use a smaller type for flags storage in the cache

[apt.git] / doc / acquire-additional-files.txt

# Acquire additional files in 'update' operations

The download and verification of data from multiple sources in different
compression formats, with partial downloads and patches is an involved
process which is hard to implement correctly and securely.

APT frontends share the code and binaries to make this happen in libapt
with the Acquire system, supported by helpers shipped in the apt package
itself and additional transports in individual packages like
apt-transport-https.

For its own operation libapt needs or can make use of Packages, Sources
and Translation-* files, which it will acquire by default, but
a repository might contain more data files (e.g.  Contents) a frontend
might want to use and would therefore need to be downloaded as well
(e.g. apt-file).

This file describes the configuration scheme such a frontend can use to
instruct the Acquire system to download those additional files.

# The Configuration Stanza

The Acquire system uses the same configuration settings to implement the
files it downloads by default. These settings are the default, but if
they would be written in a configuration file the configuration
instructing the Acquire system to download the Packages files would look
like this (see also apt.conf(5) manpage for configuration file syntax):

	Acquire::IndexTargets::deb::Packages {
		MetaKey "$(COMPONENT)/binary-$(ARCHITECTURE)/Packages";
		ShortDescription "Packages";
		Description "$(SITE) $(RELEASE)/$(COMPONENT) $(ARCHITECTURE) Packages";

		flatMetaKey "Packages";
		flatDescription "$(SITE) $(RELEASE) Packages";

		Optional "false";
	};

All files which should be downloaded (nicknamed 'Targets') are mentioned
below the Acquire::IndexTargets scope. 'deb' is here the type of the
sources.list entry the file should be acquired for. The only other
supported value is hence 'deb-src'. Beware: You can't specify multiple
types here and you can't download the same (evaluated) MetaKey from
multiple types!

After the type you can pick any valid and unique string which preferable
refers to the file it downloads (In the example we picked 'Packages').
This string is used as identifier for the target class and accessible as
'Created-By' e.g. in the "apt-get indextargets" output as detailed below.

All targets have three main properties you can define:
* MetaKey: The identifier of the file to be downloaded as used in the
  Release file.  It is also the relative location of the file from the
  Release file.  You can neither download from a different server
  entirely (absolute URI) nor access directories above the Release file
  (e.g. "../../").
* ShortDescription: Very short string intended to be displayed to the
  user e.g.  while reporting progress. apt will e.g. use this string in
  the last line to indicate progress of e.g. the download of a specific
  item.
* Description: A preferable human understandable and readable identifier
  of which file is acquired exactly. Mainly used for progress reporting
  and error messages. apt will e.g. use this string in the Get/Hit/Err
  progress lines.

Additional optional properties:
* flat{MetaKey,Description}: APT supports two types of repositories:
  dists-style repositories which are the default and by far the most
  common which are named after the fact that the files are in an
  elaborated directory structure.  In contrast a flat-style repositories
  lumps all files together in one directory.  Support for these flat
  repositories exists mainly for legacy purposes only.  It is therefore
  recommend to not set these values.
* Optional: The default value is 'true' and should be kept at this
  value.  If enabled the acquire system will skip the download if the
  file isn't mentioned in the Release file. Otherwise this is treated as
  a hard error and the update process fails. Note that failures while
  downloading (e.g. 404 or hash verification errors) are failures,
  regardless of this setting.


The acquire system will automatically choose to download a compressed
file if it is available and uncompress it for you, just as it will also
use pdiff patching if provided by the repository and enabled by the
user. You only have to ensure that the Release file contains the
information about the compressed files/pdiffs to make this happen.
NO properties have to be set to enable this.

# More examples

The stanzas for Translation-* files as well as for Sources files would
look like this:

Acquire::IndexTargets {
	deb::Translations {
		MetaKey "$(COMPONENT)/i18n/Translation-$(LANGUAGE)";
		ShortDescription "Translation-$(LANGUAGE)";
		Description "$(SITE) $(RELEASE)/$(COMPONENT) Translation-$(LANGUAGE)";

		flatMetaKey "$(LANGUAGE)";
		flatDescription "$(SITE) $(RELEASE) Translation-$(LANGUAGE)";
	};

	deb-src::Sources {
		MetaKey "$(COMPONENT)/source/Sources";
		ShortDescription "Sources";
		Description "$(SITE) $(RELEASE)/$(COMPONENT) Sources";

		flatMetaKey "Sources";
		flatDescription "$(SITE) $(RELEASE) Sources";

		Optional "false";
	};
};

# Substitution variables

As seen in the examples, properties can contain placeholders filled in
by the acquire system. The following variables are known; note that
unknown variables have no default value nor are they touched: They are
printed as-is.

* $(SITE): An identifier of the site we access as seen in sources.list,
  e.g. "http://example.org/debian" or "file:/path/to/a/repository". You
  can't use this field in {,flat}MetaKey, it is for description proposes
  only.
* $(RELEASE): This is usually an archive- or codename, e.g. "stable" or
  "stretch".  Note that flat-style repositories do not have a archive-
  or codename per-se, so the value might very well be just "/" or so.
  Again, as seen in the sources.list.
* $(COMPONENT): as given in the sources.list, e.g. "main", "non-free" or
  "universe".  Note that flat-style repositories again do not really
  have a meaningful value here.
* $(LANGUAGE): Values are all entries (expect "none") of configuration
  option Acquire::Languages, e.g. "en", "de" or "de_AT".
* $(ARCHITECTURE): Values are all entries of configuration option
  APT::Architectures (potentially modified by sources.list options),
  e.g. "amd64", "i386" or "armel" for the 'deb' type. In type 'deb-src'
  this variable has the value "source".

Note that while more variables might exist in the implementation, these
are to be considered undefined and their usage strongly discouraged. If
you have a need for other variables contact us.

# Accessing files

Do NOT hardcode specific file locations, names or compression types in
your application! You will notice that the configuration options give
you no choice over where the downloaded files will be stored. This is by
design so multiple applications can download and use the same file
rather than each and every one of them potentially downloads and uses
its own copy somewhere on disk.

"apt-get indextargets" can be used to get the location as well as other
information about all files downloaded (aka: you will see Packages,
Sources and Translation-* files here as well). Provide a line of the
default output format as parameter to filter out all entries which do
not have such a line. With --format, you can further more define your
own output style. The variables are what you see in the output, just all
uppercase and wrapped in $(), as in the configuration file.

To get all the filenames of all Translation-en files you can e.g. call:
	apt-get indextargets --format '$(FILENAME)' "Created-By: Translations" "Language: en"

The line-based filtering and the formating is rather crude and feature-
less by design, so it is recommend to use dedicated and more powerful
tools like 'grep-dctrl'.

Accessing this information via libapt is done by reading the
sources.lists (pkgSourceList), iterating over the metaIndex objects this
creates and calling GetIndexTargets() on them. See the sourcecode of
"apt-get indextargets" for a complete example.

Note that by default targets are not listed if they weren't downloaded.
If you want to see all targets, you can use the --no-release-info, which
also removes the Codename, Suite, Version, Origin, Label and Trusted
fields from the output as these also display data which needs to be
downloaded first and could hence be inaccurate [on the pro-side: This
mode is faster as it doesn't require a valid binary cache to operate].
The most notable difference perhaps is in the Filename field through: By
default it indicates an existing file, potentially compressed (Hint:
libapt users can use FileFd to open compressed files transparently). In
the --no-release-info mode the indicated file doesn't need to exist and
it will always refer to an uncompressed file, even if the index would be
(or is) stored compressed.

Remarks on fields only available in (default) --release-info mode:
* Trusted: Denotes with a 'yes' or 'no' if the data in this file is
  authenticated by a trustchain rooted in a trusted gpg key. You should
  be careful with untrusted data and warn the user if you use it.
* Codename, Suite, Version, Origin and Label are fields from the Release
  file, are only present if they are present in the Release file and
  contain the same data.

Remarks on other available fields:
* MetaKey, ShortDesc, Description, Site, Release: as defined
  by the configuration and described further above.
* Created-By: configuration entity responsible for this target
* Target-Of: type of the sources.list entry
* URI, Repo-URI: avoid using. Contains potentially username/password.
  Prefer 'Site', especially for display.
* Optional: Decodes the option of the same name from the configuration.
  Note that it is using 'yes' and 'no' instead of 'true' and 'false'.
* Language, Architecture, Component: as defined further above, but with
  the catch that they might be missing if they don't effect the target
  (aka: They weren't used while evaluating the MetaKey template).

Again, additional fields might be visible in certain implementations,
but you should avoid using them and instead talk to us about a portable
implementation.

# Multiple application requiring the same files

It is highly encouraged that applications talk to each other and to us
about which files they require. It is usually best to have a common
package ship the configuration needed to get the files, but specific
needs might require specific solutions. Again: talk to us.

# Acquiring files not mentioned in the Release file

You can't. This is by design as these files couldn't be verified to not
be modified in transit, corrupted by the download process or simple if
they are present at all on the server, which would require apt to probe
for them. APT did this in the past for legacy reasons, we do not intend
to go back to these dark times.

This is also why you can't request files from a different server. It
would have the additional problem that this server might not even be
accessible (e.g. proxy settings) or that local sources (file:/, cdrom:/)
start requesting online files…

In other words: We would be opening Pandora's box.