/repology-rules

Package normalization ruleset for Repology

Primary LanguagePythonGNU General Public License v3.0GPL-3.0

Repology ruleset

CI

There can be a huge discrepancy in how packages for a single project are named and versioned in different repositories, so Repology needs a flexible ruleset in order to overcome the differences, match packages, and make versions comparable.

TL;DR

You are welcome to submit pull requests with the rules you need. Here's a quick pointer of how to add specific rules:

You want to merge differently named packages into a single entry?

  • Choose a target name (prefer the least ambiguous and/or most widely used name)
  • Open the corresponding yaml file under 800.renames-and-merges/ (if there's no existing yaml file relevant to your package, use the file named with the first letter of your target name, like a.yaml)
  • Add a rule like - { setname: <target name>, name: <original name> }

You want to mark incorrect versions of a specific package?

  • Open the corresponding yaml file under 900.version-fixes/
  • Add a rule like: - { name: <package name>, ver: <bad version>, ignore: true }
  • Consider using a verpat with a regular expression to match similar bad versions which may appear in the future. Examples:
    • verpat: "20[0-9]{6}" to match dates (20110323)
    • verpat: "20[0-9]{2}\\.[0-9]{2}\\.[0-9]{2}" same, but for delimited date, (2010.03.23)
    • verpat: ".*20[0-9]{6}.*" to match dates anywhere in the version (1.0.20110323)
    • verpat: "[0-9a-f]{7}" match something resembling a git commit (a7b823f)
    • verpat: "[0-9]{4,}" match something resembling a build or revision number (12345)

You want to split different projects with the same name

  • Open the corresponding yaml file under 850.split-ambiguities/
  • Add a group of rules to distinguish packages by upstream URL (wwwpart, wwwpat and sourceforge conditions are allowed). Such group must end with a catch-all rule for packages not matched by specific rules. Example: 
    - { name: <ambiguous name>, wwwpart: <url part for project A>, setname: <name A> }`
- { name: <ambiguous name>, wwwpart: <url part for project B>, setname: <name B> }`
- { name: <ambiguous name>, addflag: unclassified }

Contributing

Things to know if you're submitting a pull request or have push access to this repository.

  • Repology is currently set up to automatically pull the latest ruleset from the master branch in this repository on each update, so everything committed here will be automatically applied to Repology in several hours.
  • Repology runs make check after updating the repository, and if it fails, rolls back to the latest good commit, so it's somewhat protected from a broken ruleset.
  • In the worst case, a broken ruleset will prevent Repology from updating until the problem is resolved.
  • Still, please run make check before committing, and/or install the git hook in scripts/pre-push, which runs it for you (you can copy it into .git/hooks or just run make install-hook).
  • The checker script requires the Python modules voluptuous and PyYAML. pip install PyYAML voluptuous should install them for you.
  • In general, stay close to the style already used in the ruleset, use existing rules as examples, keep it simple and have fun!
  • If in doubt, you can always just submit a report from the package's page on the website and avoid all the work!

Rule basics

Rules are stored in a set of files in YAML format, a flexible human-friendly markup format for structured data. Each rule is a single item of a big array, and may be written in a single or multiple lines (depending on what's more convenient for the particular case). For example, the following rule renames etracer into extreme-tuxracer:

- { name: etracer, setname: extreme-tuxracer }

which is the same as:

- name: etracer
  setname: extreme-tuxracer

Each rule has a set of keywords which specify how a package is matched (by name, version, repository, category etc.) and how it is modified (package is renamed, version scheme is changed, flags are applied, etc.).

Rule order matters, as multiple rules may match a single package, and they are applied in order. Furthermore, changes applied by earlier rules affect further matches: for instance, if a package is renamed, the new name will be matched for the following rules.

While rules are basically arbitrary, it's practical though to attribute each rule to a specific class of action, the most distinctive of which are:

  • Rename or merge rules. Match a name, and set another name. The main purpose is to merge differently-named packages into the same project. For example, etracer, extremetuxracer, extreme-tuxracer → extreme-tuxracer.
  • Split rules. Match a name and some additional property (version, homepage or repository), and set another name. Used to split similarly-named packages of different projects. For example, clementine → clementine-wm, clementine-player.
  • Version fixes. Match a name but do not change it; instead, change versions or set some version-related flags. Used to fix incorrect versioning scheme (v1.0 → 1.0), mark some versions as devel (such as beta versions), or ignore some versions (e.g. snapshots like 20130523 when there are official versions like 1.0).

Ruleset structure

The ruleset is split into several distinctive parts, mostly based on the functional class of rules described above. They are arranged in such a way that when adding a rule into a specific part you don't need to be aware of the rest of the ruleset.

  • 100.prefix-suffix - normalization of repository specific prefixes and suffixes which are not part of the meaningful package name. Such as removal of lib32- prefixes.

  • 2xx.handpicked - a block where access to unmodified package names is needed, such as manual whitelists or blacklists.

  • [45]xx.wildcard - wildcard rules which affect a lot of packages. These mostly handle modules for specific languages such as Perl (which may be named like p5-Foo-Bar or libfoo-bar-perl in different repositories) by adding distinctive prefix (perl: in this case) to them, so they do not conflict with modules for other languages and other software.

    There are three subsets here:

    • pure rules which are known to not have any false positives (e.g. packages from CPAN are always perl modules).
    • exceptions for the wildcard rules
    • wildcard rules themselves

  • 750.exceptions - the small set of remaining exceptions. If a package needs a rule here, it's most positively incorrectly named.

  • 800.renames-and-merges - pure merge rules

  • 850.split-ambiguities - pure split rules

  • 900.version-fixes - pure version fixes

  • 950.split-branches - additional split section for projects which have multiple development branches which are incompatible and may present in a single repository at the same time for compatibility purposes. For example, gtk2 and gtk3.

  • There are also some fixme subsets which are remainings of the previous generation of the ruleset. These files will eventually be refactored and removed.

This may seem complex, but in practice the mostly used rulesets are 800, 850 and 900, which cleanly correspond to three functional classes of rules described in the previous section.

Other parts of the ruleset may need attention when new repositories are introduced.

Rule syntax

As already mentioned, the keywords that comprise rules are related to either matching packages, or modifying them. Below are detailed descriptions for all of them.

Conditions

ruleset

Each repository that Repology supports has a set of rulesets associated with it. For instance, all Debian-based distros have the ruleset debuntu. This may be used to only match packages in specific repositories, but without the need to chase a specific repository version. You may look up repositories and their details in the repos.d directory of the main Repology repository.

You may specify a list of rulesets to match any of them.

- { ruleset: freebsd, ... }

- { ruleset: [ arch, openbsd ], ... }

noruleset

Disable rule matching for specified ruleset(s).

# applies to all Debian derivatives, but not Deepin
- { ruleset: debuntu, noruleset: deepin, ... }

family

Deprecated. Same as ruleset, and may be just changed into it.

category

Matches package category(ies). Note that category information is not available for all repositories, and each repository may have its own set of categories.

- { category: games, ... }

- { category: [ mail-client, mail-filter, mail-mta ], ... }

categorypat

Matches package category(ies) against a regular expression. The whole category is matched, match is case insensitive.

- { categorypat: "emacs[0-9]+Packages" }

maintainer

Matches package maintainer(s). The matching is case-insensitive.

- { maintainer: "nobody@nowhere.com" }

name

Match exact package name(s).

- { name: firefox, ... }

- { name: [postgresql-client, postgresql-server, postgresql-contrib], ... }

namepat

Matches package name against a regular expression. The whole name is matched. May contain captures.

- { namepat: "swig[0-9]+", ... }

ver

Matches exact package version(s).

- { name: firefox, ver: "50.0.1", ... }

notver

The opposite of ver: matches if the package version is none of specified version(s).

- { name: firefox, notver: ["50.0.1", "50.0.2"] }

verpat

Matches a package version name against a regular expression. The whole version is matched. Note that you need to escape periods, which mean "any symbol" in regular expressions. Matching is case-insensitive.

- { name: firefox, verpat: "50\\.[0-9]+", ... }

- { name: firefox, verpat: "50\\..*", ... }

vercomps

Matches the number of components (dot-separated parts) of a version.

- { name: gimp, vercomps: 3, ...} # matches 1.2.3, but not 1.2 or 1.2.3.4

verlonger

Matches versions longer than a given number of components (dot-separated parts).

Mostly useful to match broken version schemes that add extra version components.

- { name: gimp, verlonger: 3, ...} # 2.9.8.12345 is something unofficial

vergt, verge, verlt, verle, vereq, verne

Compares version to a given one and matches if it is:

  • vergt: greater (>)
  • verge: greater or equal (≥)
  • verlt: lesser (<)
  • verle: lesser or equal (≤)
  • vereq: equal
  • verne: not equal
# match git >= 2.16
- { name: git, verge: "2.16", ...}

Be careful when using this with regard to pre-release versions: 1.0beta1 is lesser than 1.0, so it won't match verge: 1.0. You may use verpat instead.

relgt, relge, rellt, relle, releq, relne

Similar to the verXX family, but checks how a package version relates to a specified release. A release includes all pre-releases and post-releases with a given prefix; e.g. releq: "1.0" would match 1.0alpha1, 1.0, 1.0patch, 1.0.1, but not 0.99 and 1.1.

wwwpat

Matches the package homepage against a regular expression. Note that unlike namepat and verpat, a partial match is allowed here. Also note that dots should be escaped with double slash, as . means "any character" in regular expressions.

- { name: firefox, wwwpat: "mozilla\\.org", ... }

wwwpart

Matches when a package homepage contains given substring. This is usually more practical than wwwpat as in most cases you just need to match an URL part and don't need complex patterns, and you don't need to worry about escaping here. Matching is case-insensitive.

- { name: firefox, wwwpart: "mozilla.org", ... }

sourceforge

Matches when a package homepage is a sourceforge page for a given project name (https://<project>.sourceforge.net, https://sourceforge.net/project/<project> etc.):

- { name: aterm, sourceforge: aterm, ... }

summpart

Matches when a package summary contains a given substring. Useful as an alternative to wwwpart for cases where the package homepage is not available. Matching is case-insensitive.

- { name: firefox, summpart: "browser", ... }

is_p_is_patch

Matches when a package has the p_is_patch flag set (see the p_is_patch action below).

Actions

setname

Effectively rename the package. You may use the $0 placeholder to substitute original name, or $1, $2 etc. to substitute the contents of the corresponding captures of the regular expression used in namepat. Note that you don't need to use neither name nor namepat for $0 to work, but you must have namepat with corresponding captures to use $1 and so on.

# etracer→extreme-tuxracer
- { name: etracer, setname: extreme-tuxracer }

# aspell-dict-en→aspell-ru, aspell-dict-ru→aspell-ru etc.
- { namepat: "aspell-dict-(.*)", setname: "aspell-$1" }

# all packages in dev-perl Gentoo category are prepended `perl:`
# Locale-Msgfmt→perl:Locale-Msgfmt
- { ruleset: gentoo, category: dev-perl, setname: "perl:$0" }

setver

Changes the version of the package. As with setname, you may use the placeholders $0, $1, etc.

# remove bogus leading version component
- { verpat: "0\\.(.*)", setver: $1 }

remove

Set to true to completely remove a package. It will not appear anywhere in Repology. Set to false to undo.

# a metapackage which does not refer to any real project, we don't need it
- { name: "x11-fonts", remove: true }

devel

Set to true to mark the version of a matched package as a development or unstable version, so it does not make the latest stable version be marked as outdated. Set to false to undo.

# mark versions with odd second component as devel
- { name: gnome-terminal, verpat: "[0-9]+\\.[0-9]*[13579]\\..*", devel: true }

altver

A project may use two parallel versioning schemes, one of which contains additional version components, such as a build number:

0.17, 0.17.13509, 0.17.13541, 0.18, 0.18.16131

Normally, 0.18.16131 would be considered more recent than 0.18, but if these refer to the same version, this is not desired behavior. In such case, a version scheme containing extra components (e.g. one which compares greater) may be marked as altver, which would allow both 0.18 and 0.18.16131 to be considered the latest, and both to be marked as outdated by the presence of either 0.19 or 0.19.x.

- { name: freecad, verlonger: 3, altver: true }

altscheme

Similar to altver, but for the case where versioning schemes do not have a common prefix and are totally incompatible:

3.2.1, 3207, 3.2.2, 3211

Marking either of the schemes with this flag results in completely independent processing, which would allow both 3.2.2 and 3211 to be treated as the newest version.

- { name: sublime-text, verpat: "[0-9]+", altscheme: true }

ignore, incorrect, untrusted, noscheme, snapshot, successor, debianism, rolling

Set to true to ignore specific package versions. This is meant for the cases where comparison is not possible - ignored versions are excluded from comparison and do not affect the status of other versions. There are multiple ignore flavors:

  • rolling - the package is always fetched from the latest snapshot or VCS master/trunk. Its version has no meaning (like Gentoo's 9999), and may contain repository-specific formats such as a commit hash, revision or date.
  • noscheme - there's no official versioning scheme. Repositories may use random versions or dates, there's no point comparing them.
  • incorrect - known incorrect version (e.g. version which was not released yet)
  • untrusted - used for repositories which are known for providing incorrect versions, to ignore them proactively. It's a common pattern to create a pair of incorrect rules matching specific versions, and an untrusted rule for the following versions in a given repository.
  • ignored - general ignore action
  • successor - currently an alias for devel, used to convey the additional meaning of this being a fork of an unmaintained original project
  • debianism - currently an alias for devel, used to convey the additional meaning of this package using a distribution maintained at Debian (probably with version addendum)
  • snapshot - currently alias for ignored
# Fedora was known to use "6.0.0" version before it was actually released
# mark as incorrect and prevent future problems
- { name: llvm, ver: "6.0.0", ruleset: fedora, incorrect: true }
- { name: llvm, ruleset: fedora, untrusted: true }

p_is_patch

Set to true to indicate that this project uses p letter in the version to indicate post- or patch releases. This fixes version comparison, as by default p is treated as pre-release.

# sudo 1.8.21p2 > 1.8.21
- { name: sudo, p_is_patch: true }

any_is_patch

Set to true to indicate that this project uses any letter in the version to indicate post- releases.

# rb here denotes a patchset, treat is as such
- { name: webalizer, verpat: ".*rb.*", any_is_patch: true }

sink

Set to true to force the package version to compare lower than any other package version. Useful to handle upstream versioning schema change when new versions compare lower than legacy ones. Set to false to undo.

# when 0.20 follows 0.193:
- { version: "0.193", sink: true }

Result: 0.20 (newest) > 0.193 (outdated)

outdated

Set to true to force the package to be outdated, even if it classifies as the most recent. Note that this does not lead to another version being selected as newest. Useful to convey that a version is outdated even when there are no newer versions (for instance, when a project is superceded by another project). Set to false to undo.

# when 0.20 follows 0.193:
- { version: "0.193", outdated: true }

Result: 0.193 (outdated) > 0.20 (outdated)

legacy

Set to true to force the package to be legacy instead of outdated. Set to false to undo. Useful when a specific repository purposely contains an outdated version of a specific project for compatibility purposes.

- { name: ruby-slack-notifier-1, ruleset: aur, legacy: true }

nolegacy

Set to true to prevent the package from ever having legacy status. This is useful for marking packages which declare to be of development version, but are nevertheless outdated.

- { name: ffmpeg-git, nolegacy: true }

warning

Output a given warning when matched.

# will catch unexpected versions
- { name: gtk, verpat: "1\\..*", setname: gtk1 }
- { name: gtk, verpat: "2\\..*", setname: gtk2 }
- { name: gtk, verpat: "3\\..*", setname: gtk3 }
- { name: gtk, verpat: "4\\..*", setname: gtk4 }
- { name: gtk, warning: "Neither of gtk1,2,3,4 - need a new rule or some weirdness is going on" }

# will trigger a warning if new project called "tesseract" appears
# ...or website changes, or just a package without website defined appears,
# so it'll require another condition
- { name: tesseract, setname: tesseract-game, wwwpart: tesseract.gg }
- { name: tesseract, setname: tesseract-ocr, wwwpart: tesseract-ocr }
- { name: tesseract, warning: "Please add rule for tesseract" }

addflavor

Flavors are used to distinguish a set of packages denoting multiple versions of a project and a set of packages denoting a multiple parts or variants of a project. Consider an example:

  • foo1 1.0 and foo2 2.0 merged into foo. In this case they denote multiple versions of the same project, flavors are not needed here and foo1 will have legacy status.
  • foo-client 1.0 and foo-server 1.1 merged into foo. In this case they denote parts of the same project, which are expected to be of the same version. Flavors should be used in this case, so foo-client will have the outdated status.

Flavors are plain strings and may be arbitrary, for example client and server in the last example. You may specify a flavor explicitly, or use the true value to make the flavor be taken from the package name.

- { name: postgresql-client, setname: postgresql, addflavor: client }
- { name: postgresql-server, setname: postgresql, addflavor: server }

# This works too
- { name: [postgresql-client, postgresql-server], setname: postgresql, addflavor: true }

setflavor

Same as addflavor, but replaces flavor instead to appending to flavors list.

resetflavors

Set to true to remove all previously added flavors.

last

Set to true to stop ruleset processing right after the current rule.

Consider this a legacy feature; it should not be needed.

replaceinname

Takes a pattern and replacement strings, and applies them to the package name. Used for low-level normalization.

# slashes in package names are not allowed
- { replaceinname: { "/": "-" } }

# also useful for some repositories
- { replaceinname: { " ": "-" } }

tolowername

Converts a package name to lowercase. This is called once in the very beginning of the ruleset. The purpose of having this as a rule action is to be able to have exceptions, e.g. packages which should be distinguished solely by the case of their names.

- { tolowername: true }

setsubrepo

Changes the subrepo property of the package. As with setname, you may use the placeholders $0, $1, etc.

# split subrepo name from package name
- { namepat: "([^-]+)-(.*)", setsubrepo: $1, setname: $2 }

Conditional rules

For additional flexibility, a mechanism exists to toggle some rules based on the previous rules.

addflag

Sets a virtual flag (arbitrary string) which only exists for the duration of rule processing, and may be checked in the following rules.

- { name: python, addflag: not_python_module }

flag, noflag

Only matches if the specified flag is (or is not) set.

- { name: python, addflag: not_python_module }
...
# will add "python:" prefix to all packages in category "python",
# but not for "python" package
- { category: python, noflag: not_python_module, setname: "python:$0" }

Annotations

These annotations do not affect package processing, but are related to ruleset maintenance.

maintenance

Indicates that a rule needs manual maintenance. For example, when a development version cannot be determined from the version schema, one would need to revisit and update the version occasionally.

- { name: tor, verge: "0.3.4", devel: true, maintenance: true }

precious

Indicates that a rule should not be removed even if it doesn't match any packages. That is, a rule is likely to be useful sometime in the future.

disposable

Indicates that a rule may be removed if it doesn't match any packages.

Author

License

GPLv3 or later, see COPYING.