Guidelines minor changes
Here is a new version of the Guidelines. This'll be in the next dpkg;
in the meantime the two main info file pieces are below, and the texi
and info files are on their way to master.
Apart from the fairly minor changes listed below, this file contains
the information from:
README.etc-skel
descriptions.txt
README.init
mailers.txt
maintainer-script-args.txt
virtual-dependencies.txt
dependency-ordering.txt
dpkg-upgrades+errors.txt (well, this was actually out of date - I
deleted the info from it.)
Thanks to Erick Branderhorst. These files should be moved to some
kind of `obsolete' directory for the moment, and deleted after a while.
I need an texi2ascii converter.
Other changes:
I have removed the following paragraph:
* If changes to the source code are made, please use a `define'. If
they are changes required to compile or function under Linux in
general, use `LINUX'. If it is a cosmetic or functional change,
use `DEBIAN'.
No-one seems to do it anyway, and this is what having the original
source is for. I've replaced it with:
* If changes to the source code are made that are applicable to
Linux systems or systems in general please try to get them
included in the upstream version of the package by supplying the
upstream authors with the changes in whatever form they prefer.
I've added to the paragraph
You should retain the source package version numbering in the
`Version' field--the version number should be the same for the
Debianised source tree and all the binary packages generated from
it. See below for details of version numbers.
a rider saying that the version numbers must be made to sort
correctly.
I intend write a script to handle the new source package format that I
wrote about here earlier, and eventually this will be documented in
the Guidelines.
Ian.
This is Info file guidelines.info, produced by Makeinfo-1.63 from the
input file guidelines.texi.
START-INFO-DIR-ENTRY
* debian-guidelines: (debian-guidelines). How to make Debian packages.
END-INFO-DIR-ENTRY
File: guidelines.info, Node: Top, Next: Additional Information, Prev: (dir), Up: (dir)
Debian GNU/Linux Packaging Guidelines
*************************************
This file documents the steps that must be taken in the preparation
of a Debian GNU/Linux package. All submissions to be included in the
distribution proper and all packages to be considered for `contrib' or
`non-free' availability *must* conform to the guidelines and standards
described in this document or they cannot be included or made available
at the archive with the distribution.
Please read the Guidelines carefully. If you have comments or
questions, please contact `debian-devel@pixar.com'. If you are
planning on going further than just contributing a package (i.e., if
you plan to maintain it for an extended period of time or if you are
generally interested in becoming more involved in the Project), you
should join the `debian-devel' mailing list. For more details, read
`info/mailing-lists.txt', available at any Debian GNU/Linux archive.
(This file was last updated on 26th January 1996. Please check the
most recent `dpkg' package at any Debian GNU/Linux archive for a
potentially more up to date copy.)
* Menu:
* Additional Information:: Where other info is to be found.
* Package Copyright:: A few words about the importance of
understanding the package copyright.
* Package Content:: Requirements for the package content.
* Source Package:: Creating the source package.
* Binary Package:: Creating the binary package.
* Control Files:: The binary package control files.
* Appendix:: More specific details about some aspects.
File: guidelines.info, Node: Additional Information, Next: Package Copyright, Prev: Top, Up: Top
Additional Information
**********************
These Guidelines are intended to be fairly general. More specific
information is available about certain aspects of building packages,
such as how to use the features of Init under Debian GNU/Linux and how
to interact with some more technical details of dpkg's operation. This
information can be found in the directory `doc/package-developer' at
any Debian GNU/Linux archive. At the time of this writing, the
following documents are available:
`virtual-package-names-list.text'
The list of virtual package names currently in use, together with
the procedure for getting new virtual package names allocated.
`auto-deconfiguration.txt'
How dpkg can sometimes automatically deconfigure packages in order
to do bulk installations smoothly.
`dpkg-essential-flag.txt'
How to tell dpkg a package is essential and should not be removed.
(This is for the use of base system packages only.)
`dpkg-disappear-replace.txt'
What happens when a package appears to have been completely
replaced.
In the future, we hope also to make available:
`copyright.txt'
How to choose a good copyright notice to attach to new programs.
`version-ordering.txt'
The algorithm with which packages' version numbers are compared.
Also, you should download the sample files and the sample package
(GNU Hello) available in `standards/samples'. You may use any of this
material as a starting point for new packages. The following sample
files, incidentally, are available:
* debian.README
* debian.control
* debian.postinst
* debian.postrm
* debian.rules
Some more detailed information about certain topics is available in
the appendix to this document (*note Appendix::.).
File: guidelines.info, Node: Package Copyright, Next: Package Content, Prev: Additional Information, Up: Top
Package Copyright
*****************
Please study the copyright of your submission *carefully* and
*understand it* before proceeding! If you have doubts or questions,
please ask!
In order to understand how we classify and distribute certain
packages, it is important to understand the distinction between being
freely available and being freely redistributable.
Being "freely available", quite simply, means that the software can
be made available freely, at least for non-commercial purposes and in
its original, unmodified form. This includes packages made available
freely that have restrictions on non-commercial use, redistribution of
modifications, etc. Being freely available, therefore, has nothing to
do with being able to modify and redistribute the software. It only
means that you can get a copy of the software without having to pay
(and it does not necessarily mean that you can *use* the software
without having to pay--shareware is an example of freely available
software).
"freely redistributable", while generally being freely available,
goes beyond just being freely available. Freely redistributable means
that that the software, in addition to being able to be made available
freely, must be able to be freely modified and redistributed without
restriction.
All submissions to be included in the distribution proper *must* be
freely redistributable.
In addition to the distribution, the Project maintains two separate
archives of software packages with the distribution: the `contrib'
archive and the `non-free' archive.
`contrib' is an archive of user-contributed packages that are not
maintained by the Project, packages that were once maintained by the
Project but that are no longer actively maintained, and packages that
are maintained by the Project but that are not yet considered ready for
inclusion in the distribution proper (i.e., ALPHA and BETA packages).
As above, all submissions for inclusion in the `contrib' archive *must*
be freely redistributable.
`non-free' is an archive of packages with either restrictive or
unclear terms of copying or modification. If a package has *any*
restrictions on modification or redistribution, it can not be included
in the distribution or `contrib' archive. It can only be included in
the `non-free' archive, and then only if it is freely available.
In summary, in order to be included in the distribution proper or the
`contrib' archive, a package must be *freely redistributable*. Anyone
must be able to make copies of it, modify it, redistribute it with
their modifications in place, include it on a CD-ROM, or generally sell
it. To be included in the `non-free' archive, a package may have
restrictions, as long as the package remains *freely available*. We
must be available to make it available freely at the archive, and anyone
must be able to make copies of it and use it for at least
non-commercial, personal purposes. Software that will typically be
included in `non-free' are software that does not allow commercial
distribution, software that does not allow modification or
redistribution of modifications, commercial "demos", and "shareware".
When in doubt, send mail to `iwj10@cus.cam.ac.uk' and
`imurdock@debian.org'. Be prepared to provide us with the copyright
statement. Software covered by the GPL, public domain software and
BSD-like copyrights are safe; be wary of the phrases "commercial use
prohibited" and "distribution restricted".
Every package submission *must* be accompanied by verbatim copy of
its copyright (with the exceptions of public domain packages and those
covered by the UCB BSD licence or the GNU GPL or LGPL; in these cases
simply indicate which is appropriate). This information must be
included in a file installed to the directory `/usr/doc/copyright'.
See below for details.
File: guidelines.info, Node: Package Content, Next: Source Package, Prev: Package Copyright, Up: Top
Package Content
***************
The following requirements apply equally to both the binary and
source packages. In either case, when files have been installed, they
must conform to the requirements described in this section.
The primary rule in Debian GNU/Linux is to follow the Linux "File
System Standard" ("FSSTND"). The location of installed files *must*
comply *fully* with the FSSTND. The latest version of this document
can be found alongside the Guidelines or at `tsx-11.mit.edu' in
`/pub/linux/docs/linux-standards/fsstnd'. Specific questions about
following the standard should be addressed to Daniel Quinlan, the
FSSTND coordinator, at `quinlan@yggdrasil.com'.
In addition to the FSSTND, all Debian GNU/Linux packages must follow
the guidelines below.
* Directories should be mode 755 or (for group-writability) mode
2775, with the exception of special "system" directories that need
to be another mode. The ownership of the directory should be
consistent with its mode--if a directory is mode 2775, it should
be owned by the group that needs write access to it, of course.
Use common sense in assigning permissions and ownerships to
directories, and make sure that what is done is secure if it is
"non-standard".
* Normal binaries should be mode 755 and owned by `root.root'. If
there is a good reason to use a different mode or ownership, you
may do so, but you must try to be as consistent as possible with
the rest of the system. If you need to use a different mode or
ownership, please discuss it with `imurdock@debian.org'.
* Setuid binaries should normally be mode 4755 (not 4711!) and, of
course, owned by the appropriate user.
* Setgid binaries should normally be mode 2755 (not 2711!) and, of
course, owned by the appropriate group.
* Library files should generally be mode 644 and owned by
`root.root'. If the package requires different permissions or
ownerships to function correctly, they should be used instead.
* Manual pages should be mode 644 and owned by `root.root'. The
`nroff' source must be installed. You should *not* install a
preformatted "cat page", and you should only use sections 1 to
9--see the FSSTND for more details.
* Info documents should be mode 644, owned by `root.root', and
compressed with `gzip -9' when installed. The package must call
`install-info' to update the Info `dir' file. This should be done
in the post-installation script (`postinst'), like this:
install-info --quiet /usr/info/foobar.info
The entries should be removed by the pre-removal script (`prerm'),
like this:
install-info --quiet --remove /usr/info/foobar.info
It is also a good idea to specify a section for the Info `dir'
entry. This is done with the `--section' switch. To determine
which section to use, you should use look at `/usr/info/dir' on
your system and choose the most relevant (or create a new section
if none of the current sections are relevant).
If `install-info' cannot find a description entry in the Info file
you will have to supply one. See `install-info'(8) for details.
* If a package contains any shared libraries you will have to invoke
`ldconfig' in both the `postinst' and `prerm' scripts to correctly
update the library links. See `ldconfig'(8) for details.
* Any additional documentation that comes with the package can be
installed at the discretion of the package maintainer. Text
documentation should be mode 644, owned by `root.root', installed
to `/usr/doc', and compressed with `gzip -9' unless it is small.
If a subdirectory of `/usr/doc' is warranted, please do create one.
Please do not install DVI, PostScript, or large textual
documentation to `/usr/doc'. However, please do upload such
documentation as a separate package so that it can be made
available with the distribution. If a user has the need for the
documentation, they can easily get it from the archive, CD-ROM,
etc., but it should not take up disk space on the machines of the
user who do not need or want it installed.
* Create a file named `/usr/doc/copyright/<package>' which gives
details of the authorship and copyright of the package. If the
package is distributed under the GNU General Public Licence, the
GNU Library General Public Licence or the Regents of the
University of California at Berkeley (BSD) licence, please say so
instead of including a copy of the licence. The files `BSD',
`GPL', and `LGPL' will be available in the `/usr/doc/copyright'
directory for you to refer to. `/usr/doc/copyright/<package>'
should not be compressed.
*All* authorship and copyright information from the original source
package must be included in the `/usr/doc/copyright/<package>'
file.
* Any example files (for example, sample configuration files) should
be placed in the directory `/usr/doc/examples'. If the file is
normally a hidden file, such as `.emacs', then please call it
`dot.emacs', to avoid confusion. Again, you may create a
subdirectory if it is needed.
* All symbolic links should be relative, not absolute. Absolute
links, in general, cause problems when a file system is not
mounted where it "normally" resides (for example, when mounted via
NFS). In certain cases, however, relative links may also cause
similar problems. I have generally made links into `/etc' and
`/var' absolute and all other links relative. There may be other
cases in which absolute links are necessary.
Therefore, in the `Makefile' or `debian.rules', do not do:
install: all
[...]
ln -fs /usr/bin/gcc /usr/bin/cc
[...]
Instead, do:
ln -fs gcc /usr/bin/cc
or
( cd /usr/bin ; ln -fs gcc cc )
Please do not create hard links in the manual page directories. In
these cases, you should use relative symbolic links or files that
`.so' (roff for `source') others instead.
* All command scripts should have a `#!' line naming the shell to be
used to interpret them.
* In the case of Perl scripts this should be `#!/usr/bin/perl' or
sometimes `#!/bin/perl', as follows: if the script is a critical
one that may be called when the `/usr' partition is unmounted or
broken it should use `/bin/perl'. Otherwise (especially if the
script is not specifically targetted at Debian) it should use
Perl's standard location, `/usr/bin/perl'.
* Generally the following compilation parameters should be used:
CC = gcc
CFLAGS = -O2 -g -Wall # sane warning options vary between programs
LDFLAGS = # none (or -N, if appropriate; see below)
install -s (or strip)
Note that all installed binaries should be stripped, either by
using the `-s' flag to `install', or by calling `strip' on the
binaries after they have been copied into the `debian-tmp' but
before the tree is made into a package.
Make sure that you do not link with `-g', as this makes a.out
compilers produce huge statically linked binaries. The `-g' flag
is useful on compilation so that you have available a full set of
debugging symbols in your built source tree, in case anyone should
file a bug report involving (for example) a core dump.
`-N' should only be used on binaries that are very small (less than
8K with the `-N' option, roughly) and are not likely to have
multiple instances in memory. Do not use `-N' on daemons, no
matter how small they are.
It is up to the package maintainer to decide what compilation
options are best for the package. Certain binaries (such as
computationally-intensive programs) may function better with
certain flags (`-O3', for example); feel free to use them. Please
use good judgment here. Don't add flags for the sake of adding
flags; only add flags if there is good reason to do so.
* Please check with the base system maintainer (Ian Murdock) before
using users or groups other than `root' and others specified in
this document.
File: guidelines.info, Node: Source Package, Next: Binary Package, Prev: Package Content, Up: Top
Source Package
**************
The source package should contain a file called `debian.rules' which
contains at least the following targets, to be invoked in the top level
directory:
build
binary
clean
`debian.rules' should start with
#!/usr/bin/make -f
and be executable. It is a good idea to arrange for it not to fail
obscurely when invoked in the wrong directory, for example by testing
for the existence of a file in the source directory.
* The `build' target should perform all non-interactive configuration
and compilation of the package. If a package has an interactive
pre-build configuration routine, the source package should be built
*after* this has taken place.
For some packages, notably ones where the same source tree is
compiled in different ways to produce two binary packages, the
`build' target does not make much sense. For these packages it is
good enough to provide two (or more) targets (`build-a' and
`build-b' or whatever) for each of the ways of building the
package, and a `build' target that does nothing. The `binary'
target will have to build the package in each of the possible ways
and make the binary package out of each.
* The `binary' target of `debian.rules' should be all that is
necessary for the user to build the binary package. The binary
package should be created using `dpkg' and placed in the parent of
the top level directory. The next section describes how to
construct binary packages from the `binary' target.
* The `clean' target should undo the effects of the `build' target
and the `binary' target, except that it should leave alone any
`../<package>-<version>.deb' file created by a run of `binary'.
* Additional targets may exist in `debian.rules'. We recommend using
`source' and `diff' targets to build the Debianised source package
and the Debianisation context diff, respectively. These files
should be placed in `../foo-<version>.tar.gz' and
`../foo-<version>.diff.gz'. The `install' target, for installing
into a running system direct from the Debianised source tree, is
no longer required. The sample `debian.rules' provides `source'
and `diff' targets that should work with little or no alteration,
providing that the package-specific variables at the top of the
script have been properly defined.
* If you need to edit a `Makefile' where `configure' scripts are
used, you should edit the `.in' files rather than editing the
`Makefile' directly. This allows the user to reconfigure the
package if necessary. You should *not* configure the package and
edit the generated `Makefile'! This makes it impossible for
someone else to later reconfigure the package.
* Please document your changes to the source package so that future
package maintainers know what has been changed. To do this,
include a description of your changes in the `debian.README'
(which, as described above, should already contain authorship and
copyright information!) and include relevant information such as
your name, electronic mail address, date, etc.
* If changes to the source code are made that are applicable to Linux
systems or systems in general please try to get them included in
the upstream version of the package by supplying the upstream
authors with the changes in whatever form they prefer.
If changes to the source code are made, please use a `define'. If
they are changes required to compile or function under Linux in
general, use `LINUX'. If it is a cosmetic or functional change,
use `DEBIAN'.
* Create the source package using `tar', and use `gzip -9' to
compress it. Source packages should be named in the form
<package>-<version>.tar.gz--for example, `fileutils-3.9-3.tar.gz'.
NB, here `<version>' is the full Debian version number, in the
form `<original_version>-<debian_revision>' (see below), but the
tarfile should unpack into a directory named
`<package>-<original_version>' (again, see the section below on
version numbering).
* Create the context diff against the original package using `diff
-cNr', and use `gzip -9' to compress it. Context diffs should be
named in the form <package>-<version>.diff.gz--for example,
`fileutils-3.9-3.diff.gz'.
Please note that the package and patch filenames do *not* need to
fit in MS-DOS 8+3. They will be made available under an alternative
8+3 name in the archive by the archive maintainer, using a symlink.
File: guidelines.info, Node: Binary Package, Next: Control Files, Prev: Source Package, Up: Top
Binary Package
**************
The `binary' target of the source package `debian.rules' file should
do the following (see the sample `debian.rules' for an implementation
that you are free to modify and use in your own packages, of course):
* Create an empty directory in the top-level directory of the source
package (deleting it first, if necessary), and install the files
belonging to this package in that directory. For example, the
directory could be called `debian-tmp' and would probably contain
directories `debian-tmp/usr/bin', `debian-tmp/usr/lib', etc.
(`debian-tmp' is the name traditionally used, and it is used in
the sample `debian.rules' file, so we will use that name in the
Guidelines.)
* Make sure that all the files under `debian-tmp' have the correct
ownerships and permissions (*note Package Content::., for more
information about file locations, ownerships, and permissions.)
* Create a subdirectory of `debian-tmp' called `DEBIAN'. This
directory contains the package control information, including at
the very least the master information file named `control'. The
next section describes the semantics and syntax of the files
required and allowed here.
* Run `dpkg' to create the binary package, using something like
dpkg --build debian-tmp
This will create a file called `debian-tmp.deb', from the
`debian-tmp' directory. You should rename this file to
`../<package>-<version>.deb' after it is built.
After the `binary' target has done all this, the
`<package>-<version>.deb' file in the parent directory is the
binary distribution. This file may be distributed and installed on
any Debian GNU/Linux system with `dpkg' in the same manner and
using the same methods as all packages are installed to the system.
* If a single source package corresponds to several binary packages,
there should usually be a `debian.rules' file with a single
`binary' target that builds all the binary packages involved and
move all packages to the parent directory of that containing the
source package.
In this case, you should choose binary package names which are
meant to make clear the close relationship between the binary
packages and which source package the binary packages came from
(for example, the `texinfo' source package build two binary
packages: `texidoc' and `texinfo'). You should place the
appropriate binary package name in the `Package' field of the
control file (not the source package name), and you should
consider whether the other binary packages that come from the same
source tree should be mentioned in the `Depends', `Recommends' or
`Suggests' fields. You should put the source package name in the
`Source' field.
You should retain the source package version numbering in the
`Version' field, if possible--the version number should be the
same for the Debianised source tree and all the binary packages
generated from it. It is more important, though, that the version
numbers sort correctly. See below for details of version numbers.
File: guidelines.info, Node: Control Files, Next: Appendix, Prev: Binary Package, Up: Top
Control Files
*************
Each binary package contains, in addition to the files that comprise
the actual package, a set of text files that control how `dpkg'
installs, configures, upgrades, removes, etc. the package. These files
are called "control files". When creating the package, the control
files should placed in a directory called `DEBIAN', as described
earlier (*note Binary Package::., for further information).
The control information files are:
`control'
The master package control information file.
`conffiles'
A list of package configuration files.
`preinst'
The package pre-installation script.
`postinst'
The package post-installation script.
`prerm'
The package pre-removal script.
`postrm'
The package post-removal script.
Of these, only `control' is required. The various installation
scripts, and the configuration files list, will only be used if they are
present.
* Menu:
* control::
* conffiles::
* Installation and Removal Scripts::
* Dependencies and Conflicts::
* Package Classification Fields::
File: guidelines.info, Node: control, Next: conffiles, Prev: Control Files, Up: Control Files
control
=======
The `control' file contains a number of fields. Each field begins
with a field name, such as `Package' or `Version' (case insensitive),
followed by a colon and optionally some spaces or tabs (a single space
is conventional). Then comes the body of the field, which may be
several lines long; each continuation line must start with at least one
space or tab. (These are the same rules as apply to RFC822 mail
headers.) Blank lines are not permitted in the control file.
The required fields in the control file are the following:
`Package'
The name of the package.
`Description'
The description of the package. How to write an extended and more
usefull description field can be found in *note How to write the
Description control file field::..
`Maintainer'
The name and e-mail address of the maintainer of the package.
`Version'
The version number in the format
`<original_version>-<debian_revision>'.
Each field has a particular format and meaning for the package
installation tools.
The value of `Package' should be the name of the package. Package
names must start with an alphanumeric, must be at least two characters,
and may contain only alphanumerics and the characters - + . @ : = % _
(that is, hyphen, plus, stop, at, colon, equals, percent and
underscore). They are not case sensitive.
The `Maintainer' field should be in the form
Joe J. Bloggs <jbloggs@foo.com>
Note that this will not be useable as an email address if the name
given contains full stop characters, because of a silly restriction in
the Internet mail standards. If you want to use this as an email
address in a program you should check for full stops and change the
string to the form `jbloggs@foo.com (Joe J. Bloggs)' if you find any.
The `Version' field should be the version number of the package.
For most packages which are not written specifically for Debian, this
should be in the form
Version: <original_version>-<debian_revision>
where `<original_version>' is the original package version number in
whatever form the original package uses and `<debian_revision>'
indicates which "debianisation" this is (this should usually be a plain
number or perhaps a two numbers separated by a full stop, and should be
incremented each time the package is changed or updated).
Packages which are written specifically for Debian do not have a
debian_revision, and their version number should simply be version
(which should not contain any hyphens, to avoid confusion).
There is an ordering imposed on version numbers, described in
`version-ordering.txt'. This ordering is designed to `do the right
thing' in most circumstances; if your package has an version number in
an unusual format you may need to reformat it somewhat to get the
ordering right. This is important because `dpkg' is (for example)
reluctant to downgrade packages.
The optional fields in the control file are the following:
`Depends'
The names of prerequisite packages.
`Recommends'
The names of related, recommended packages.
`Suggests'
The names of related, optional packages.
`Conflicts'
The names of packages which conflict with this package.
`Provides'
The names of virtual packages which this package provides.
`Priority'
The `priority' of the package, as shown and used by `dselect'.
`Section'
The `section' of the package, as shown and used by `dselect', and
used as a location for the package in the distribution.
`Essential'
A boolean field used by the base packages.
See below for details of the semantics and syntax of these fields.
Most packages will need at least a `Depends' field.
An example of a `control' file would be:
Package: smail
Version: 3.1.29.1-13
Maintainer: Ian Jackson <iwj10@cus.cam.ac.uk>
Recommends: pine | mailx | elm | emacs | mail-user-agent
Suggests: metamail
Depends: cron, libc5
Conflicts: sendmail
Provides: mail-transport-agent
Description: Electronic mail transport system.
Smail is the recommended mail transport agent (MTA) for Debian.
.
An MTA is the innards of the mail system - it takes messages from
user-friendly mailer programs and arranges for them to be delivered
locally or passed on to other systems as required.
.
In order to make use of it you must have one or more user level
mailreader programs such as elm, pine, mailx or Emacs (which has Rmail
and VM as mailreaders) installed. If you wish to send messages other
than just to other users of your system you must also have appropriate
networking support, in the form of IP or UUCP.
In this case, `mail-user-agent' is a virtual package representing
any user mailer program; the actual package names `pine' is quoted for
the reasons described in `dependency-ordering.txt', and the others
because older versions of those packages do not have the appropriate
`Provides' field.
File: guidelines.info, Node: conffiles, Next: Installation and Removal Scripts, Prev: control, Up: Control Files
conffiles
=========
The contents of `conffiles' is simply a list of configuration files
in the package. When installing the package, `dpkg' uses an
intelligent method to update these files. This will ensure that
package-specific configuration files are not overwritten when a package
is upgraded, unless the user wishes the installation tools to do so.
Typically, files listed in conffiles are package-specific
configuration files, which (according to the Linux Filesystem Standard)
are stored in `/etc'. For example, the `sendmail' package may contain
the file `/etc/sendmail.cf', which we do not wish to overwrite
automatically when the user upgrades the sendmail package. Only those
files listed in `DEBIAN/conffiles' will be updated intelligently when a
package is upgraded; all other files in the package will be overwritten
by the upgrade process.
Configuration files which will be functional as shipped and will
probably need little or no local editing should simply be listed the
`conffiles' file; in this case you need read no further.
For packages whose configuration files will need modification on
most systems there are two sensible approaches. Which one is chosen
depends on how hard the configuration problem is and how much time the
package maintainer has available.
One option is for you to ship a minimal `best-effort' file in
`/etc', and list the file in `conffiles'. This will mean that the user
will have to go and edit the file themselves to get the package to work
properly, of course. The next time they upgrade the package, if you
haven't changed the file version, their old file will be left in place.
If you have modified your version then the user will get a prompt
asking them which version of the file they want, theirs or yours. They
will then usually have to resolve the discrepancies manually.
The other option is to be preferred, if you can do it: do not put a
copy of the configuration file in the package at all. Instead, you
check in the postinst whether the file exists, and if it doesn't you
prompt the user for the information you need to create a good one. This
is obviously harder work.
You also have to remember that you will have to keep up with your
package's changes: if you discover a bug in the program which generates
the configuration file, or if the format of the file changes from one
version to the next, you will have to arrange for the postinst script to
do something sensible--usually this will mean editing the installed
configuration file to remove the problem or change the syntax. You will
have to do this very carefully, since the user may have changed the
file, perhaps to fix the very problem that your script is trying to deal
with--you will have to detect these situations and deal with them
correctly.
If you do go down this route it's probably a good idea to make the
program that generates the configuration file(s) a separate program in
`/usr/sbin', by convention called package`config', and then run that if
appropriate from the post-installation script. The package`config'
program should not unquestioningly overwrite an existing
configuration--if its mode of operation is geared towards setting up a
package for the first time (rather than any arbitrary reconfiguration
later) you should have it check whether the configuration already
exists, and require a `--force' flag to overwrite it.
`conffiles' should almost certainly list all the files contained in
your package in the `/etc' directory. There may also be other files
somewhere that the user is expected to edit, which should also be
included. Note, however, that the FSSTND specifies that configuration
files must be in `/etc'. No Debian package should contain
configuration files in `/usr/etc', and all programs should refer to
configuration files in `/etc'.
For example, the TCP/IP package might use a conffiles which contains
/etc/init.d/netbase
/etc/gateways
/etc/protocols
/etc/services
/etc/hosts.allow
/etc/hosts.deny
/etc/rpc
and so on; the files
/etc/hosts
/etc/inetd.conf
/etc/host.conf
/etc/networks
/etc/resolv.conf
might be generated by an interactive configuration program, and would
then not be included in the package or listed in the `conffiles'.
File: guidelines.info, Node: Installation and Removal Scripts, Next: Dependencies and Conflicts, Prev: conffiles, Up: Control Files
Installation and Removal Scripts
================================
The scripts `preinst', `postinst', `prerm', and `postrm' are
optional (Bash or Perl) scripts. As the names would indicate, if these
scripts exist, they will be executed before installing the package,
after installation, before package removal, and after removal,
respectively.
They are given arguments which indicate the precise situation and
action being performed--see `maintainer-script-args.txt' for details of
exactly when each of the scripts is invoked and what its arguments are.
Extra arguments and situations may be added later, so you should not
test the number of arguments to your script to determine the situation,
and you should choose the sense of your `if it is this then do this
otherwise do that' tests carefully.
These scripts can be used to perform any site-specific package
configuration.
Because the scripts will be exectued by the dpkg front-end, it is
guaranteed that the scripts will be executed interactively. User input
from the scripts should be read from standard input, not the user's
terminal. Similarly, output should be sent to standard output.
If your maintainer scripts need to prompt for passwords and/or do
full-screen interaction should do these things to and from `/dev/tty',
since `dpkg' will at some point redirect scripts' standard input and
output so that it can log the installation process. Likewise, because
these scripts may be executed with standard output redirected into a
pipe for logging purposes, Perl scripts should set unbuffered output by
setting `$|=1' so that the output is printed immediately rather than
being buffered.
The scripts must be idempotent, and they must clean up after
themselves properly. Ie, they must do the right thing if run multiple
times, even if previous runs failed halfway through. This is so that if
any errors occur, or if the `dpkg' run is interrupted, the user can
recover by rerunning `dpkg', and/or by upgrading to a new version and
then rerunning the failed operation.
These scripts should avoid producing output which it is unnecessary
for the user to see and should rely on `dpkg' to stave off boredom on
the part of a user installing many packages. This means, amongst other
things, using the `--quiet' option on `install-info'.
Packages should try to minimise the amount of prompting they need to
do, and they should ensure that the user will only every be asked each
question once. This means that packages should try to use appropriate
shared configuration files (such as `/etc/papersize' and
`/etc/news/server'), rather than each prompting for their own list of
required pieces of information.
It also means that an upgrade should not ask the same questions
again, unless the user has used `dpkg --purge' to remove the package's
configuration. The answers to configuration questions should be stored
in an appropriate place in `/etc' so that the user can modify them, and
how this has been done should be documented.
If a package has a vitally important piece of information to pass to
the user (such as "don't run me as I am, you must edit the following
configuration files first or you risk your system emitting
badly-formatted messages"), it should display this in the `postinst'
script and prompt the user to hit Return to acknowledge the message.
Copyright messages do not count as vitally important (they belong in
`/usr/doc/copyright'; neither do instructions on how to use a program
(these should be in on line documentation, where all the users can see
them).
They should return a zero exit status for success, or a nonzero one
for failure. Note that if a script is a `#!/bin/sh' script it should
probably start with `set -e', to avoid continuing after errors--see
`bash'(1) for details. Perl scripts should check for errors when
making calls such as `open', `print', `close', `rename' and `system'.
If these scripts exist they should be left in the `DEBIAN' directory
with execute permission enabled and should contain an appropriate `#!'
line, such as `#!/bin/bash' for a `bash' script or `#!/bin/perl' for a
Perl script (see above).
File: guidelines.info, Node: Dependencies and Conflicts, Next: Package Classification Fields, Prev: Installation and Removal Scripts, Up: Control Files
Conflicts, Depends, Suggests, Recommends and Provides
=====================================================
The `Depends' field lists packages that are required for this
package to provide a significant amount of functionality. The package
maintenance software will not allow a package to be installed without
also installing packages listed in its `Depends' field, and will run
the `postinst' scripts of packages listed in `Depends' fields before
those of the packages which depend on them, and run the `prerm' scripts
before.
Packages containing dynamically-linked executable binaries (this
includes almost all C programs) should include a `Depends' field which
mentions the shared C library required for the program to run. For
a.out binaries linked against `libc.so.4' the relevant package name is
`libc'; for ELF binaries linked against `libc.so.5' the relevant
package name is `libc5'.
The `Recommends' field lists packages that would be found together
with this one in all but unusual installations. The user-level package
maintenance program `dselect' will warn the user if they select a
package without those listed in its `Recommends' field. Note that
`Recommends' fields do not currently have any implications for the
order in which the maintainer scripts are run.
The `Suggests' field lists packages that are related to this one and
can perhaps enhance its usefulness, but without which installing this
package is perfectly reasonable. The package maintenance software will
not moan at the user for not selecting `Suggests' related packages, but
may use the information in the `Suggests' field to assist the user
during package selection.
The syntax of `Depends', `Recommends' and `Suggests' is a list of
groups of alternative packages. Each group is a list of packages
separated by vertical bar (or `pipe') symbols, `|'. The groups are
separated by commas. Each package is a package name optionally
followed by a version number specification in parentheses. A version
number may start with a `>=', in which case that version or any later
will match, or `<=' for that version or any earlier version. A version
number starting with a `>>' or `<<' will respectively match any later
or earlier version. If a version number or a version number starting
with `=' is specified an exact match is required. Commas are to be read
as `AND', and pipes as `OR', with pipes binding more tightly.
Versions of dpkg before 1.0.9 used `<' and `>' for `<=' and `>='
(these are still supported for backward compatibility), and did not
support `<<' and `>>'.
The `Conflicts' field lists packages that conflict with this one,
for example by containing files with the same names (an example would
be Smail vs. Sendmail). The package maintenance software will not
allow conflicting packages to be installed. Two conflicting packages
should each include a `Conflicts' line mentioning the other.
The syntax of `Conflicts' is a list of package names (with optional
version numbers), separated by commas (and optional whitespace). In
the `Conflicts' field the comma should be read as `OR'.
The `Provides' field lists the names of any `virtual packages' of
which this packages is to be considered an instantiation. Virtual
packages are used to allow packages to refer to a service they require
(such as the availability of `/usr/sbin/sendmail') without having to
know the names of all the relevant packages. The virtual package names
defined in `Provides' fields may be used in other packages' `Depends',
`Recommends', `Suggests' and `Conflicts' fields. For more information
about how to use virtual packages and which virtual package names to
use read *note Virtual dependencies::. and
`doc/package-developer/virtual-package-names-list.text'.
The syntax of `Provides' is a list of package names separated by
commas (and optional whitespace).
File: guidelines.info, Node: Package Classification Fields, Prev: Dependencies and Conflicts, Up: Control Files
Priority, Section and Essential
===============================
The `Priority' and `Section' fields are used by `dselect' when
displaying the list of packages to the user. There is no need to put
them into a package, since these are usually set by the distribution
maintainers in the `Packages' file.
However, if a user installs a package which is not part of the
standard distribution, or without downloading and updating from a new
`Packages' file, the information about the priority and section of a
package will be absent, and the `dselect' package listing will have the
package listed under `unclassified'. It is permissible for a package
to include `Section' or `Priority' fields to improve this; however, if
you do this you should make sure you keep the information up to date so
that users are not shown conflicting information. The `Section' field
can also be used by the distribution maintainers as a suggestion about
which section you think is most appropriate for your package.
The `Essential' field should only appear in packages in the
installation's base system. If it is set to `yes' then `dpkg' will not
remove the package even if asked to, and will make certain minor
modifications to its installation procedures. The only other legal
value is `no', which is equivalent to the absence of the field.
File: guidelines.info, Node: Appendix, Prev: Control Files, Up: Top
Appendix
********
* Menu:
* configuration files - /etc/skel vs /usr/doc/examples::
* How to write the Description control file field::
* Configuration of init::
* Maintainer script arguments and how `dpkg' does things::
* Mail processing packages::
* Virtual dependencies::
This is Info file guidelines.info, produced by Makeinfo-1.63 from the
input file guidelines.texi.
START-INFO-DIR-ENTRY
* debian-guidelines: (debian-guidelines). How to make Debian packages.
END-INFO-DIR-ENTRY
File: guidelines.info, Node: configuration files - /etc/skel vs /usr/doc/examples, Next: How to write the Description control file field, Prev: Appendix, Up: Appendix
configuration files - /etc/skel vs /usr/doc/examples
====================================================
There seems to be a certain amount of confusion about `/etc/skel'
and `/usr/doc/examples'. The most important thing to remember is the
following:
Files in `/etc/skel' will *automatically* be copied into *new* user
accounts by `adduser'. They should not be referenced there by any
program. Files in `/usr/doc/examples' should not be installed
automatically.
Therefore, if the program in question need a dotfile to exist in
advance in `$HOME' to work *sensibly* that dotfile should be installed
in `/etc/skel' (and listed in conffiles; *note conffiles::.).
However, programs that require dotfiles in order to operate sensibly
(dotfiles that they do not create themselves automatically, that is) are
a bad thing, and that programs should be configured by the Debian
default installation as close to normal as possible.
Therefore, if a program in a Debian package needs to be configured in
some way in order to operate sensibly that configuration should be done
in a site-wide global configuration file elsewhere in `/etc' (and that
file should be listed in conffiles). Only if the program doesn't
support a site-wide default configuration should a default per-user file
be placed in `/etc/skel' (and listed in conffiles; *note conffiles::.).
The idea is as follows:
The sysadmin should ideally not have to do any configuration other
than that done (semi-)automatically by the postinst script.
However, if they wish to change their configuration themselves
(because the configuration they want is beyond the scope of the
autoconfiguration, or because the autoconfiguration doesn't exist yet,
or because they just want to do it themselves for any reason) then
`/usr/doc/examples' exists as *documentation* for their benefit.
The only time these files should be read are by the sysadmin using
their favourite editor or pager, or *perhaps* (in very complex packages)
by the postinst as a template to build on or modify.
`/etc/skel' is part of the *implementation* of this configuration.
It contains the files that are copied into new user accounts. It
should probably be as empty as we can make it.
Examples:
`.profile'
`/etc/skel' should not contain a `.profile' file. Anything that
needs to be done there should be done in `/etc/profile'. Anything
that should not go in `/etc/profile' (users can't avoid running
`/etc/profile') probably should not be in the default
configuration. bash has generally good default behaviour.
`.bash_logout'
Likewise, bash functions perfectly happily without a
`.bash_logout', so none should be provided, since anything in it is
a deviation from the sensible default behaviour.
`.xsession'
`/etc/skel' should not contain a `.xsession'. `xdm''s system-wide
startup file `/usr/lib/X11/xdm/Xsession' supports a system-wide
default user configuration (which should probably be
`/etc/X11/Xsession' or some such) which may be overridden by
`.xsession' in the user's home directory. Therefore there is no
need for a `.xsession' to be installed by default and none should
be provided.
Instead, a sensible `/etc/X11/Xsession' should be provided, and if
desired this can be used as a template by users who wish to install
their own configuration, or alternatively a more comprehensive
example with much commented-out interesting stuff could be put in
`/usr/doc/examples'.
If the sysadmin wishes to change the system-wide default they
should probably do this by editing `/etc/X11/Xsession' rather than
creating the file in `/etc/skel', because the former will affect
all user accounts that haven't explicitly overridden things by
creating their own file while the latter will only affect new
accounts.
All the configuration necessary for a program to function should be
provided. Therefore sysadmins will not need to go through
`/usr/doc/examples' while editing configuration files in `/etc'
except in extreme cases (like INN) where the configuration was too
difficult to do automatically (?? erick).
`site-wide defaults'
Site-wide defaults should not go in `/etc/skel'. In the case of
twm, for example, the system-wide default should be in
`/etc/X11/system.twmrc'. (The default location for this in X11R5,
btw, is in `/usr/lib/X11' somewhere, but we can't put it on `/usr'
because of CDROM distributions, etc - hence the FSSTND's mandate
to put configuration files in `/etc'.)
`.twmrc'
There should be no `.twmrc' file in `/etc/skel'. You can have one
in `/usr/doc/examples' if you *like*, but why bother if
`system.twmrc' is a good example (and indeed is the one the user is
using before they create their own)?
`m4'
`/usr/doc/examples' isn't mainly for example *configuration
files*. It's for any kind of example file distributed with a
package. For example, GNU m4 comes with a whole pile of example
m4 macro scripts, which is exactly what `/usr/doc/examples' is for.
Summary
Files that should be installed in new user accounts should be in
`/etc/skel', as that will ensure that they *are* installed in new user
accounts! However, we should try to avoid the need for this.
`/usr/doc/examples' is just what it says: documentation in the form
of examples. If a sysadmin is required to go and read these files for
their system to work they should be told about it. For example, here
is what the Smail postinst script says right at the start:
I can do certain kinds of automatic configuration of your
mail system, by asking you a number of questions. Later you
may to confirm and/or correct your answers. In any case,
comprehensive information on configuring Smail is in
smail(5) and in /usr/doc/examples/smail and
/usr/doc/smail-admin-guide.
File: guidelines.info, Node: How to write the Description control file field, Next: Configuration of init, Prev: configuration files - /etc/skel vs /usr/doc/examples, Up: Appendix
How to write the Description control file field
===============================================
The format of the `Description' field is as follows:
Description: <single line synopsis>
<extended description over several lines>
The extended description has several kinds of line:
* Those starting with a single space are part of a paragraph.
Successive lines of this form will be word-wrapped when displayed.
The leading space will usually be stripped off.
* Those starting with two or more spaces. These will be displayed
verbatim. If the display cannot be panned horizontally the
displaying program will linewrap them `hard' (ie, without taking
account of word breaks). If it can they will be allowed to trail
off to the right. None, one or two initial spaces may be deleted,
but the number of spaces deleted from each line will be the same
(so that you can have indenting work correctly, for example).
* Those containing a single space followed by a single full stop
character. These are rendered as blank lines. This is the *only*
way to get a blank line - see below.
* Those containing a space, a full stop and some more characters.
These are for future expansion. *Do not* use them.
IMPORTANT and not so important TIPS:
* *Always* start extended description lines with at least *one*
whitespace character. Fields in the control file and in the
Packages file are separated by field names starting in the first
column, just as in RFC822. Forgetting the whitespace will cause
`dpkg-deb' (>=0.93.23) to produce a syntax error when trying to
build the package. If you force it to build anyway `dpkg' will
refuse to install the resulting mess.
* *Do not* include any completely *empty* lines. These separate
different records in the Packages file, and are forbidden in
control files. See the previous paragraph for what happens if you
get this wrong.
* The single line synopsis should be kept brief - certainly under 80
characters. `dselect' displays the *first 49* characters if
you're using an 80-column terminal.
* Do not include the package name in the synopsis line. The display
software knows how to display this already, and you do not need to
state it. Remember that in many situations the user may only see
the synopsis line - make it as informative as you can.
* The extended description should describe what the package does and
how it relates to the rest of the system (in terms of, for
example, which subsystem it is which part of).
* Put important information first, both in the synopis and extended
description. Sometimes only the first part of the synopsis or of
the description will be displayed. You can assume that there will
usually be a way to see the whole extended description.
* You may include information about dependencies and so forth in the
extended description, if you wish.
* Do not use tab characters. Their effect is not predictable.
Example control file for Smail:
Package: smail
Version: 3.1.29.1-13
Maintainer: Ian Jackson <iwj10@cus.cam.ac.uk>
Recommends: pine | mailx | elm | emacs | mail-user-agent
Suggests: metamail
Depends: cron, libc5
Conflicts: sendmail
Provides: mail-transport-agent
Description: Electronic mail transport system.
Smail is the recommended mail transport agent (MTA) for Debian.
.
An MTA is the innards of the mail system - it takes messages from
user-friendly mailer programs and arranges for them to be delivered
locally or passed on to other systems as required.
.
In order to make use of it you must have one or more user level
mailreader programs such as elm, pine, mailx or Emacs (which has Rmail
and VM as mailreaders) installed. If you wish to send messages other
than just to other users of your system you must also have appropriate
networking support, in the form of IP or UUCP.
File: guidelines.info, Node: Configuration of init, Next: Maintainer script arguments and how `dpkg' does things, Prev: How to write the Description control file field, Up: Appendix
Configuration of init
=====================
The `/etc/init.d' directory contains the scripts executed by init(8)
when init state (or "runlevel") is changed. This includes the boot
process, when the multi-user state begins. Several of these scripts
are included with init and are intended to be executed *once*, usually
at boot time. An example is `/etc/init.d/boot', which is executed at
boot time to check and mount file systems, activate swap, load kernel
modules, etc.-everything that needs to be done before the multi-user
state begins. `/etc/init.d' also contains the scripts that are
executed when entering runlevel 0 (halt), runlevel 1 (single-user) and
runlevel 6 (reboot).
Packages can (and should) place scripts in `/etc/init.d' to start or
stop services at boot time or during a change of runlevel. These
scripts should be named `/etc/init.d/'<package>, and they should accept
one of two arguments: "start", which starts the services, or "stop",
which stops the services. This script should not fail obscurely when
the configuration files remain but the package has been removed, as the
default in dpkg is to leave configuration files on the system after the
package has been removed. Only when it is executed with the `-purge'
option will dpkg remove configuration files. Therefore, you should
include a `test' statement at the top of the script, like this:
test -f <program-executed-later-in-script> || exit 0
These scripts should be referenced, when appropriate, by symbolic
links in the `/etc/rc?.d' directories, as below.
When changing runlevels, init looks in the directory `/etc/rc<n>.d'
for the scripts it should execute, where <n> is the runlevel that is
being changed to. Please note that the "scripts" in `/etc/rc?.d' are
not actually scripts; they are symbolic links, referencing actual
scripts in `/etc/init.d'. For simplicity, we refer to them as
"scripts".
First, the scripts prefixed with a "K" are executed, followed by the
scripts prefixed with an "S". The "K" scripts are responsible for
killing certain services and the "S" scripts for starting certain
services upon *entering* the runlevel. For example, if we are changing
from runlevel 2 to runlevel 3, init will first execute all of the "K"
prefixed scripts it finds in `/etc/rc3.d' (to kill services), and then
all of the "S" prefixed scripts it finds in `/etc/rc3.d' (to start
services). The "K" scripts will execute the file it references with an
argument of "stop", and the "S" scripts will execute this file with an
argument of "start".
After the "K" or "S" prefix, there should be a number specified, and
this number should be between 00 and 99. The number determines the
order in which the scripts are run. For example, the "K20" scripts will
be executed before the "K30" scripts. You can use this number to make
sure that a certain service is started before another. For example, on
some machines, the program `setserial' may need to properly set an IRQ
before the `ppp' program uses a modem to connect to a network. In this
case, the script that runs `setserial' should have a lower number than
the script that starts `ppp' so that it runs first:
`/etc/rc2.d/S10setserial'
`/etc/rc2.d/S20ppp'
If it does not matter when or in which order the script is run, use
the number "20". If it does, then you should talk to the maintainer of
the `sysvinit' package or post to `debian-devel', and they will help
you choose a number.
In Debian GNU/Linux, we try to ship our software in as much of a
"default" state as possible. Therefore, unless there is a good reason
for doing differently, we ask that you start and stop the services in
each of the multi-user state runlevels (2, 3, 4, and 5). If a service
needs to be stopped before a file system can be unmounted (an example is
process accounting or quota services), then be sure to stop them in the
halt runlevel (0), the single-user runlevel (1) and the reboot runlevel
(6).
The system administrator will have the opportunity to customize
runlevels by simply adding, moving, or removing the symbolic links in
`/etc/rc?.d'. This is why we default to running everything in the
multi-user state-a reasonable default-and the administrator can easily
customize init to be as complex and sophisticated as he or she wants it
to be beyond this.
We provide a script, `update-rc.d', to make it easier for package
maintainers to arrange for the proper creation and removal of
`/etc/rc?.d' symbolic links from their postinst and postrm scripts.
You should use this script to make changes to `/etc/rc?.d' and *never*
include any `/etc/rc.?.d' symbolic links in the actual archive.
* In the postinst script, you need only do the following to setup
`/etc/rc?.d'. You should redirect standard output to `/dev/null',
as `update-rc.d' produces insignificant output:
update-rc.d <package> default >/dev/null
where <package> is the name of the file as it appears in
`/etc/init.d'. It will use the default number of "20", as
mentioned above. If you need to use a different number, you can
specify it after "default":
update-rc.d <package> default 30 >/dev/null
* In the postrm script, you need only do the following *if and only
if* it is called with the `purge' argument:
if [ "$1" (?? erick) = "purge" ]
then
update-rc.d <package> remove >/dev/null
fi
Important Note:
---------------
*Do not* include the `/etc/rc?.d/*' symbolic links in the archive!
*This will cause problems!* You should create them with update-rc.d,
as above.
*Do not* include the `/etc/rc?.d/*' symbolic links in conffiles!
*This will cause problems!* *Do*, however, include the `/etc/init.d'
scripts in conffiles.
Example:
--------
The process accounting package wants to make sure that process
accounting is started at boot time and that it is stopped before the
system is halted, enters the single-user state, or is rebooted (all so
(?? erick: "all so" or "also") the `/var' file system can be properly
unmounted). It puts a script that does this in `/etc/init.d', naming
the script appropriately "acct". This script accepts one of two
arguments: either "start", which starts process accounting, or "stop",
which stops it. To ensure that it does not fail obscurely when the
configuration files remain but the package has been removed, we include
a `test' statement at the top of the script:
#! /bin/sh
#
# Start process accounting.
. /etc/init.d/functions
test -f /usr/sbin/accton || exit 0
case "$1" in
start)
echo "Starting process accounting"
/usr/sbin/accton /var/account/pacct
;;
stop)
echo "Stopping process accounting"
/usr/sbin/accton
;;
*)
echo "Usage: /etc/init.d/acct {start|stop}"
exit 1
esac
exit 0
You may find a skeletal script from which to base your `/etc/init.d'
scripts in `/etc/init.d/skeleton'.
We want to stop then (re)start process accounting when entering a
multi-user state-runlevels 2, 3, 4, and 5-and we want to stop it when
leaving such a state-runlevels 0 (halt), 1 (single) and 6 (reboot).
These are good defaults, and we accomplish this by including the
following in the postinst:
update-rc.d acct default >/dev/null
When the user removes the acct packages with the `-purge' option, we
want to make sure the `/etc/rc?.d' symbolic links are properly removed,
so we include the following in the postrm:
update-rc.d acct remove >/dev/null
Otherwise, the `/etc/rc?.d' symbolic links will remain on the system
along with `/etc/init.d/acct' script.
File: guidelines.info, Node: Maintainer script arguments and how `dpkg' does things, Next: Mail processing packages, Prev: Configuration of init, Up: Appendix
Maintainer script arguments and how `dpkg' does things
======================================================
This appendix describes exactly how maintainer scripts are called,
with what arguments, in what order, and what `dpkg' does in between.
In all cases version numbers are <version>-<revision>, if the package
has both, or just <version>. `upgrade' is used even when the new
version number looks lower than the old.
Summary
-------
<new preinst> install
<new preinst> install <old-version>
<new preinst> upgrade <old-version>
<old preinst> abort-upgrade <new-version>
<postinst> configure
<old postinst> abort-upgrade <new version>
<conflictor's postinst> abort-remove in-favour <package> <new version>
<deconfigured's postinst> abort-deconfigure \
in-favour <package-being-installed-but-failed> <version>
removing <conflicting-package> <version>
<prerm> remove
<old prerm> upgrade <new version>
<new prerm> failed-upgrade <old-vppersion>
<conflictor's prerm> remove in-favour <package> <new version>
<deconfigured's prerm> deconfigure \
in-favour <package-being-installed> <version> \
removing <conflicting-package> <version>
<postrm> remove
<postrm> purge
<old postrm> upgrade <new-version>
<new postrm> failed-upgrade <old-version>
<new postrm> abort-install
<new postrm> abort-install <old-version>
<new postrm> abort-upgrade <old-version>
<disappearer's postrm> disappear <overwriter> <new version>
Details of unpack phase of installation or upgrade
--------------------------------------------------
The procedure on installation/upgrade/overwrite/disappear (ie, when
running `dpkg --unpack', or the unpack stage of `dpkg --install') is as
follows. In each case if an error occurs the actions in are general
run backwards - this means that the maintainer scripts are run with
different arguments in reverse order. These are the `error unwind'
calls listed below.
1.
a. If a version the package is already installed, call
<old prerm> upgrade <new version>
b. If this gives an error (ie, a non-zero exit status), dpkg will
attempt instead:
<new prerm> failed-upgrade <old-version>
error unwind, for both the above cases:
<old postinst> abort-upgrade <new version>
2. If a `conflicting' package is being removed at the same time:
a. If any packages depended on that conflicting package and
`--auto-deconfigure' is specified, call, for each such
package:
<deconfigured's prerm> deconfigure \
in-favour <package-being-installed> <version> \
removing <conflicting-package> <version>
error unwind:
<deconfigured's postinst> abort-deconfigure \
in-favour <package-being-installed-but-failed> <version>
removing <conflicting-package> <version>
The deconfigured packages are marked as requiring
configuration, so that if -install is used they will be
configured again if possible.
b. To prepare for removal of the conflicting package, call:
<conflictor's prerm> remove in-favour <package> <new version>
error unwind:
<conflictor's postinst> abort-remove in-favour <package> <new version>
3.
a. If the package is being upgraded, call
<new preinst> upgrade <old-version>
b. otherwise, if the package had some configuration files from a
previous version installed (ie, it is in the conffiles-only
state):
<new preinst> install <old-version>
c. otherwise (ie, the package was completely purged):
<new preinst> install
error unwind versions, respectively:
<new postrm> abort-upgrade <old-version>
<new postrm> abort-install <old-version>
<new postrm> abort-install
4. The new package's files are unpacked, overwriting any that may be
on the system already, for example any from the old package or
from another package (backups of the old files are left around,
and if anything goes wrong dpkg will attempt to put them back as
part of the error unwind).
5.
a. If the package is being upgraded, call
<old postrm> upgrade <new-version>
b. If this fails, dpkg will attempt:
<new postrm> failed-upgrade <old-version>
error unwind, for both cases:
<old preinst> abort-upgrade <new-version>
This is the point of no return - if dpkg gets this far, it
won't back off past this point if an error occurs. This will
leave the package in a fairly bad state, which will require a
successful reinstallation to clear up, but it's when dpkg starts
doing things that are irreversible.
6. Any files which were in the old version of the package but not in
the new are removed.
7. The new file list replaces the old.
8. The new maintainer scripts replace the old.
9. Any packages all of whose files have been overwritten during the
installation, and which aren't required for dependencies, are
considered to have been removed. For each such package,
a. dpkg calls:
<disappearer's postrm> disappear <overwriter> <new version>
b. The package's maintainer scripts are removed.
c. It is noted in the status database as being in a sane state,
namely not installed (any conffiles it may have are ignored).
Note that disappearing packages do not have their prerm
called, because dpkg doesn't know in advance that the package
is going to vanish.
10. Any files in the package we're unpacking that are also listed in
the file lists of other packages are removed from those lists.
(This will lobotomise the file list of the `conflicting' package
if there is one.)
11. The backup files made at 4. are deleted.
12. The new package's status is now sane, and recorded as `unpacked'.
Here is another point of no return - if the conflicting package's
removal fails we do not unwind the rest of the installation; the
conflicting package is left in a half-removed limbo.
13. If there was a conflicting package we go and do the removal
actions, starting from point 2. of the removal, below.
Details of configuration
------------------------
When we configure a package (this happens with `dpkg --install', or
with `--configure'), we first update the conffiles and then call:
<postinst> configure
(I'm planning to make available as an argument the version of the
most recent successfully-configured version of the package.) (?? erick:
third person)
No attempt is made to unwind after errors during configuration.
Details of removal and/or configration purging
----------------------------------------------
1. <prerm> remove
2. The package's files are removed (except conffiles).
3. <postrm> remove
4. All the maintainer scripts except the postrm are removed.
If we aren't purging the package we stop here. Note that packages
which have no postrm and no conffiles are automatically purged
when removed, as there is no difference except for the dpkg status.
5. The conffiles and any backup files (`~'-files, `#*#' files,
`%'-files, .dpkg-{old,new,tmp}, etc.) are removed.
6. <postrm> purge
7. The package's file list is removed.
No attempt is made to unwind after errors during removal.
File: guidelines.info, Node: Mail processing packages, Next: Virtual dependencies, Prev: Maintainer script arguments and how `dpkg' does things, Up: Appendix
Mail processing packages
========================
Debian packages which process electronic mail (whether
mail-user-agents (MUA) or alternative mail-transport-agents (MTA))
*must* make sure that they are compatible with the configuration
decisions below. Failure to do this may result in lost mail, broken
`From:' lines, and other serious brain damage!
* The mail spool is `/var/spool/mail' and the interface to send a
mail message is `/usr/sbin/sendmail' (as per the FSSTND). The mail
spool is part of the base and not part of the MTA package.
* Mailboxes are locked using the `.lock' lockfile convention, rather
than fcntl, flock or lockf.
* Mailboxes are generally 660 `<user>.mail' unless the user has
chosen otherwise. A MUA may remove a mailbox (unless it has
nonstandard permissions) in which case the MTA or another MUA must
recreate it if needed. Mailboxes must be writeable by group mail.
* The mail spool is 2775 mail.mail, and MUA's need to be setgid mail
to do the locking mentioned above (and obviously need to avoid
accessing other users' mailboxes using this privilege).
* `/etc/aliases' is the source file for the system mail aliases (e.g.
postmaster, usenet, etc.) - it is the one which the sysadmin and
postinst scripts may edit.
* The convention of writing `forward to <address>' in the mailbox
itself is not supported. Use a `.forward' file instead.
* The location for the `rmail' program used by UUCP for incoming mail
is `/usr/sbin/rmail', as per the FSSTND. Likewise, `rsmtp', for
receiving batch-SMTP-over-UUCP, is in `/usr/sbin/rsmtp' if it is
supported.
* Smail is not using HoneyDanBer UUCP, whose uux apparently accepts
-a and -g options.
* If you need to know what name to use (for example) on outgoing
news and mail messages which are generated locally, you should use
the file `/etc/mailname'. It will contain the portion after the
username and `@' sign for email addresses of users on the machine
(followed by a newline).
A package should check for the existence of this file. If it exists
it should use it without comment (1). If it does not exist it should
prompt the user for the value and store it in `/etc/mailname' as well
as using it in the package's configuration. The prompt should make it
clear that the name will not just be used by that package. E.g., in
the same situation the INN package says:
Please enter the `mail name' of your system. This is the hostname
portion of the address to be shown on outgoing news and mail messages.
The default is `$syshostname', your system's host name.
Mail name [`$syshostname']:
($syshostname is the output of `hostname -fqdn').
---------- Footnotes ----------
(1) An MTA's prompting configuration script may wish to prompt the
user even if it finds this file exists.
File: guidelines.info, Node: Virtual dependencies, Prev: Mail processing packages, Up: Appendix
Virtual dependencies
====================
Virtual packages are in the same namespace as real packages, and may
have the same name. The meaning of a virtual package in a
dependency/conflicts list is exactly that of listing all the real
packages which state that they are an instantiation of that virtual
package.
This is done with a new Provides field in the control file, with a
syntax much like the Conflicts field.
The idea is that we can have something like:
Package: elm
Depends: mta
Package: smail
Provides: mta
Conflicts: mta
Package: sendmail
Provides: mta
Conflicts: mta
The result is equivalent to elm having said
Package: elm (?? erick: added this line)
Depends: smail | sendmail
(There'll be a special case to say that a package may conflict with a
virtual package which it provides - clearly ...)
If there are both a real and a virtual package of the same name then
the dependency may be satisfied (or the conflict caused) by either the
real package or any of the virtual packages which provide it. This is
so that, for example, supposing we have
Package: lout
Optional: ghostview
(this is a fictional example - the Lout package should not mention
ghostview), and someone else comes up with a nice PostScript previewer,
then they can just say
Package: marvelpostview
Provides: ghostview
and all will work in the interim (until, say, the Lout maintainer
changes things).
If a dependency or a conflict has a version number attached then only
real packages will be considered to see whether the relationship is
satisfied (or prohibited, for a conflict) - it is assumed that a real
package which provides virtual package is not of the `right' version.
If there is demand it can be arranged that a package which provides a
virtual package may mention a version number, though this is unlikely to
be helpful:
Provides: mta (2.0)
If you want to specify which of a set of real packages should be the
default to satisfy a particular dependency on a virtual package, you can
simply list the real package as alternative before the virtual one:
Package: xbaseR6
Recommended: xsvga | x-server
Provides: x-base, xr6shlib
Package: xsvga
Recommended: x-base
Provides: x-server
Package: x8514
Recommended: x-base
Provides: x-server
Reply to: