Did you know ... | Search Documentation: |
Pack logtalk -- logtalk-3.85.0/manuals/_sources/devtools/packs.rst.txt |
.. _library_packs:
packs
This tool provides predicates for downloading, installing, upgrading, and uninstalling third-party libraries and applications, generically known as packs. Collections of pack specifications are made available using registries. Registries can be local to a system, publicly shared, or private to a company (e.g. only available in a VPN). There is no concept of a central registry. Users decide which registries they trust and want to use and add them using their published URLs. The tool supports both pack checksums and signatures and takes several steps to sanitize registry and pack specifications. As other Logtalk developer tools, portability is a main goal. This tool can be used with any supported Prolog backend and run in both POSIX and Windows systems. Moreover, this tool can be used not only for handling Logtalk packs but also Prolog only packs, thus providing a solution for sharing portable resources between multiple systems.
A list of public Logtalk and Prolog pack registries is available at:
https://github.com/LogtalkDotOrg/pack-registries
This tool is the beta stage of development. Feedback is most welcome.
On POSIX systems (Linux, macOS, ...), the following shell commands are required:
sha256sum
(provided by GNU coreutils
)curl
(default file downloader)wget
(alternative file downloader)bsdtar
(provided by libarchive
or libarchive-tools
)gpg
(provided by gnupg2
)git
direnv
(when using virtual environments)
The tool uses bsdtar
instead of GNU tar
so that it can
uncompress .zip
archives (unzip
doesn't provide the desired
options that allows a simple and reliable solution for ignoring the
non-predictable name of the wrapper directory).
On Windows systems, the following shell commands are required:
certutil.exe
curl.exe
(default file downloader)wget.exe
(alternative file downloader)tar.exe
gpg.exe
git.exe
Set-PsEnv
(when using virtual environments)
In recent Windows 10 builds, only wget
, gpg
, git
, and
Set-PsEnv
should require installation. You can download the GnuPG
software from:
You can download Git from:
You can download Wget from:
https://eternallybored.org/misc/wget/
You can also use Chocolatey to install the commands above:
::
choco install gnupg git wget
To install `Set-PsEnv https://github.com/rajivharris/Set-PsEnv`__ from the PowerShell Gallery:
::
PS> Install-Module -Name Set-PsEnv
On macOS systems, Apple bundles both curl
and BSD tar
(under the
name tar
; you can simply create a bsdtar
alias or install a more
recent version). The required commands can be easily installed using
MacPorts:
::
$ sudo port install coreutils wget libarchive gnupg2 git direnv
Or using Homebrew:
::
$ brew install coreutils wget libarchive gnupg2 git direnv
On Linux systems, use the distribution own package manager to install any missing command. For example, in recent Ubuntu versions:
::
$ sudo apt update $ sudo apt install coreutils curl wget libarchive-tools gnupg2 git direnv
This tool API documentation is available at:
`../../docs/library_index.html#packs <../../docs/library_index.html#packs>`__
This tool can be loaded using the query:
::
| ?- logtalk_load(packs(loader))
.
To run the tool tests, use the query:
::
| ?- logtalk_load(packs(tester))
.
The tests can be run without interfering with the user packs setup.
The packs
tool loads at startup all the currently defined registry
and pack specifications (from the registries/packs storage directory;
see below). When no registry/pack setup exists, a new one is
automatically created.
The tool provides two main objects, registries
and packs
, for
handling, respectively, registries and packs. Both objects accept a
help/0 message that describes the most common queries.
The tool uses a directory specified using the logtalk_packs
library
alias when defined (in a settings file or in a backend Prolog
initialization file). When this library alias is not defined, the tool
uses the value of the LOGTALKPACKS
environment variable when
defined. Otherwise it defaults to the ~/logtalk_packs
directory. The
actual directory can be retrieved by the query:
::
| ?- packs::logtalk_packs(Directory)
.
...
This directory holds sub-directories for registries, packs, and
archives. These sub-directories are automatically created when loading
the packs
tool if they don't exist . Users shouldn't manually modify
the contents of these directories. Multiple and independent
registry/pack setups are possible using virtual environments as
explained next.
Your registries and packs setup can be saved and restored (e.g. in a
different system) by using the packs::save/1-2
and
packs::restore/1-2
predicates, as explained in the next section
about virtual environments. If necessary, before restoring, the
packs::reset/0
predicate can be called to delete any defined
registries and installed packs.
An application may require specific pack versions. These requirements may differ between applications. Different applications may also have conflicting requirements. Therefore, a virtual environment where an application requirements are fulfilled may be required to develop and/or run it. A virtual environment is essentially a registries/packs storage directory.
Defining the logtalk_packs
library alias in a settings file or
defining the LOGTALKPACKS
environment variable before starting
Logtalk allows easy creation and switching between virtual environments.
By using a per application settings file (or a per application
environment variable definition) each application can thus use its own
virtual environment. The settings.lgt
file can define the
logtalk_packs
library alias using code such as:
::
:- initialization((
logtalk_load_context(directory, Directory),
assertz(logtalk_library_path(logtalk_packs, Directory))
))
.
The definition of the logtalk_packs
library alias must always be
an atom and thus never use library notation (i.e. it must never depend
on other library aliases).
When a virtual environment also requires a specific Logtalk version
(e.g. the version used to test and certify it), this can be installed as
a pack from the official
`talkshow https://github.com/LogtalkDotOrg/talkshow`__ registry and
used by (re)defining the LOGTALKHOME
and LOGTALKUSER
environment
variables to point to its pack directory (which can be queried by using
the packs::directory/2
message).
Experimental lgtenv.sh
and lgtenv.ps1
scripts are included to
simplify creating virtual environments. For example:
::
$ lgtenv -d ~/my_venv -c -p logtalk_packs $ cd ~/my_venv direnv: loading ~/my_venv/.envrc direnv: export +LOGTALKPACKS
Type lgtenv -h
for details on the script options.
These scripts require, respectively,
`direnv https://github.com/direnv/direnv` and
`Set-PsEnv https://github.com/rajivharris/Set-PsEnv` to be
installed. These utilities load and unload environment variables when
changing the current directory. On Windows systems, when using the
lgtenv.ps1
script, you also need to redefine the PowerShell prompt
in a profile file (e.g. $HOME\Documents\PowerShell\Profile.ps1
) to
mimic the functionality of direnv
of automatically loading an
existing .env
file when changing to its directory. For example:
::
function prompt { Set-PsEnv 'PS ' + $(Get-Location) + '> ' }
A virtual environment setup (i.e. the currently defined registries and
installed packs) can be saved into a file (e.g. requirements.lgt
)
using the packs::save/1
predicate:
::
| ?- packs::save('requirements.lgt')
.
...
This query saves a listing of all the installed packs and their
registries. Using the saved file, the virtual environment setup can then
be restored using the packs::restore/1-2
predicates. The file uses a
simple format with registry/2, pack/3, pinned_registry/1,
and pinned_pack/1 facts (in this order) and can be manually created
or edited if necessary. For example:
::
registry(talkshow, 'https://github.com/LogtalkDotOrg/talkshow.git')
.
pack(talkshow, logtalk, 3:45:0)
.
pack(talkshow, lflat, 2:1:0)
.
These files can be distributed with applications so that users can easily fulfill application requirements by running once the query:
::
| ?- packs::restore('requirements.lgt')
.
After, the application loader.lgt
file can then load the required
packs using their loader files:
::
:- initialization((
% load required packs
logtalk_load(foo(loader))
,
logtalk_load(bar(loader))
,
...
% load application files
...
)).
Note that restoring encrypted registries or encrypted packs requires entering the required passphrases. Although the restore/2 predicate accepts a list of options that include the gpg/1 option, this only allows specifying a single and common passphrase when interactive entering of passphrases is not convenient or possible.
A registry is a git remote repo that can be cloned, a downloadable archive, or a local directory containing a Logtalk loader file that loads source files defining the registry itself and the packs it provides. The registry name is ideally a valid unquoted atom. The registry directory must contain at least two Logtalk source files:
_registry
suffix, implementing the registry_protocol
. This
naming convention helps preventing name conflicts.loader.lgt
or loader.logtalk
) that loads
the registry object file and all pack object files.
An example of a registry specification object would be:
::
:- object(jdoe_awesome_packs_registry,
implements(registry_protocol))
.
:- info([
version is 1:0:0,
author is 'John Doe',
date is 2021-10-18,
comment is 'John Doe awesome packs registry spec.'
])
.
name(jdoe_awesome_packs)
.
description('John Doe awesome packs')
.
home('https://example.com/jdoe_awesome_packs')
.
clone('https://github.com/jdoe/jdoe_awesome_packs.git')
.
archive('https://github.com/jdoe/jdoe_awesome_packs/archive/main.zip')
.
:- end_object.
Optionally, the registry object can also define a note(Action, Note)
predicate. The Action
argument is an atom: add
, update
, or
delete
. The Note
argument is also an atom. The tool will print
any available notes when executing one of the registry actions. See the
registry_protocol
documentation for more details.
The registry directory should also contain LICENSE
and README.md
files (individual packs can use a different license, however). The path
to the README.md
file is printed when the registry is added. It can
also be queried using the registries::directory/2
predicate. The
NOTES.md
file name can also be used in alternative to the
recommended README.md
file name.
Summarizing the required directory structure using the above example (note that the registry and pack specification files are named after the objects):
::
jdoe_awesome_packs LICENSE README.md jdoe_awesome_packs_registry.lgt loader.lgt foo_pack.lgt bar_pack.lgt ...
With the contents of the loader.lgt
file being:
::
:- initialization((
logtalk_load(jdoe_awesome_packs_registry),
logtalk_load(foo_pack),
logtalk_load(bar_pack),
...
))
.
It would be of course possible to have all objects in a single source file. But having a file per object and a loader file helps maintenance and it's also a tool requirement for applying safety procedures to the source file contents and thus successfully loding the registry and pack specs.
As registries are git repos in the most common case and thus adding them performs a git repo cloning, they should only contain the strictly required files.
Registries can be added using the registries::add/1-3
predicates,
which take a registry URL. Using the example above:
::
| ?- registries::add('https://github.com/jdoe/jdoe_awesome_packs.git')
.
HTTPS URLs must end with either a .git
extension or a an archive
extension (same valid extensions as for pack archives, including gpg
encrypted). Git cloning URLs are preferred as they simplify updating
registries. But a registry can also be made available via a local
directory (using a file://
URL) or a downloadable archive (using a
https://
URL).
For registries made available using an archive, the
registries::add/2-3
predicates must be used as the registry name
cannot in general be inferred from the URL basename or from the archived
directory name. The registry argument must also be the declared registry
name in the registry specification object. For example:
::
| ?- registries::add(
jdoe_awesome_packs,
'https://github.com/jdoe/jdoe_awesome_packs/archive/main.zip'
)
.
When a registry may be already defined, you can use the update(true)
option to ensure that the registry will be updated to its latest
definition:
::
| ?- registries::add(
jdoe_awesome_packs,
'https://github.com/jdoe/jdoe_awesome_packs/archive/main.zip',
[update(true)]
)
.
The added registries can be listed using the registries::list/0
predicate:
::
| ?- registries::list.
% Defined registries: % jdoe_awesome_packs (git) % ...
The registries::describe/1
predicate can be used to print the
details of a registry:
::
| ?- registries::describe(jdoe_awesome_packs)
.
% Registry: jdoe_awesome_packs % Description: John Doe awesome packs % Home: https://example.com/jdoe_awesome_packs % Cloning URL: https://github.com/jdoe/jdoe_awesome_packs.git % Archive URL: https://github.com/jdoe/jdoe_awesome_packs/archive/main.zip
To update all registries, use the registries::update/0
predicate. To
update a single registry, use the registries::update/1-2
predicates.
After updating, you can use the packs::outdated/0-1
predicates to
list any outdated packs.
Registries can also be deleted using the registries::delete/1-2
predicate. By default, any registries with installed packs cannot be
deleted. If you force deletion (by using the force(true)
option),
you can use the packs::orphaned/0
predicate to list any orphaned
packs that are installed.
See the tool API documentation on the `registries <../../docs/registries_0.html>`__ object for other useful predicates.
To simplify registry development and testing, use a local directory and
a file://
URL when calling the registries::add/1
predicate. For
example:
::
| ?- registries::add('file:///home/jdoe/work/my_pack_collection')
.
If the directory is a git repo, the tool will clone it when adding it. Otherwise, the files in the directory are copied to the registry definition directory. This allows the registry to be added and deleted without consequences for the original registry source files.
To check your registry specifications, use the registries::lint/0-1
predicates after adding the registry.
A pack is specified using a Logtalk source file defining an object that
implements the pack_protocol
. The source file should be named after
the pack with a _pack
suffix. This naming convention helps
preventing name conflicts, notably with the pack own objects. The file
must be available from a declared pack registry (by having the registry
loader file loading it). The pack name is preferably a valid unquoted
atom. An example of a pack specification object would be:
::
:- object(lflat_pack,
implements(pack_protocol))
.
:- info([
version is 1:0:0,
author is 'Paulo Moura',
date is 2021-10-18,
comment is 'L-FLAT - Logtalk Formal Language and Automata Toolkit pack spec.'
])
.
name(lflat)
.
description('L-FLAT - Logtalk Formal Language and Automata Toolkit')
.
license('MIT')
.
home('https://github.com/l-flat/lflat')
.
version(
2:1:0,
stable,
'https://github.com/l-flat/lflat/archive/refs/tags/v2.1.0.tar.gz',
sha256 - '9c298c2a08c4e2a1972c14720ef1498e7f116c7cd8bf7702c8d22d8ff549b6a1',
[logtalk @>= 3:42:0],
all
)
.
version(
2:0:2,
stable,
'https://github.com/l-flat/lflat/archive/refs/tags/v2.0.2.tar.gz',
sha256 - '8774b3863efc03bb6c284935885dcf34f69f115656d2496a33a446b6199f3e19',
[logtalk @>= 3:36:0],
all
)
.
:- end_object.
The license/1 argument must be an atom and should whenever possible be a license identifier as specified in the `SPDX standard https://spdx.org/licenses/`__.
Optionally, the pack object can also define a
note(Action, Version, Note)
predicate. The Action
argument is an
atom: install
, update
, or uninstall
. The Note
argument
is also an atom. The tool will print any available notes when executing
one of the registry actions. See the pack_protocol
documentation for
more details.
The pack sources must be available either as a local directory (when
using a file://
URL) or for downloading as a supported archive. The
checksum for the archive must use the SHA-256 hash algorithm
(sha256
). The pack may optionally be signed. Supported archive
formats and extensions are:
.zip
.tgz
, .tar.gz
.tbz2
, .tar.bz2
Also, for encrypted packs, all the extensions above with a .gpg
suffix (e.g. .zip.gpg
).
The pack sources should contain LICENSE
, README.md
(or
NOTES.md
), and loader.lgt
(or loader.logtalk
) files.
Ideally, it should also contain a tester.lgt
(tester.logtalk
)
file. The path to the README.md
file is printed when the pack is
installed or updated. It can also be queried using the
packs::directory/2
predicate.
Packs can be gpg
encrypted, with a choice of passphrase-based
encryption, key-based encryption, or both. Encrypted pack archives must
always have a .gpg
extension. For example, to encrypt a pack archive
with a symmetric cipher using a passphrase:
::
$ tar -cvzf - my_pack | gpg -c --cipher-algo AES256 > v1.2.1.tar.gz.gpg
In this case, the passphrase would need to be securely communicated to any users installing or updating the pack.
See the gpg
documentation for full details on encrypting and
decrypting archives. If you get a "gpg: problem with the agent:
Inappropriate ioctl for device" error message with the command above,
try:
::
$ export GPG_TTY=$(tty)
Packs can be gpg
signed. Detached signature files are assumed and
expected to share the name of the archive and use .asc
or .sig
extensions. For example, if the pack archive name is v1.0.0.tar.gz
,
the signature file must be named v1.0.0.tar.gz.asc
or
v1.0.0.tar.gz.sig
. When the checksig(true)
option is used, the
signature file is automatically downloaded using a URL constructed from
the pack archive URL. When both .asc
and .sig
files exist, the
.asc
file is used. An example of signing a pack and creating the
.asc
file (assuming the default signing key) is:
::
$ gpg --armor --detach-sign v1.0.0.tar.gz
To create instead a .sig
file:
::
$ gpg --detach-sign v1.0.0.tar.gz
See the gpg
documentation for full details on signing archives and
sharing the public keys required to verify the signatures.
Typically, pack archive download URLs are HTTPS URLs and handled using
curl
. It's also possible to use git archive
to download pack
archives, provided that the server supports it (as of this writing,
Bitbucket and GitLab public hosting services support it but not GitHub).
Using git archive
is specially useful when the packs registry in
hosted in a server using Single Sign-On (SSO) for authentication. In
this case, HTTPS URLs can only be handled by curl
by passing a token
(see below for an example). When the user have setup SSH keys to
authenticate to the packs registry server, git archive
simplifies
pack installation, providing a better user experience. For example:
::
version(
1:0:1,
stable,
'git@gitlab.com:me/foo.git/v1.0.1.zip',
sha256 - '0894c7cdb8968b6bbcf00e3673c1c16cfa98232573af30ceddda207b20a7a207',
[logtalk @>= 3:36:0],
all
)
.
The pseudo-URL must be the concatenation of the SSH repo cloning URL with the archive name. The archive name must be the concatenation of a valid tag with a supported archive extension. SSH repo cloning URLs use the format:
::
git@<hostname>:path/to/project.git
They can usually be easily copied from the hosting service repo webpage. To compute the checksum, you must first download the archive. For example:
::
$ git archive --output=foo-v1.0.1.zip --remote=git@gitlab.com:me/foo.git v1.0.1 $ openssl sha256 foo-v1.0.1.zip
Be sure to use a format that is supported by both the packs
tool and
the git archive
command (the format is inferred from the
--output
option). Do not download the archive from the web interface
of the git hosting service in order to compute the checksum. Different
implementations of the archiving and compressing algorithms may be used
resulting in mismatched checksums.
Users installing packs available using git archive
URLs are advised
to run a SSH agent to avoiding being prompted for passwords when
installing or updating packs. They must also upload their SSH public
keys to the pack provider hosts.
A pack may specify multiple versions. Each version is described using a
version/6 predicate clause as illustrated in the example above. The
versions must be listed ordered from newest to oldest. For details, see
the pack_protocol
API documentation.
Listing multiple versions allows the pack specification to be updated (by updating its registry) without forcing existing users into installing (or updating to) the latest version of the pack.
Pack dependencies on other packs can be specified using a list of
Registry::Pack Operator Version
terms where Operator
is a
standard term comparison operator:
Registry::Pack @>= Version
- the pack requires a dependency with
version equal or above the specified one. For example,
logtalk @>= 3:36:0
means that the pack requires Logtalk 3.36.0 or
later version.Registry::Pack @=< Version
- the pack requires a dependency with
version up to the specified one. For example, common::bits @=< 2:1
means that the pack requires a common::bits
pack up to 2.1. This
includes all previous versions and also all patches for version 2.1
(e.g. 2.1.7, 2.1.8, ...) but not version 2.2 or newer.Registry::Pack @< Version
- the pack requires a dependency with
version older than the specified one. For example,
common::bits @< 3
means that the pack requires a common::bits
2.x or older version.Registry::Pack @> Version
- the pack requires a dependency with
version newer than the specified one. For example,
common::bits @> 2:4
means that the pack requires a
common::bits
2.5 or newer version.Registry::Pack == Version
- the pack requires a dependency with a
specific version. For example, common::bits == 2:1
means that the
pack requires a common::bits
pack version 2.1.x (thus, from
version 2.1.0 to the latest patch for version 2.1).Registry::Pack \== Version
- the pack requires a dependency with
any version other than then the one specified. For example,
common::bits \== 2.1
means that the pack requires a
common::bits
pack version other than any 2.1.x version.
To specify range dependencies by using two consecutive elements with
the lower bound followed by the upper bound. For example,
common::bits @>= 2, common::bits @< 3
means all common::bits
2.x
versions but not older or newer major versions.
It's also possible to specify alternative dependencies using the
(;)/2
operator. For example,
(common::bits == 1:9; common::bits @>= 2:3)
means either
common::bits
1.9.x versions or 2.3.x and later versions.
Alternatives should be listed in decreasing order of preference.
When a pack also depends on a Logtalk or backend version, the name
logtalk
or the backend identifier atom can be used in place of
Registry::Pack
(see below for the table of backend specifiers). For
example, logtalk @>= 3.36.0
.
When a pack also depends on an operating-system version (e.g. a pack
containing shared libraries with executable code), the
os(Name,Machine)
compound term can also be used in place of
Registry::Pack
. For example, os('Darwin',x86_64) @>= '23.0.0'
.
Note that, in this case, the release is an atom. The operating-system
data (name, machine, and release) is queried using the corresponding
os
library predicates (see the library documentation for details).
Ideally, packs are fully portable and can be used with all Logtalk
supported Prolog backends. This can be declared by using the atom
all
in the last argument of the version/6 predicate (see example
above).
When a pack can only be used with a subset of the Prolog backends, the last argument of the version/6 predicate is a list of backend identifiers (atoms):
b
ciao
cx
eclipse
gnu
ji
xvm
quintus
sicstus
swi
tau
trealla
xsb
yap
To simplify pack development and testing, define a local registry and add to it a pack specification with the development version available from a local directory. For example:
::
version(
0:11:0,
beta,
'file:///home/jdoe/work/my_awesome_library',
none,
[],
all
)
.
If the directory is a git repo, the tool will clone it when installing the pack. Otherwise, the files in the directory are copied to the pack installation directory. This allows the pack to be installed, updated, and uninstalled without consequences for the pack source files.
You can also use a local archive instad of a directory. For example:
::
version(
1:0:0,
stable,
'file:///home/jdoe/work/my_awesome_library/v1.0.0.tar.gz',
sha256 - '1944773afba1908cc6194297ff6b5ac649a844ef69a69b2bcdf267cfa8bfce1e',
[],
all
)
.
Packs that are expected to be fully portable should always be checked by
loading them with the portability
flag set to warning
.
To check your packs specifications, use the packs::lint/0-2
predicates after adding the registry that provides the packs.
Packs must be available from a defined registry. To list all packs that
are available for installation, use the packs::available/0
predicate:
::
| ?- packs::available.
To list all installed packs, call the packs::installed/0
predicate:
::
| ?- packs::installed.
To list only the installed packs from a specific registry, call instead
the packs::installed/1
predicate. For example:
::
| ?- packs::installed(talkshow)
.
To know more about a specific pack, use the packs::describe/1-2
predicates. For example:
::
| ?- packs::describe(bar)
.
The packs::describe/2
predicate can be used when two or more
registries provide packs with the same name. For example:
::
| ?- packs::describe(reg, bar)
.
To install the latest version of a pack, we can use the
packs::install/1-4
predicates. In the most simple case, when a pack
name is unique among registries, we can use the packs::install/1
predicate. For example:
::
| ?- packs::install(bar)
.
Any pack dependencies are also checked and installed or updated if necessary. Other install predicates are available to disambiguate between registries and to install a specific pack version.
Packs becomes available for loading immediately after successful
installation (no restarting of the Logtalk session is required). For
example, after the pack bar
is installed, you can load it at the
top-level by typing:
::
| ?- {bar(loader)
}.
or load it from a loader file using the goal
logtalk_load(bar(loader))
.
After updating the defined registries, outdated packs can be listed
using the packs::outdated/0
predicate. You can update all outdated
packs by calling the packs::update/0
predicate or update a single
pack using the packs::update/1-2
predicates. For example:
::
| ?- packs::update(bar)
.
By default, updating a pack fails if it would break any dependent pack
(the force(true)
option, described below, can be used to force
updating in this case).
The tool provides versions of the pack install, update, and uninstall predicates that accept a list of options:
verbose(Boolean)
(default is false
)clean(Boolean)
(default is false
)update(Boolean)
(default is false
)force(Boolean)
(default is false
)compatible(Boolean)
(default is true
)checksum(Boolean)
(default is true
)checksig(Boolean)
(default is false
)git(Atom)
(extra command-line options; default is ''
)downloader(Atom)
(downloader utility; default is curl
)curl(Atom)
(extra command-line options; default is ''
)wget(Atom)
(extra command-line options; default is ''
)gpg(Atom)
(extra command-line options; default is ''
)tar(Atom)
(extra command-line options; default is ''
)
Note that, by default, only compatible packs can be installed. To
install a pack that is incompatible with the current Logtalk version,
backend version, or operating-system version, use the install/4 or
update/3 predicates with the option compatible(false)
.
When installing large packs over unreliable network conditions, you may
try switching the default downloader utility from curl
to wget
.
When a pack may be already installed, you can use the update(true)
option to ensure that the installation will by updated to the specified
version:
::
| ?- packs::install(reg, bar, 1:1:2, [update(true)])
.
When using a checksig(true)
option to check a pack signature, is
strongly advised that you also use the verbose(true)
option. For
example:
::
| ?- packs::install(reg, bar, 1:1:2, [verbose(true), checksig(true)])
.
Note that the public key used to sign the pack archive must be already present in your local system.
Downloading pack archives may require passing extra command-line options
to curl
for authentication. A common solution is to use a personal
access token. The details depend on the server software. An example when
using GitHub:
::
| ?- packs::install(reg, bar, 1:1:2, [curl('--header "Authorization: token foo42"')])
.
Another example when using GitLab:
::
| ?- packs::install(reg, bar, 1:1:2, [curl('--header "PRIVATE-TOKEN: foo42"')])
.
Pack archives may be gpg
encrypted. Encryption can be
passphrase-based, key-based, or both. When using only passphrase-based
encryption, the archive passphrase must be entered (if not cached) when
installing or updating a pack. In this case, the passphrase can be
entered interactively or using the gpg/1 option. For example:
::
| ?- packs::install(reg, bar, 1:1:2, [gpg('--batch --passphrase test123')])
.
See the gpg
documentation for details. When using the gpg/1
option, you should be careful to not leak passphrases in e.g. the query
history.
To uninstall a pack that you no longer need, use the
packs::uninstall/1-2
predicates. By default, only packs with no
dependent packs can be uninstalled. You can print or get a list of the
packs that depend on a given pack by using the packs::dependents/1-3
predicates. For example:
::
| ?- packs::dependents(reg, bar, Dependents)
.
See the tool API documentation on the `packs <../../docs/packs_0.html>`__ object for other useful predicates.
The path to the pack README.md
file is printed when the pack is
installed or updated. It can also be retrieved at any time by using the
readme/2 predicate. For example:
::
| ?- packs::readme(lflat, Path)
.
Additional documentation may also be available from the pack home page,
which can be printed by using the describe/1-2
predicates. For
example:
::
| ?- packs::describe(lflat)
.
% Registry: ... % Pack: lflat % Description: L-FLAT - Logtalk Formal Language and Automata Toolkit % License: MIT % Home: https://github.com/l-flat/lflat % Versions: ...
The pack API documentation can be generated using the lgtdoc
tool
library and directory predicates (depending on the pack source files
organization). For example:
::
| ?- {lflat(loader)
},
{lgtdoc(loader)
},
logtalk::expand_library_path(lflat, Path)
,
lgtdoc::rdirectory(Path)
.
...
This query creates a xml_docs
directory in the current directory.
The XML documentation files can then be converted into a final format,
e.g. HTML, using one of the lgtdoc
tool provided scripts. For
example:
::
$ cd xml_docs $ lgt2html
For more details and alternatives, see the lgtdoc
tool
documentation.
It is also possible to add API documentation and diagrams for all the
installed packs to the Logtalk distribution API documentation and
diagrams by calling the update_html_docs
and update_svg_diagrams
scripts with the -i
option. See the scripts documentation for more
details.
Registries and packs can be pinned after installation to prevent
accidental updating or deleting, e.g. when using the batch update/0
predicate. This is useful when your application requires a specific
version or for security considerations (see below). For example, if we
want the bar
pack to stay at its current installed version:
::
| ?- packs::pin(bar)
.
yes
After, any attempt to update or uninstall the pack will fail with an error message:
::
| ?- packs::update(bar)
.
! Cannot update pinned pack: bar
no
| ?- packs::uninstall(bar)
.
! Cannot uninstall pinned pack: bar
no
To enable the pack to be updated ou uninstalled, the pack must first be
unpinned. Alternatively, the force(true)
option can be used. Note
that if you force update a pinned pack, the new version will be
unpinned.
It's also possible to pin (or unpin) all defined registries or installed packs at once by using the pin/0 (or unpin/0) predicates. But note that registries added after or packs installed after will not be automatically pinned.
Logtalk packs (as most Logtalk libraries, tools, and examples) are
expected to have a tester.lgt
or tester.logtalk
tests driver
file at the root of their directory, which can be used for both
automated and manual testing. For example, after installing the foo
pack:
::
| ?- {foo(tester)
}.
To test all installed packs, you can use the logtalk_tester
automation script from the installed packs directory, which you can
query using the goal:
::
| ?- packs::prefix(Directory)
.
Note that running the packs tests, like simply loading the pack, can result in calling arbitrary code, which can potentially harm your system. Always take into account the security considerations discussed below.
New pack registries should be examined before being added, specially if public and from a previously unknown source. The same precautions should be taken when adding or updating a pack. Note that a registry can always index third-party packs.
Pack checksums are checked by default. But pack signatures are only checked if requested as packs are often unsigned. Care should be taken when adding public keys for pack signers to your local system.
Registry and pack spec files plus the registry loader file are compiled
by term-expanding them so that only expected terms are actually loaded
and only expected logtalk_load/2 goals with expected relative file
paths are allowed. Predicates defining URLs are discarded if the URLs
are neither https://
nor file://
URLs or if they contain
non-allowed characters (currently, only alpha-numeric ASCII characters
plus the ASCII /
, .
, -
, and _
characters are accepted).
But note that this tool makes no attempt to audit pack source files
themselves.
Registries and packs can always be pinned so that they are not accidentally updated to a version that you may not had the chance to audit.
_registry
or _pack
suffix. Save the objects in files
named after the objects.$LOGTALKPACKS
directory (or the default
~/logtalk_packs
directory) on your regular backups.
This tool can also be used to install Prolog packs that don't use
Logtalk. After installing a pl_pack
Prolog pack from a pl_reg
registry, it can be found in the $LOGTALKPACKS/packs/pl_reg/pl_pack
directory. When the LOGTALKPACKS
environment variable is not
defined, the pack directory is by default
~/logtalk_packs/packs/pl_reg/pl_pack
.
Different Prolog systems provide different solutions for locating Prolog
code. For example, several Prolog systems adopted the Quintus Prolog
file_search_path/2 hook predicate. For these systems, a solution
could be to add a fact to this predicate for each installed Prolog pack.
For example, assuming a pl_pack
Prolog pack:
::
:- multifile(file_search_path/2)
.
:- dynamic(file_search_path/2)
.
file_search_path(library, '$LOGTALKPACKS/packs/pl_pack')
.
If the Prolog system also supports reading an initialization file at startup, the above definition could be added there.
Load the tutor
tool to get help with selected warnings printed by
the packs
tool.
Using the verbose(true)
option on Windows systems may not provide
the shell commands output depending on the backend.
On Windows systems, the reset, delete, and uninstall predicates may fail to delete all affected folders and files due to a operating-system bug. Depending on the backend, this bug may cause some of the tests to fail. For details on this bug, see:
https://github.com/microsoft/terminal/issues/309
The workaround is to use the Windows File Explorer to delete the left-over folders and files.
When using Ciao Prolog 1.20.0, a workaround is used for this system non-standard support for multifile predicates.
When using GNU Prolog 1.5.0 as the backend on Windows, you may get an error on directory_files/2 calls. For details and a workaround, see:
https://github.com/didoudiaz/gprolog/issues/4
This issue is fixed in the latest GNU Prolog git version.
Using SICStus Prolog as the backend on Windows doesn't currently work in version 4.7.0 and earlier versions. The underlying issues are fixed in the SICStus Prolog 4.7.1 version.
XSB have an odd bug (likely in its parser) when reading files that may
cause a pack installed version to be reported as the end_of_file
atom.
Some tests fail on Windows when using ECLiPSe or XSB due to file path representation issues.