the directory /pub/UploadQueue/.
To get the files processed there, they need to be signed with a key in the
Debian Developers keyring or the Debian Maintainers keyring
(see http://wiki.debian.org/DebianMaintainer).
Please note that you should transfer the changes file last. Otherwise, your upload may be rejected because the archive maintenance software will parse the changes file and see that not all files have been uploaded.
You may also find the Debian packages dupload or dput useful when uploading packages.These handy programs help automate the process of uploading packages into Debian.
For removing packages, please see ftp://ftp.upload.debian.org/pub/UploadQueue/README and the Debian package dcut.
It is sometimes useful to upload a package immediately, but to want this package to arrive in the archive only a few days later. For example, when preparing a Non-Maintainer Upload, you might want to give the maintainer a few days to react.
An upload to the delayed directory keeps the package in
the deferred uploads queue.
When the specified waiting time is over, the package is moved into
the regular incoming directory for processing.
This is done through automatic uploading to
ftp.upload.debian.org
in upload-directory
DELAYED/
(X
-dayX
between 0 and 15). 0-day is uploaded
multiple times per day to ftp.upload.debian.org
.
With dput, you can use the --delayed
parameter to put the package into one of the queues.
DELAY
Do NOT upload a package to the security
upload queue (oldstable-security
, stable-security
,
etc.) without prior authorization from the security team. If the
package does not exactly meet the team's requirements, it will cause many
problems and delays in dealing with the unwanted upload. For details, please
see Section 5.8.5, “Handling security-related bugs”.
There is an alternative upload queue in Europe at ftp://ftp.eu.upload.debian.org/pub/UploadQueue/. It operates in
the same way as ftp.upload.debian.org
, but should be faster
for European developers.
Packages can also be uploaded via ssh to
ssh.upload.debian.org
; files should be put
/srv/upload.debian.org/UploadQueue
. This queue does
not support delayed uploads.
The Debian archive maintainers are responsible for handling package uploads.
For the most part, uploads are automatically handled on a daily basis by the
archive maintenance tools, dak process-upload. Specifically,
updates to existing packages to the unstable
distribution are
handled automatically. In other cases, notably new packages, placing the
uploaded package into the distribution is handled manually. When uploads are
handled manually, the change to the archive may take some time to occur. Please
be patient.
In any case, you will receive an email notification indicating that the package has been added to the archive, which also indicates which bugs will be closed by the upload. Please examine this notification carefully, checking if any bugs you meant to close didn't get triggered.
The installation notification also includes information on what section the package was inserted into. If there is a disparity, you will receive a separate email notifying you of that. Read on below.
Note that if you upload via queues, the queue daemon software will also send you a notification by email.
The debian/control
file's Section
and
Priority
fields do not actually specify where the file will
be placed in the archive, nor its priority. In order to retain the overall
integrity of the archive, it is the archive maintainers who have control over
these fields. The values in the debian/control
file are
actually just hints.
The archive maintainers keep track of the canonical sections and priorities for
packages in the override file
. If there is a disparity
between the override file
and the package's fields as
indicated in debian/control
, then you will receive an
email noting the divergence when the package is installed into the archive.
You can either correct your debian/control
file for your
next upload, or else you may wish to make a change in the override
file
.
To alter the actual section that a package is put in, you need to first make
sure that the debian/control
file in your package is
accurate. Next, submit a
bug against ftp.debian.org
requesting
that the section or priority for your package be changed from the old section
or priority to the new one. Use a Subject like
override: PACKAGE1:section/priority, [...],
PACKAGEX:section/priority
, and include the justification for the
change in the body of the bug report.
For more information about override files
, see
dpkg-scanpackages(1) and http://www.debian.org/Bugs/Developer#maintincorrect.
Note that the Section
field describes both the section as
well as the subsection, which are described in Section 4.6.1, “Sections”. If the section is main, it should be omitted.
The list of allowable subsections can be found in http://www.debian.org/doc/debian-policy/ch-archive.html#s-subsections.
Every developer has to be able to work with the Debian bug tracking system. This includes knowing how to file bug reports properly (see Section 7.1, “Bug reporting”), how to update them and reorder them, and how to process and close them.
The bug tracking system's features are described in the BTS documentation for developers. This includes closing bugs, sending followup messages, assigning severities and tags, marking bugs as forwarded, and other issues.
Operations such as reassigning bugs to other packages, merging separate bug reports about the same issue, or reopening bugs when they are prematurely closed, are handled using the so-called control mail server. All of the commands available on this server are described in the BTS control server documentation.
If you want to be a good maintainer, you should periodically check the Debian bug tracking system (BTS) for
your packages. The BTS contains all the open bugs against your packages. You
can check them by browsing this page:
http://bugs.debian.org/
.
yourlogin
@debian.org
Maintainers interact with the BTS via email addresses at
bugs.debian.org
. Documentation on available
commands can be found at http://www.debian.org/Bugs/, or,
if you have installed the doc-debian
package, you can look at the local files /usr/share/doc/debian/bug-*
.
Some find it useful to get periodic reports on open bugs. You can add a cron job such as the following if you want to get a weekly email outlining all the open bugs against your packages:
# ask for weekly reports of bugs in my packages
0 17 * * fri echo "index maint address
" | mail request@bugs.debian.org
Replace address
with your official Debian maintainer
address.
When responding to bugs, make sure that any discussion you have about bugs is
sent both to the original submitter of the bug, and to the bug itself (e.g.,
<
). If you're writing a new mail and you
don't remember the submitter email address, you can use the
123
@bugs.debian.org><
email to contact the submitter
and to record your mail within the bug log (that means you
don't need to send a copy of the mail to 123
-submitter@bugs.debian.org><
).
123
@bugs.debian.org>
If you get a bug which mentions FTBFS, this means Fails to build from source. Porters frequently use this acronym.
Once you've dealt with a bug report (e.g. fixed it), mark it as
done
(close it) by sending an explanation message to
<
. If you're fixing a bug by changing
and uploading the package, you can automate bug closing as described in Section 5.8.4, “When bugs are closed by new uploads”.
123
-done@bugs.debian.org>
You should never close bugs via the bug server
close
command sent to <control@bugs.debian.org>
.
If you do so, the original submitter will not receive any information about why
the bug was closed.
As a package maintainer, you will often find bugs in other packages or have bugs reported against your packages which are actually bugs in other packages. The bug tracking system's features are described in the BTS documentation for Debian developers. Operations such as reassigning, merging, and tagging bug reports are described in the BTS control server documentation. This section contains some guidelines for managing your own bugs, based on the collective Debian developer experience.
Filing bugs for problems that you find in other packages is one of the civic obligations of maintainership, see Section 7.1, “Bug reporting” for details. However, handling the bugs in your own packages is even more important.
Here's a list of steps that you may follow to handle a bug report:
Decide whether the report corresponds to a real bug or not. Sometimes users are just calling a program in the wrong way because they haven't read the documentation. If you diagnose this, just close the bug with enough information to let the user correct their problem (give pointers to the good documentation and so on). If the same report comes up again and again you may ask yourself if the documentation is good enough or if the program shouldn't detect its misuse in order to give an informative error message. This is an issue that may need to be brought up with the upstream author.
If the bug submitter disagrees with your decision to close the bug, they may
reopen it until you find an agreement on how to handle it. If you don't find
any, you may want to tag the bug wontfix
to let people know
that the bug exists but that it won't be corrected. If this situation is
unacceptable, you (or the submitter) may want to require a decision of the
technical committee by reassigning the bug to tech-ctte
(you may use the clone command of the BTS
if you wish to keep it reported against your package). Before doing so, please
read the recommended
procedure.
If the bug is real but it's caused by another package, just reassign the bug to
the right package. If you don't know which package it should be reassigned to,
you should ask for help on IRC or
on <debian-devel@lists.debian.org>
. Please inform the maintainer(s) of the package
you reassign the bug to, for example by Cc:ing the message that does the
reassign to <
and explaining
your reasons in that mail. Please note that a simple reassignment is
not e-mailed to the maintainers of the package
being reassigned to, so they won't know about it until they look at
a bug overview for their packages.
packagename
@packages.debian.org>
If the bug affects the operation of your package, please consider cloning the bug and reassigning the clone to the package that really causes the behavior. Otherwise, the bug will not be shown in your package's bug list, possibly causing users to report the same bug over and over again. You should block "your" bug with the reassigned, cloned bug to document the relationship.
Sometimes you also have to adjust the severity of the bug so that it matches our definition of the severity. That's because people tend to inflate the severity of bugs to make sure their bugs are fixed quickly. Some bugs may even be dropped to wishlist severity when the requested change is just cosmetic.
If the bug is real but the same problem has already been reported by someone else, then the two relevant bug reports should be merged into one using the merge command of the BTS. In this way, when the bug is fixed, all of the submitters will be informed of this. (Note, however, that emails sent to one bug report's submitter won't automatically be sent to the other report's submitter.) For more details on the technicalities of the merge command and its relative, the unmerge command, see the BTS control server documentation.
The bug submitter may have forgotten to provide some information, in which case
you have to ask them for the required information. You may use the
moreinfo
tag to mark the bug as such. Moreover if you can't
reproduce the bug, you tag it unreproducible
. Anyone who
can reproduce the bug is then invited to provide more information on how to
reproduce it. After a few months, if this information has not been sent by
someone, the bug may be closed.
If the bug is related to the packaging, you just fix it. If you are not able
to fix it yourself, then tag the bug as help
. You can also
ask for help on <debian-devel@lists.debian.org>
or
<debian-qa@lists.debian.org>
. If it's an upstream problem, you have to
forward it to the upstream author. Forwarding a bug is not enough, you have to
check at each release if the bug has been fixed or not. If it has, you just
close it, otherwise you have to remind the author about it. If you have the
required skills you can prepare a patch that fixes the bug and send it to the
author at the same time. Make sure to send the patch to the BTS and to tag the
bug as patch
.
If you have fixed a bug in your local copy, or if a fix has been committed to
the VCS repository, you may tag the bug as pending
to let
people know that the bug is corrected and that it will be closed with the next
upload (add the closes:
in the
changelog
). This is particularly useful if you are
several developers working on the same package.
Once a corrected package is available in the archive, the bug should be closed indicating the version in which it was fixed. This can be done automatically, read Section 5.8.4, “When bugs are closed by new uploads”.
As bugs and problems are fixed in your packages, it is your responsibility as the package maintainer to close these bugs. However, you should not close a bug until the package which fixes the bug has been accepted into the Debian archive. Therefore, once you get notification that your updated package has been installed into the archive, you can and should close the bug in the BTS. Also, the bug should be closed with the correct version.
However, it's possible to avoid having to manually close bugs after the upload
— just list the fixed bugs in your debian/changelog
file, following a certain syntax, and the archive maintenance software will
close the bugs for you. For example:
acme-cannon (3.1415) unstable; urgency=low * Frobbed with options (closes: Bug#98339) * Added safety to prevent operator dismemberment, closes: bug#98765, bug#98713, #98714. * Added man page. Closes: #98725.
Technically speaking, the following Perl regular expression describes how bug closing changelogs are identified:
/closes:\s*(?:bug)?\#\s*\d+(?:,\s*(?:bug)?\#\s*\d+)*/ig
We prefer the closes: #
syntax, as it is the most concise entry and the easiest to integrate with the
text of the XXX
changelog
. Unless specified different by the
-v
-switch to dpkg-buildpackage,
only the bugs closed in the most recent changelog entry are closed (basically,
exactly the bugs mentioned in the changelog-part in the
.changes
file are closed).
Historically, uploads identified as non-maintainer
upload (NMU) were tagged fixed
instead of being
closed, but that practice was ceased with the advent of version-tracking. The
same applied to the tag fixed-in-experimental
.
If you happen to mistype a bug number or forget a bug in the changelog entries,
don't hesitate to undo any damage the error caused. To reopen wrongly closed
bugs, send a reopen
command
to the bug tracking system's control address,
XXX
<control@bugs.debian.org>
. To close any remaining bugs that were
fixed by your upload, email the .changes
file to
<
, where XXX
-done@bugs.debian.org>XXX
is the bug number, and put Version: YYY
and an empty line as the first two
lines of the body of the email, where YYY
is the
first version where the bug has been fixed.
Bear in mind that it is not obligatory to close bugs using the changelog as
described above. If you simply want to close bugs that don't have anything to
do with an upload you made, do it by emailing an explanation to
<
. Do not close bugs in the changelog entry of a version if
the changes in that version of the package don't have any bearing on the bug.
XXX
-done@bugs.debian.org>
For general information on how to write your changelog entries, see Section 6.3, “Best practices for debian/changelog
”.
Due to their sensitive nature, security-related bugs must be handled carefully.
The Debian Security Team exists to coordinate this activity, keeping track of
outstanding security problems, helping maintainers with security problems or
fixing them themselves, sending security advisories, and maintaining
security.debian.org
.
When you become aware of a security-related bug in a Debian package, whether or
not you are the maintainer, collect pertinent information about the problem,
and promptly contact the security team, preferably by filing a ticket in
their Request Tracker.
See http://wiki.debian.org/rt.debian.org#Security_Team.
Alternatively you may email <team@security.debian.org>
.
DO NOT UPLOAD any packages for
stable
without contacting the team. Useful information
includes, for example:
Whether or not the bug is already public.
Which versions of the package are known to be affected by the bug. Check each
version that is present in a supported Debian release, as well as
testing
and unstable
.
The nature of the fix, if any is available (patches are especially helpful)
Any fixed packages that you have prepared yourself (send only the
.diff.gz
and .dsc
files and read Section 5.8.5.4, “Preparing packages to address security issues” first)
Any assistance you can provide to help with testing (exploits, regression testing, etc.)
Any information needed for the advisory (see Section 5.8.5.3, “Security Advisories”)
As the maintainer of the package, you have the responsibility to maintain it, even in the stable release. You are in the best position to evaluate patches and test updated packages, so please see the sections below on how to prepare packages for the Security Team to handle.
The security team maintains a central database, the Debian Security Tracker. This contains all public information that is known about security issues: which packages and versions are affected or fixed, and thus whether stable, testing and/or unstable are vulnerable. Information that is still confidential is not added to the tracker.
You can search it for a specific issue, but also on package name. Look for your package to see which issues are still open. If you can, please provide more information about those issues, or help to address them in your package. Instructions are on the tracker web pages.
Unlike most other activities within Debian, information about security issues must sometimes be kept private for a time. This allows software distributors to coordinate their disclosure in order to minimize their users' exposure. Whether this is the case depends on the nature of the problem and corresponding fix, and whether it is already a matter of public knowledge.
There are several ways developers can learn of a security problem:
they notice it on a public forum (mailing list, web site, etc.)
someone files a bug report
someone informs them via private email
In the first two cases, the information is public and it is important to have a fix as soon as possible. In the last case, however, it might not be public information. In that case there are a few possible options for dealing with the problem:
If the security exposure is minor, there is sometimes no need to keep the problem a secret and a fix should be made and released.
If the problem is severe, it is preferable to share the information with other vendors and coordinate a release. The security team keeps in contact with the various organizations and individuals and can take care of that.
In all cases if the person who reports the problem asks that it not be disclosed, such requests should be honored, with the obvious exception of informing the security team in order that a fix may be produced for a stable release of Debian. When sending confidential information to the security team, be sure to mention this fact.
Please note that if secrecy is needed you may not upload a fix to
unstable
(or
anywhere else, such as a public VCS repository). It is not sufficient to
obfuscate the details of the change, as the code itself is public, and can (and
will) be examined by the general public.
There are two reasons for releasing information even though secrecy is requested: the problem has been known for a while, or the problem or exploit has become public.
The Security Team has a PGP-key to enable encrypted communication about sensitive issues. See the Security Team FAQ for details.
Security advisories are only issued for the current, released stable
distribution, and not for testing
or unstable
. When released, advisories are sent to the
<debian-security-announce@lists.debian.org>
mailing list and posted on
the security web
page. Security advisories are written and posted by the security team.
However they certainly do not mind if a maintainer can supply some of the
information for them, or write part of the text. Information that should be in
an advisory includes:
A description of the problem and its scope, including:
The type of problem (privilege escalation, denial of service, etc.)
What privileges may be gained, and by whom (if any)
How it can be exploited
Whether it is remotely or locally exploitable
How the problem was fixed
This information allows users to assess the threat to their systems.
Version numbers of affected packages
Version numbers of fixed packages
Information on where to obtain the updated packages (usually from the Debian security archive)
References to upstream advisories, CVE identifiers, and any other information useful in cross-referencing the vulnerability
One way that you can assist the security team in their duties is to provide them with fixed packages suitable for a security advisory for the stable Debian release.
When an update is made to the stable release, care must be taken to avoid changing system behavior or introducing new bugs. In order to do this, make as few changes as possible to fix the bug. Users and administrators rely on the exact behavior of a release once it is made, so any change that is made might break someone's system. This is especially true of libraries: make sure you never change the API or ABI, no matter how small the change.
This means that moving to a new upstream version is not a good solution. Instead, the relevant changes should be back-ported to the version present in the current stable Debian release. Generally, upstream maintainers are willing to help if needed. If not, the Debian security team may be able to help.
In some cases, it is not possible to back-port a security fix, for example when large amounts of source code need to be modified or rewritten. If this happens, it may be necessary to move to a new upstream version. However, this is only done in extreme situations, and you must always coordinate that with the security team beforehand.
Related to this is another important guideline: always test your changes. If you have an exploit available, try it and see if it indeed succeeds on the unpatched package and fails on the fixed package. Test other, normal actions as well, as sometimes a security fix can break seemingly unrelated features in subtle ways.
Do NOT include any changes in your package which are not directly related to fixing the vulnerability. These will only need to be reverted, and this wastes time. If there are other bugs in your package that you would like to fix, make an upload to proposed-updates in the usual way, after the security advisory is issued. The security update mechanism is not a means for introducing changes to your package which would otherwise be rejected from the stable release, so please do not attempt to do this.
Review and test your changes as much as possible. Check the differences from
the previous version repeatedly (interdiff from the
patchutils
package and
debdiff from devscripts
are useful tools for this, see Section A.2.2, “debdiff”).
Be sure to verify the following items:
Target the right distribution
in your debian/changelog
.
For stable
this is stable-security
and
for testing
this is testing-security
, and for the previous
stable release, this is oldstable-security
. Do not target
distribution
-proposed-updates
or
stable
!
The upload should have urgency=high.
Make descriptive, meaningful changelog entries. Others will rely on them to
determine whether a particular bug was fixed. Add closes:
statements for any Debian bugs filed.
Always include an external reference, preferably a CVE
identifier, so that it can be cross-referenced. However, if a CVE
identifier has not yet been assigned, do not wait for it but continue the
process. The identifier can be cross-referenced later.
Make sure the version number is proper.
It must be greater than the current package, but less than package versions in
later distributions. If in doubt, test it with dpkg
--compare-versions
. Be careful not to re-use a version number that
you have already used for a previous upload, or one that conflicts with a
binNMU. The convention is to append
+
codename
1
, e.g.
1:2.4.3-4+lenny1
, of course increasing 1 for any subsequent
uploads.
Unless the upstream source has been uploaded to
security.debian.org
before (by a previous security update),
build the upload with full upstream source
(dpkg-buildpackage -sa
). If there has been a previous
upload to security.debian.org
with the same upstream
version, you may upload without upstream source (dpkg-buildpackage
-sd
).
Be sure to use the exact same
*.orig.tar.{gz,bz2,xz}
as used in the
normal archive, otherwise it is not possible to move the security fix into the
main archives later.
Build the package on a clean system which only
has packages installed from the distribution you are building for. If you do not
have such a system yourself, you can use a debian.org machine (see
Section 4.4, “Debian machines”) or setup a chroot (see
Section A.4.3, “pbuilder
” and Section A.4.2, “debootstrap
”).
Do NOT upload a package to the security
upload queue (oldstable-security
, stable-security
,
etc.) without prior authorization from the security team. If the
package does not exactly meet the team's requirements, it will cause many
problems and delays in dealing with the unwanted upload.
Do NOT upload your fix to
proposed-updates
without coordinating with the security team.
Packages from security.debian.org
will be copied into
the proposed-updates
directory automatically. If a package
with the same or a higher version number is already installed into the archive,
the security update will be rejected by the archive system. That way, the
stable distribution will end up without a security update for this package
instead.
Once you have created and tested the new package and it has been approved by
the security team, it needs to be uploaded so that it can be installed in the
archives. For security uploads, the place to upload to is
ftp://security-master.debian.org/pub/SecurityUploadQueue/
.
Once an upload to the security queue has been accepted, the package will automatically be built for all architectures and stored for verification by the security team.
Uploads which are waiting for acceptance or verification are only accessible by the security team. This is necessary since there might be fixes for security problems that cannot be disclosed yet.
If a member of the security team accepts a package, it will be installed on
security.debian.org
as well as proposed for the proper
distribution
-proposed-updates
on ftp-master.debian.org
.
Some archive manipulation operations are not automated in the Debian upload process. These procedures should be manually followed by maintainers. This chapter gives guidelines on what to do in these cases.
Sometimes a package will change its section. For instance, a package from the
non-free
section might be GPL'd in a later version, in which case the package
should be moved to `main' or `contrib'.[3]
If you need to change the section for one of your packages, change the package
control information to place the package in the desired section, and re-upload
the package (see the Debian Policy Manual for
details). You must ensure that you include the
.orig.tar.{gz,bz2,xz}
in your upload (even if you are not uploading
a new upstream version), or it will not appear in the new section together with
the rest of the package. If your new section is valid, it will be moved
automatically. If it does not, then contact the ftpmasters in order to
understand what happened.
If, on the other hand, you need to change the subsection
of one of your packages (e.g., ``devel'', ``admin''), the procedure is slightly
different. Correct the subsection as found in the control file of the package,
and re-upload that. Also, you'll need to get the override file updated, as
described in Section 5.7, “Specifying the package section, subsection and priority”.
If for some reason you want to completely remove a package (say, if it is an
old compatibility library which is no longer required), you need to file a bug
against ftp.debian.org
asking that the package be removed;
as all bugs, this bug should normally have normal severity.
The bug title should be in the form RM:
, where package
[architecture list]
--
reason
package
is the package to be removed and reason
is a
short summary of the reason for the removal request.
[architecture list]
is optional and only needed
if the removal request only applies to some architectures, not all. Note
that the reportbug will create a title conforming
to these rules when you use it to report a bug against the
ftp.debian.org
pseudo-package.
If you want to remove a package you maintain, you should note this in
the bug title by prepending ROM
(Request Of Maintainer).
There are several other standard acronyms used in the reasoning for a package
removal, see http://ftp-master.debian.org/removals.html
for a complete list. That page also provides a convenient overview of
pending removal requests.
Note that removals can only be done for the unstable
,
experimental
and stable
distribution. Packages are not removed from
testing
directly. Rather, they will be removed
automatically after the package has been removed from
unstable
and no package in
testing
depends on it.
There is one exception when an explicit removal request is not necessary: If a (source or binary) package is no longer built from source, it will be removed semi-automatically. For a binary-package, this means if there is no longer any source package producing this binary package; if the binary package is just no longer produced on some architectures, a removal request is still necessary. For a source-package, this means that all binary packages it refers to have been taken over by another source package.
In your removal request, you have to detail the reasons justifying the request. This is to avoid unwanted removals and to keep a trace of why a package has been removed. For example, you can provide the name of the package that supersedes the one to be removed.
Usually you only ask for the removal of a package maintained by yourself. If
you want to remove another package, you have to get the approval of its
maintainer. Should the package be orphaned and thus have no maintainer,
you should first discuss the removal request on <debian-qa@lists.debian.org>
. If
there is a consensus that the package should be removed, you should
reassign and retitle the O:
bug filed against the
wnpp
package instead of filing a new bug as
removal request.
Further information relating to these and other package removal related topics may be found at http://wiki.debian.org/ftpmaster_Removals and http://qa.debian.org/howto-remove.html.
If in doubt concerning whether a package is disposable, email
<debian-devel@lists.debian.org>
asking for opinions. Also of interest is
the apt-cache program from the apt
package. When invoked as apt-cache
showpkg
, the program will show
details for package
package
, including reverse depends.
Other useful programs include apt-cache rdepends,
apt-rdepends, build-rdeps (in the
devscripts
package) and
grep-dctrl. Removal of
orphaned packages is discussed on <debian-qa@lists.debian.org>
.
Once the package has been removed, the package's bugs should be handled. They
should either be reassigned to another package in the case where the actual
code has evolved into another package (e.g. libfoo12
was
removed because libfoo13
supersedes it) or closed if the
software is simply no longer part of Debian.
When closing the bugs,
to avoid marking the bugs as fixed in versions of the packages
in previous Debian releases, they should be marked as fixed
in the version <most-recent-version-ever-in-Debian>+rm
.
In the past, it was possible to remove packages from
incoming
. However, with the introduction of the new
incoming system, this is no longer possible. Instead, you have to upload a new
revision of your package with a higher version than the package you want to
replace. Both versions will be installed in the archive but only the higher
version will actually be available in unstable
since the
previous version will immediately be replaced by the higher. However, if you
do proper testing of your packages, the need to replace a package should not
occur too often anyway.
When the upstream maintainers for one of your packages chose to
rename their software (or you made a mistake naming your package),
you should follow a two-step process to rename it. In the first
step, change the debian/control
file to
reflect the new name and to replace, provide and conflict with the
obsolete package name (see the Debian
Policy Manual for details). Please note that you
should only add a Provides
relation if all
packages depending on the obsolete package name continue to work
after the renaming. Once you've uploaded the package and the package
has moved into the archive, file a bug against ftp.debian.org
asking to remove the package with the
obsolete name (see Section 5.9.2, “Removing packages”). Do not forget
to properly reassign the package's bugs at the same time.
At other times, you may make a mistake in constructing your package and wish to
replace it. The only way to do this is to increase the version number and
upload a new version. The old version will be expired in the usual manner.
Note that this applies to each part of your package, including the sources: if
you wish to replace the upstream source tarball of your package, you will need
to upload it with a different version. An easy possibility is to replace
foo_1.00.orig.tar.gz
with
foo_1.00+0.orig.tar.gz
or
foo_1.00.orig.tar.bz2
. This restriction gives each
file on the ftp site a unique name, which helps to ensure consistency
across the mirror network.
If you can no longer maintain a package, you need to inform others, and see
that the package is marked as orphaned. You should set the package maintainer
to Debian QA Group <packages@qa.debian.org>
and
submit a bug report against the pseudo package wnpp
. The bug report should be titled O:
indicating that the package is now
orphaned. The severity of the bug should be set to
package
-- short
description
normal
; if the package has a priority of standard or
higher, it should be set to important. If you feel it's necessary, send a copy
to <debian-devel@lists.debian.org>
by putting the address in the
X-Debbugs-CC: header of the message (no, don't use CC:, because that way the
message's subject won't indicate the bug number).
If you just intend to give the package away, but you can keep maintainership
for the moment, then you should instead submit a bug against wnpp
and title it RFA:
. package
-- short
description
RFA
stands for
Request For Adoption
.
More information is on the WNPP web pages.
A list of packages in need of a new maintainer is available in the Work-Needing and Prospective Packages list (WNPP). If you wish to take over maintenance of any of the packages listed in the WNPP, please take a look at the aforementioned page for information and procedures.
It is not OK to simply take over a package that you feel is neglected — that would be package hijacking. You can, of course, contact the current maintainer and ask them if you may take over the package. If you have reason to believe a maintainer has gone AWOL (absent without leave), see Section 7.4, “Dealing with inactive and/or unreachable maintainers”.
Generally, you may not take over the package without the assent of the current maintainer. Even if they ignore you, that is still not grounds to take over a package. Complaints about maintainers should be brought up on the developers' mailing list. If the discussion doesn't end with a positive conclusion, and the issue is of a technical nature, consider bringing it to the attention of the technical committee (see the technical committee web page for more information).
If you take over an old package, you probably want to be listed as the
package's official maintainer in the bug system. This will happen
automatically once you upload a new version with an updated
Maintainer
field, although it can take a few hours after
the upload is done. If you do not expect to upload a new version for a while,
you can use Section 4.10, “The Package Tracking System” to get the bug reports.
However, make sure that the old maintainer has no problem with the fact that
they will continue to receive the bugs during that time.
Debian supports an ever-increasing number of architectures. Even if you are not a porter, and you don't use any architecture but one, it is part of your duty as a maintainer to be aware of issues of portability. Therefore, even if you are not a porter, you should read most of this chapter.
Porting is the act of building Debian packages for architectures that are
different from the original architecture of the package maintainer's binary
package. It is a unique and essential activity. In fact, porters do most of
the actual compiling of Debian packages. For instance, when a maintainer
uploads a (portable) source packages with binaries for the i386
architecture, it will be built for each of the other architectures,
amounting to 11 more builds.
Porters have a difficult and unique task, since they are required to deal with a large volume of packages. Ideally, every source package should build right out of the box. Unfortunately, this is often not the case. This section contains a checklist of ``gotchas'' often committed by Debian maintainers — common problems which often stymie porters, and make their jobs unnecessarily difficult.
The first and most important thing is to respond quickly to bug or issues raised by porters. Please treat porters with courtesy, as if they were in fact co-maintainers of your package (which, in a way, they are). Please be tolerant of succinct or even unclear bug reports; do your best to hunt down whatever the problem is.
By far, most of the problems encountered by porters are caused by packaging bugs in the source packages. Here is a checklist of things you should check or be aware of.
Make sure that your Build-Depends
and
Build-Depends-Indep
settings in
debian/control
are set properly. The best way to validate
this is to use the debootstrap
package
to create an unstable
chroot environment (see Section A.4.2, “debootstrap
”).
Within that chrooted environment, install the build-essential
package and any package
dependencies mentioned in Build-Depends
and/or
Build-Depends-Indep
. Finally, try building your package
within that chrooted environment. These steps can be automated by the use of
the pbuilder program which is provided by the package of the
same name (see Section A.4.3, “pbuilder
”).
If you can't set up a proper chroot, dpkg-depcheck may be of assistance (see Section A.6.7, “dpkg-depcheck”).
See the Debian Policy Manual for instructions on setting build dependencies.
Don't set architecture to a value other than all
or
any
unless you really mean it. In too many cases,
maintainers don't follow the instructions in the Debian Policy Manual. Setting your
architecture to only one architecture (such as i386
or amd64
) is usually incorrect.
Make sure your source package is correct. Do dpkg-source -x
to make sure your source
package unpacks properly. Then, in there, try building your package from
scratch with dpkg-buildpackage.
package
.dsc
Make sure you don't ship your source package with the
debian/files
or debian/substvars
files. They should be removed by the clean
target of
debian/rules
.
Make sure you don't rely on locally installed or hacked configurations or
programs. For instance, you should never be calling programs in
/usr/local/bin
or the like. Try not to rely on programs
being setup in a special way. Try building your package on another machine,
even if it's the same architecture.
Don't depend on the package you're building being installed already (a sub-case of the above issue). There are, of course, exceptions to this rule, but be aware that any case like this needs manual bootstrapping and cannot be done by automated package builders.
Don't rely on the compiler being a certain version, if possible. If not, then make sure your build dependencies reflect the restrictions, although you are probably asking for trouble, since different architectures sometimes standardize on different compilers.
Make sure your debian/rules
contains separate binary-arch
and binary-indep
targets, as the Debian Policy Manual
requires. Make sure that both targets work independently, that is, that you
can call the target without having called the other before. To test this,
try to run dpkg-buildpackage -B.
If the package builds out of the box for the architecture to be ported to, you are in luck and your job is easy. This section applies to that case; it describes how to build and upload your binary package so that it is properly installed into the archive. If you do have to patch the package in order to get it to compile for the other architecture, you are actually doing a source NMU, so consult Section 5.11.1, “When and how to do an NMU” instead.
For a porter upload, no changes are being made to the source. You do not need
to touch any of the files in the source package. This includes
debian/changelog
.
The way to invoke dpkg-buildpackage is as
dpkg-buildpackage -B
-m
. Of course, set
porter-email
porter-email
to your email address. This will do a
binary-only build of only the architecture-dependent portions of the package,
using the binary-arch
target in
debian/rules
.
If you are working on a Debian machine for your porting efforts and you need to
sign your upload locally for its acceptance in the archive, you can run
debsign on your .changes
file to have
it signed conveniently, or use the remote signing mode of
dpkg-sig.
Sometimes the initial porter upload is problematic because the environment in which the package was built was not good enough (outdated or obsolete library, bad compiler, etc.). Then you may just need to recompile it in an updated environment. However, you have to bump the version number in this case, so that the old bad package can be replaced in the Debian archive (dak refuses to install new packages if they don't have a version number greater than the currently available one).
You have to make sure that your binary-only NMU doesn't render the package
uninstallable. This could happen when a source package generates
arch-dependent and arch-independent packages that have inter-dependencies
generated using dpkg's substitution variable $(Source-Version)
.
Despite the required modification of the changelog, these are called binary-only NMUs — there is no need in this case to trigger all other architectures to consider themselves out of date or requiring recompilation.
Such recompilations require special ``magic'' version numbering, so that the archive maintenance tools recognize that, even though there is a new Debian version, there is no corresponding source update. If you get this wrong, the archive maintainers will reject your upload (due to lack of corresponding source code).
The ``magic'' for a recompilation-only NMU is triggered by using a suffix
appended to the package version number, following the form
b
.
For instance, if the latest version you are recompiling against was version
number
2.9-3
, your binary-only NMU should carry a version of
2.9-3+b1
. If the latest version was 3.4+b1
(i.e, a native package with a previous recompilation NMU), your
binary-only NMU should have a version number of 3.4+b2
.[4]
Similar to initial porter uploads, the correct way of invoking
dpkg-buildpackage is dpkg-buildpackage -B
to only build the architecture-dependent parts of the package.
Porters doing a source NMU generally follow the guidelines found in Section 5.11, “Non-Maintainer Uploads (NMUs)”, just like non-porters. However, it is expected that the wait cycle for a porter's source NMU is smaller than for a non-porter, since porters have to cope with a large quantity of packages. Again, the situation varies depending on the distribution they are uploading to. It also varies whether the architecture is a candidate for inclusion into the next stable release; the release managers decide and announce which architectures are candidates.
If you are a porter doing an NMU for unstable
, the above
guidelines for porting should be followed, with two variations. Firstly, the
acceptable waiting period — the time between when the bug is submitted to
the BTS and when it is OK to do an NMU — is seven days for porters working
on the unstable
distribution. This period can be shortened
if the problem is critical and imposes hardship on the porting effort, at the
discretion of the porter group. (Remember, none of this is Policy, just
mutually agreed upon guidelines.) For uploads to stable
or
testing
, please coordinate with the appropriate release
team first.
Secondly, porters doing source NMUs should make sure that the bug they submit
to the BTS should be of severity serious
or greater. This
ensures that a single source package can be used to compile every supported
Debian architecture by release time. It is very important that we have one
version of the binary and source package for all architectures in order to
comply with many licenses.
Porters should try to avoid patches which simply kludge around bugs in the
current version of the compile environment, kernel, or libc. Sometimes such
kludges can't be helped. If you have to kludge around compiler bugs and the
like, make sure you #ifdef
your work properly; also,
document your kludge so that people know to remove it once the external
problems have been fixed.
Porters may also have an unofficial location where they can put the results of their work during the waiting period. This helps others running the port have the benefit of the porter's work, even during the waiting period. Of course, such locations have no official blessing or status, so buyer beware.
There is infrastructure and several tools to help automate package porting. This section contains a brief overview of this automation and porting to these tools; see the package documentation or references for full information.
Web pages containing the status of each port can be found at http://www.debian.org/ports/.
Each port of Debian has a mailing list. The list of porting mailing lists can be found at http://lists.debian.org/ports.html. These lists are used to coordinate porters, and to connect the users of a given port with the porters.
Descriptions of several porting tools can be found in Section A.7, “Porting tools”.
The wanna-build
system is used as a
distributed, client-server build distribution system. It is usually used in
conjunction with build daemons running the buildd
program. Build daemons
are ``slave'' hosts
which contact the central wanna-build
system to receive a list of packages that need to be built.
wanna-build
is not yet available as a
package; however, all Debian porting efforts are using it for automated
package building. The tool used to do the actual package builds, sbuild
is available as a package, see its
description in Section A.4.4, “sbuild
”. Please note that the packaged
version is not the same as the one used on build daemons, but it is close
enough to reproduce problems.
Most of the data produced by wanna-build
which is generally useful to porters is available on the
web at http://buildd.debian.org/. This data includes nightly
updated statistics, queueing information and logs for build attempts.
We are quite proud of this system, since it has so many possible uses. Independent development groups can use the system for different sub-flavors of Debian, which may or may not really be of general interest (for instance, a flavor of Debian built with gcc bounds checking). It will also enable Debian to recompile entire distributions quickly.
The wanna-build team, in charge of the buildds,
can be reached at debian-wb-team@lists.debian.org
.
To determine who (wanna-build team, release team) and how (mail, BTS)
to contact, refer to http://lists.debian.org/debian-project/2009/03/msg00096.html.
When requesting binNMUs or give-backs (retries after a failed build), please use the format described at http://release.debian.org/wanna-build.txt.
Some packages still have issues with building and/or working on some of the
architectures supported by Debian, and cannot be ported at all, or not within a
reasonable amount of time. An example is a package that is SVGA-specific (only
available for i386
and amd64
), or uses
other hardware-specific features not supported on all architectures.
In order to prevent broken packages from being uploaded to the archive, and wasting buildd time, you need to do a few things:
First, make sure your package does fail to build on architectures that it cannot support. There are a few ways to achieve this. The preferred way is to have a small testsuite during build time that will test the functionality, and fail if it doesn't work. This is a good idea anyway, as this will prevent (some) broken uploads on all architectures, and also will allow the package to build as soon as the required functionality is available.
Additionally, if you believe the list of supported architectures is pretty
constant, you should change any
to a list of supported
architectures in debian/control
. This way, the build will
fail also, and indicate this to a human reader without actually trying.
In order to prevent autobuilders from needlessly trying to build your package,
it must be included in Packages-arch-specific
, a list used
by the wanna-build script. The current version is available
as http://buildd.debian.org/quinn-diff/Packages-arch-specific;
please see the top of the file for whom to contact for changes.
Please note that it is insufficient to only add your package to
Packages-arch-specific
without making it fail to build on unsupported
architectures: A porter or any other person trying to build your package might
accidently upload it without noticing it doesn't work. If in the past some
binary packages were uploaded on unsupported architectures, request their
removal by filing a bug against ftp.debian.org
.
By default packages from the non-free
section are not built by the autobuilder
network (mostly because the license of the packages could disapprove).
To enable a package to be build you need to perform the following
steps:
Check whether it is legally allowed and technically possible to auto-build the package;
Add XS-Autobuild: yes
into the header part
of debian/control
;
Send an email to <nonfree@release.debian.org>
and explain why the
package can legitimately and technically be auto-built.
Every package has one or more maintainers. Normally, these are the people who work on and upload new versions of the package. In some situations, it is useful that other developers can upload a new version as well, for example if they want to fix a bug in a package they don't maintain, when the maintainer needs help to respond to issues. Such uploads are called Non-Maintainer Uploads (NMU).
Before doing an NMU, consider the following questions:
Does your NMU really fix bugs? Fixing cosmetic issues or changing the packaging style in NMUs is discouraged.
Did you give enough time to the maintainer? When was the bug reported to the BTS? Being busy for a week or two isn't unusual. Is the bug so severe that it needs to be fixed right now, or can it wait a few more days?
How confident are you about your changes? Please remember the Hippocratic Oath: "Above all, do no harm." It is better to leave a package with an open grave bug than applying a non-functional patch, or one that hides the bug instead of resolving it. If you are not 100% sure of what you did, it might be a good idea to seek advice from others. Remember that if you break something in your NMU, many people will be very unhappy about it.
Have you clearly expressed your intention to NMU, at least in the BTS? It is also a good idea to try to contact the maintainer by other means (private email, IRC).
If the maintainer is usually active and responsive, have you tried to contact him? In general it should be considered preferable that a maintainer takes care of an issue himself and that he is given the chance to review and correct your patch, because he can be expected to be more aware of potential issues which an NMUer might miss. It is often a better use of everyone's time if the maintainer is given an opportunity to upload a fix on their own.
When doing an NMU, you must first make sure that your intention to NMU is
clear. Then, you must send a patch with the differences between the
current package and your proposed NMU to the BTS. The
nmudiff script in the devscripts
package
might be helpful.
While preparing the patch, you should better be aware of any package-specific
practices that the maintainer might be using. Taking them into account
reduces the burden of integrating your changes into the normal package
workflow and thus increases the chances that integration will happen. A good
place where to look for for possible package-specific practices is
debian/README.source
.
Unless you have an excellent reason not to do so, you must then give some time
to the maintainer to react (for example, by uploading to the
DELAYED
queue). Here are some recommended values to use for delays:
Upload fixing only release-critical bugs older than 7 days, with no maintainer activity on the bug for 7 days and no indication that a fix is in progress: 0 days
Upload fixing only release-critical bugs older than 7 days: 2 days
Upload fixing only release-critical and important bugs: 5 days
Other NMUs: 10 days
Those delays are only examples. In some cases, such as uploads fixing security
issues, or fixes for trivial bugs that blocking a transition, it is desirable
that the fixed package reaches unstable
sooner.
Sometimes, release managers decide to allow NMUs with shorter delays for a subset of bugs (e.g release-critical bugs older than 7 days). Also, some maintainers list themselves in the Low Threshold NMU list, and accept that NMUs are uploaded without delay. But even in those cases, it's still a good idea to give the maintainer a few days to react before you upload, especially if the patch wasn't available in the BTS before, or if you know that the maintainer is generally active.
After you upload an NMU, you are responsible for the possible problems that you might have introduced. You must keep an eye on the package (subscribing to the package on the PTS is a good way to achieve this).
This is not a license to perform NMUs thoughtlessly. If you NMU when it is clear that the maintainers are active and would have acknowledged a patch in a timely manner, or if you ignore the recommendations of this document, your upload might be a cause of conflict with the maintainer. You should always be prepared to defend the wisdom of any NMU you perform on its own merits.
Just like any other (source) upload, NMUs must add an entry to
debian/changelog
, telling what has changed with this
upload. The first line of this entry must explicitely mention that this upload is an NMU, e.g.:
* Non-maintainer upload.
The way to version NMUs differs for native and non-native packages.
If the package is a native package (without a Debian revision in the version number),
the version must be the version of the last maintainer upload, plus
+nmu
, where
X
X
is a counter starting at 1
.
If
the last upload was also an NMU, the counter should be increased. For example,
if the current version is 1.5
, then an NMU would get
version 1.5+nmu1
.
If the package is not a native package, you should add a minor version number
to the Debian revision part of the version number (the portion after the last
hyphen). This extra number must start at 1
. For example,
if the current version is 1.5-2
, then an NMU would get
version 1.5-2.1
. If a new upstream version
is packaged in the NMU, the Debian revision is set to 0
, for
example 1.6-0.1
.
In both cases, if the last upload was also an NMU, the counter should
be increased. For example, if the current version is
1.5+nmu3
(a native package which has already been
NMUed), the NMU would get version 1.5+nmu4
.
A special versioning scheme is needed to avoid disrupting the maintainer's work, since using an integer for the Debian revision will potentially conflict with a maintainer upload already in preparation at the time of an NMU, or even one sitting in the ftp NEW queue. It also has the benefit of making it visually clear that a package in the archive was not made by the official maintainer.
If you upload a package to testing or stable, you sometimes need to "fork" the
version number tree. This is the case for security uploads, for example. For
this, a version of the form
+deb
should be used, where XY
uZ
X
and
Y
are the major and minor release numbers, and
Z
is a counter starting at 1
.
When the release number is not yet known (often the case for
testing
, at the beginning of release cycles), the lowest
release number higher than the last stable release number must be used. For
example, while Lenny (Debian 5.0) is stable, a security NMU to stable for a
package at version 1.5-3
would have version
1.5-3+deb50u1
, whereas a security NMU to Squeeze would get
version 1.5-3+deb60u1
. After the release of Squeeze, security
uploads to the testing
distribution will be versioned
+deb61uZ
, until it is known whether that release will be
Debian 6.1 or Debian 7.0 (if that becomes the case, uploads will be versioned
as +deb70uZ
).
Having to wait for a response after you request permission to NMU is
inefficient, because it costs the NMUer a context switch to come back to the
issue.
The DELAYED
queue (see Section 5.6.2, “Delayed uploads”)
allows the developer doing the NMU to perform all the necessary tasks at the
same time. For instance, instead of telling the maintainer that you will
upload the updated
package in 7 days, you should upload the package to
DELAYED/7
and tell the maintainer that he has 7 days to
react. During this time, the maintainer can ask you to delay the upload some
more, or cancel your upload.
The DELAYED
queue should not be used to put additional
pressure on the maintainer. In particular, it's important that you are
available to cancel or delay the upload before the delay expires since the
maintainer cannot cancel the upload himself.
If you make an NMU to DELAYED
and the maintainer updates
his package before the delay expires, your upload will be rejected because a
newer version is already available in the archive.
Ideally, the maintainer will take care to include your proposed changes (or
at least a solution for the problems they address) in that upload.
When someone NMUs your package, this means they want to help you to keep it in good shape. This gives users fixed packages faster. You can consider asking the NMUer to become a co-maintainer of the package. Receiving an NMU on a package is not a bad thing; it just means that the package is interesting enough for other people to work on it.
To acknowledge an NMU, include its changes and changelog entry in your next maintainer upload. If you do not acknowledge the NMU by including the NMU changelog entry in your changelog, the bugs will remain closed in the BTS but will be listed as affecting your maintainer version of the package.
The full name of an NMU is source NMU. There is also another type, namely the binary-only NMU, or binNMU. A binNMU is also a package upload by someone other than the package's maintainer. However, it is a binary-only upload.
When a library (or other dependency) is updated, the packages using it may need to be rebuilt. Since no changes to the source are needed, the same source package is used.
BinNMUs are usually triggered on the buildds by wanna-build.
An entry is added to debian/changelog
,
explaining why the upload was needed and increasing the version number as
described in Section 5.10.2.1, “Recompilation or binary-only NMU”.
This entry should not be included in the next upload.
Buildds upload packages for their architecture to the archive as binary-only
uploads. Strictly speaking, these are binNMUs. However, they are not normally
called NMU, and they don't add an entry to debian/changelog
.
NMUs are uploads of packages by somebody else than their assigned maintainer. There is another type of upload where the uploaded package is not yours: QA uploads. QA uploads are uploads of orphaned packages.
QA uploads are very much like normal maintainer uploads: they may fix anything,
even minor issues; the version numbering is normal, and there is no need to use
a delayed upload. The difference is that you are not listed as the Maintainer
or Uploader
for the package. Also, the changelog entry of a QA upload has a
special first line:
* QA upload.
If you want to do an NMU, and it seems that the maintainer is not active, it is
wise to check if the package is orphaned
(this information is displayed on the package's Package Tracking System page).
When doing the first QA upload to an
orphaned package, the maintainer should be set to Debian QA Group
<packages@qa.debian.org>
. Orphaned packages which did
not yet have a QA upload still have their old maintainer set. There is a list
of them at http://qa.debian.org/orphaned.html.
Instead of doing a QA upload, you can also consider adopting the package by making yourself the maintainer. You don't need permission from anybody to adopt an orphaned package, you can just set yourself as maintainer and upload the new version (see Section 5.9.5, “Adopting a package”).
Sometimes you are fixing and/or updating a package because you are member of a
packaging team (which uses a mailing list as Maintainer
or Uploader
, see Section 5.12, “Collaborative maintenance”) but you don't want to add yourself to Uploaders
because you do not plan to contribute regularly to this specific package. If it
conforms with your team's policy, you can perform a normal upload without
being listed directly as Maintainer
or Uploader
. In that case, you should
start your changelog entry with the following line:
* Team upload.
Collaborative maintenance is a term describing the sharing of Debian package
maintenance duties by several people. This collaboration is almost always a
good idea, since it generally results in higher quality and faster bug fix
turnaround times. It is strongly recommended that packages with a priority of
standard
or which are part of the base set have
co-maintainers.
Generally there is a primary maintainer and one or more co-maintainers. The
primary maintainer is the person whose name is listed in the
Maintainer
field of the debian/control
file. Co-maintainers are all the other maintainers,
usually listed in the Uploaders
field of the
debian/control
file.
In its most basic form, the process of adding a new co-maintainer is quite easy:
Setup the co-maintainer with access to the sources you build the package from.
Generally this implies you are using a network-capable version control system,
such as CVS
or Subversion
. Alioth (see
Section 4.12, “Debian's FusionForge installation: Alioth”) provides such tools, amongst others.
Add the co-maintainer's correct maintainer name and address to the
Uploaders
field in the first paragraph of the
debian/control
file.
Uploaders: John Buzz <jbuzz@debian.org>, Adam Rex <arex@debian.org>
Using the PTS (Section 4.10, “The Package Tracking System”), the co-maintainers should subscribe themselves to the appropriate source package.
Another form of collaborative maintenance is team maintenance, which is
recommended if you maintain several packages with the same group of developers.
In that case, the Maintainer
and Uploaders
field of each package must be
managed with care. It is recommended to choose between one of the two
following schemes:
Put the team member mainly responsible for the package in the Maintainer
field.
In the Uploaders
, put the mailing list address, and the team members who care
for the package.
Put the mailing list address in the Maintainer
field. In the Uploaders
field,
put the team members who care for the package. In this case, you must make
sure the mailing list accept bug reports without any human interaction (like
moderation for non-subscribers).
In any case, it is a bad idea to automatically put all team members in the
Uploaders
field. It clutters the Developer's Package Overview listing (see
Section 4.11, “Developer's packages overview”) with packages one doesn't really care for, and creates
a false sense of good maintenance. For the same reason, team members do
not need to add themselves to the Uploaders
field just because they are
uploading the package once, they can do a “Team upload” (see Section 5.11.7, “NMUs vs team uploads”). Conversely, it is a bad idea to keep a
package with only the mailing list address as a Maintainer
and no
Uploaders
.
Packages are usually installed into the testing
distribution
after they have undergone some degree of testing
in
unstable
.
They must be in sync on all architectures and mustn't have dependencies that
make them uninstallable; they also have to have generally no known
release-critical bugs at the time they're installed into testing
.
This way, testing
should always be close to
being a release candidate. Please see below for details.
The scripts that update the testing
distribution are run
twice each day, right after the installation of the updated packages; these
scripts are called britney
. They generate the
Packages
files for the testing
distribution, but they do so in an intelligent manner; they try to avoid any
inconsistency and to use only non-buggy packages.
The inclusion of a package from unstable
is conditional on
the following:
The package must have been available in unstable
for 2, 5
or 10 days, depending on the urgency (high, medium or low). Please note that
the urgency is sticky, meaning that the highest urgency uploaded since the
previous testing
transition is taken into account. Those
delays may be doubled during a freeze, or testing
transitions may be switched off altogether;
It must not have new release-critical bugs (RC bugs affecting the version
available in unstable
, but not affecting the version in
testing
);
It must be available on all architectures on which it has previously been built
in unstable
. dak ls may be of interest
to check that information;
It must not break any dependency of a package which is already available in
testing
;
The packages on which it depends must either be available in
testing
or they must be accepted into
testing
at the same time (and they will be if they fulfill
all the necessary criteria).
To find out whether a package is progressing into testing
or not, see the testing
script output on the web page of the testing
distribution, or use the program grep-excuses which
is in the devscripts
package. This
utility can easily be used in a crontab(5)
to keep yourself informed of the progression of your packages into
testing
.
The update_excuses
file does not always give the precise
reason why the package is refused; you may have to find it on your own by
looking for what would break with the inclusion of the package. The testing web page gives some
more information about the usual problems which may be causing such troubles.
Sometimes, some packages never enter testing
because the
set of interrelationship is too complicated and cannot be sorted out by the
scripts. See below for details.
Some further dependency analysis is shown on http://release.debian.org/migration/ — but be warned, this page also shows build dependencies which are not considered by britney.
For the testing
migration script, outdated means: There are
different versions in unstable
for the release architectures
(except for the architectures in fuckedarches; fuckedarches is a list of
architectures that don't keep up (in update_out.py
), but
currently, it's empty). outdated has nothing whatsoever to do with the
architectures this package has in testing
.
Consider this example:
alpha | arm | |
---|---|---|
testing | 1 | - |
unstable | 1 | 2 |
The package is out of date on alpha
in unstable
, and will
not go to testing
. Removing the package would not help at all, the
package is still out of date on alpha
, and will not
propagate to testing
.
However, if ftp-master removes a package in unstable
(here
on arm
):
alpha | arm | hurd-i386 | |
---|---|---|---|
testing | 1 | 1 | - |
unstable | 2 | - | 1 |
In this case, the package is up to date on all release architectures in
unstable
(and the extra hurd-i386
doesn't matter, as it's not a release architecture).
Sometimes, the question is raised if it is possible to allow packages in that are not yet built on all architectures: No. Just plainly no. (Except if you maintain glibc or so.)
Sometimes, a package is removed to allow another package in: This happens only
to allow another package to go in if it's ready in every
other sense. Suppose e.g. that a
cannot be installed
with the new version of b
; then a
may
be removed to allow b
in.
Of course, there is another reason to remove a package from testing
:
It's just too buggy (and having a single RC-bug is enough to be
in this state).
Furthermore, if a package has been removed from unstable
,
and no package in testing
depends on it any more, then it
will automatically be removed.
A situation which is not handled very well by britney is if package
a
depends on the new version of package
b
, and vice versa.
An example of this is:
testing | unstable | |
---|---|---|
a | 1; depends: b=1 | 2; depends: b=2 |
b | 1; depends: a=1 | 2; depends: a=2 |
Neither package a
nor package b
is
considered for update.
Currently, this requires some manual hinting from the release team. Please
contact them by sending mail to <debian-release@lists.debian.org>
if this
happens to one of your packages.
Generally, there is nothing that the status of a package in testing
means for transition of the next version from unstable
to testing
, with two exceptions:
If the RC-bugginess of the package goes down, it may go in even if it is still
RC-buggy. The second exception is if the version of the package in
testing
is out of sync on the different arches: Then any arch might
just upgrade to the version of the source package; however, this can happen
only if the package was previously forced through, the arch is in fuckedarches,
or there was no binary package of that arch present in unstable
at all during the testing
migration.
In summary this means: The only influence that a package being in
testing
has on a new version of the same package is that the new
version might go in easier.
If you are interested in details, this is how britney works:
The packages are looked at to determine whether they are valid candidates. This gives the update excuses. The most common reasons why a package is not considered are too young, RC-bugginess, and out of date on some arches. For this part of britney, the release managers have hammers of various sizes to force britney to consider a package. (Also, the base freeze is coded in that part of britney.) (There is a similar thing for binary-only updates, but this is not described here. If you're interested in that, please peruse the code.)
Now, the more complex part happens: Britney tries to update testing
with the valid candidates. For that, britney tries to add each
valid candidate to the testing distribution. If the number of uninstallable
packages in testing
doesn't increase, the package is
accepted. From that point on, the accepted package is considered to be part
of testing
, such that all subsequent installability
tests include this package. Hints from the release team are processed
before or after this main run, depending on the exact type.
If you want to see more details, you can look it up on http://ftp-master.debian.org/testing/update_output/.
The hints are available via http://ftp-master.debian.org/testing/hints/.
The testing
distribution is fed with packages from
unstable
according to the rules explained above. However,
in some cases, it is necessary to upload packages built only for
testing
. For that, you may want to upload to
testing-proposed-updates
.
Keep in mind that packages uploaded there are not automatically processed, they
have to go through the hands of the release manager. So you'd better have a
good reason to upload there. In order to know what a good reason is in the
release managers' eyes, you should read the instructions that they regularly
give on <debian-devel-announce@lists.debian.org>
.
You should not upload to testing-proposed-updates
when you
can update your packages through unstable
. If you can't
(for example because you have a newer development version in unstable
),
you may use this facility, but it is recommended that you ask for
authorization from the release manager first. Even if a package is frozen,
updates through unstable
are possible, if the upload via
unstable
does not pull in any new dependencies.
Version numbers are usually selected by adding the codename of the
testing
distribution and a running number, like
1.2squeeze1
for the first upload through
testing-proposed-updates
of package version
1.2
.
Please make sure you didn't miss any of these items in your upload:
Make sure that your package really needs to go through
testing-proposed-updates
, and can't go through
unstable
;
Make sure that you included only the minimal amount of changes;
Make sure that you included an appropriate explanation in the changelog;
Make sure that you've written testing
or
testing-proposed-updates
into your target distribution;
Make sure that you've built and tested your package in
testing
, not in unstable
;
Make sure that your version number is higher than the version in
testing
and testing-proposed-updates
,
and lower than in unstable
;
After uploading and successful build on all platforms, contact the release team
at <debian-release@lists.debian.org>
and ask them to approve your upload.
All bugs of some higher severities are by default considered release-critical;
currently, these are critical
, grave
and
serious
bugs.
Such bugs are presumed to have an impact on the chances that the package will
be released with the stable
release of Debian: in general,
if a package has open release-critical bugs filed on it, it won't get into
testing
, and consequently won't be released in
stable
.
The unstable
bug count are all release-critical bugs which
are marked to apply to package
/version
combinations that are available in unstable for a release
architecture. The testing
bug count is defined analogously.
The structure of the distribution archives is such that they can only contain
one version of a package; a package is defined by its name. So when the source
package acmefoo
is installed into testing
,
along with its binary packages acme-foo-bin
,
acme-bar-bin
, libacme-foo1
and
libacme-foo-dev
, the old version is removed.
However, the old version may have provided a binary package with an old soname
of a library, such as libacme-foo0
. Removing the old
acmefoo
will remove libacme-foo0
, which
will break any packages which depend on it.
Evidently, this mainly affects packages which provide changing sets of binary packages in different versions (in turn, mainly libraries). However, it will also affect packages upon which versioned dependencies have been declared of the ==, <=, or << varieties.
When the set of binary packages provided by a source package change in this
way, all the packages that depended on the old binaries will have to be updated
to depend on the new binaries instead. Because installing such a source
package into testing
breaks all the packages that depended on
it in testing
,
some care has to be taken now: all the depending packages must be updated and
ready to be installed themselves so that they won't be broken, and, once
everything is ready, manual intervention by the release manager or an assistant
is normally required.
If you are having problems with complicated groups of packages like this,
contact <debian-devel@lists.debian.org>
or <debian-release@lists.debian.org>
for help.
[3] See the Debian Policy Manual for guidelines on what section a package belongs in.
[4] In the past, such NMUs used the third-level number on the Debian part of the revision to denote their recompilation-only status; however, this syntax was ambiguous with native packages and did not allow proper ordering of recompile-only NMUs, source NMUs, and security NMUs on the same package, and has therefore been abandoned in favor of this new syntax.