GLEP 67: Package maintenance structure
|Author||Michał Górny <firstname.lastname@example.org>|
|Posting history||2015-11-03, 2016-05-29|
- Backwards Compatibility
- Reference implementation
Within this GLEP, the issues with the current package maintainer description system are explained, and a new system that aims to solve those issues is proposed. The current complex structure is replaced with two well-defined maintainer types — people and projects. Herds are removed in favor of projects, and project member listings are provided in XML format. Maintainer listings in metadata.xml become uniform, and can be used directly to assign bugs.
The current system used to declare maintainers in Gentoo has been criticized multiple times. A number of mailing list threads raised various issues with it and proposed alternatives. The topic has been brought before the Council, and was discussed on 2015-10-25 Council meeting (continuation from 2015-10-11).
The Council meeting has brought two important decisions:
- herds were to be deprecated and eventually removed, both in definition and in machine uses.
- A replacement structure can not be voted ad-hoc, and therefore a new GLEP was to be written providing a complete proposal.
This GLEP is a response to the second decision, aiming to provide a new maintainership structure written with the goals of simplicity, flexibility and good backwards compatibility.
The current system uses two elements in metadata.xml to describe maintainers:
- <maintainer/> element that is identified by e-mail address and can have optional name and description,
- <herd/> element that is identified by short herd name, and holds no extra information.
The system stated that <maintainer/> elements always have higher priority than <herd/>, except when maintainer descriptions stated otherwise.
This resulted in four different kinds of maintainers being listed:
- regular people (developers and proxied maintainers), described using <maintainer/>,
- projects with matching herds, described either using <herd/>, <maintainer/> or both,
- herds subordinate to projects or independent of projects, described using <herd/>,
- projects without matching herd and other e-mail aliases, described using <maintainer/>.
The herds stated using the <herd/> tag corresponded to herd maintainer listings in herds.xml file. In the past, this file could either list herd maintainers explicitly or copied them from a project; the latter has been broken by migrating projects to the wiki, however.
The maintainers stated using the <maintainer/> tag lack any explicit type indicator. Usually, project wiki pages or e-mail alias targets were used as a reference maintainer list.
There is a special case of email@example.com e-mail address that — when used as maintainer — indicates that the package has no active maintainer.
The issues with the current system expressed so far included:
- Redundancy and complexity in structure. Maintainers can be either grouped as herd maintainers, projects or e-mail aliases that do not correspond to either. Herds can be equivalent, subordinate or independent of projects.
- Redundancy in member listings. It is no longer possible to implicitly copy project members to herd maintainers, or the other way around. Therefore, maintainers of herds corresponding to projects have to be listed twice — in herds.xml and in project wiki pages. Both listings need to be kept in sync.
- Redundancy in maintainer listings. By using two different elements to express maintainers, herd maintenance could be expressed in up to three different ways.
- Implicit ordering rules in maintainer listings. The importance of maintainers is determined by descriptions, then elements used, then actual occurrence order. It is not fully machine-readable.
- Unnecessary indirection for bug assignment. If a package belongs to a herd, one needs to parse herds.xml to obtain the bug assignment e-mail address corresponding to the herd name.
- Unclear and complex member expansion rules. Maintainers of herd can be determined using herds.xml, unless the herd corresponds to a project which has different member listing on wiki page. Non-herd maintainers can be either people, projects (members from wiki page) or other groups where member listings can't be clearly obtained.
- Unclear cross-repository meaning. The herds.xml is not clearly defined throughout multiple repository chain, while metadata.xml is.
Each package defines zero or more maintainers. Each maintainer can either be a person or a project. Each maintainer is identified by a unique e-mail address that must correspond to an active bugs.gentoo.org account. Optionally, each maintainer can define a human-readable name and maintenance description.
Maintainers are described using metadata.xml <maintainer/> element. The type of the maintainer is defined by type attribute of the <maintainer/> element. The e-mail address, human-readable name and maintenance description are placed in <email/>, <name/> and <description/> sub-elements appropriately.
<pkgmetadata> <maintainer type="person"> <email>firstname.lastname@example.org</email> <name>Foo Barsky</name> <description>Proxied maintainer</description> </maintainer> <maintainer type="person"> <email>email@example.com</email> <name>Example Developer</name> </maintainer> <maintainer type="project"> <email>firstname.lastname@example.org</email> </maintainer> </pkgmetadata>
The basic project structure is defined in GLEP 39. However, the projects which are going to maintain packages have to meet the additional requirement of having a unique e-mail address with a corresponding bugs.gentoo.org account.
Each project can have zero or more subprojects, from which it can optionally inherit members. It is undefined whether a project can have more than one parent project. However, the complete project hierarchy must form an acyclic directed graph.
The project structure is exported from wiki.gentoo.org into a projects.xml file. The file consists of root <projects/> element which contains one or more <project/> element. Each <project/> element contains the following sub-elements:
- <email/> element stating the project contact e-mail (must be registered on bugs.gentoo.org),
- <name/> element stating the human-readable project name,
- <url/> element stating the project homepage URL,
- <description/> element shortly describing the project,
- zero or more <subproject/> elements listing subprojects of the particular project,
- zero or more <member/> elements listing direct project members.
Each <subproject/> element has the following attributes:
- obligatory ref="" attribute referencing the subproject by e-mail address (the e-mail address must be equal to the value of <email/> element of exactly one other <project/>),
- optional inherit-members="" attribute whose non-empty value indicates that subproject members are to be considered members of the parent project as well.
Each <member/> has the following sub-elements:
- <email/> stating the member's e-mail address,
- optional <name/> stating the member's human-readable name,
- optional <role/> stating the member's role in team.
In addition, <member/> can have optional is-lead="" attribute whose non-empty value indicates that the particular member is the project's lead.
<projects> <project> <email>email@example.com</email> <name>Portage package manager</name> <url>https://wiki.gentoo.org/wiki/Project:Portage</url> <description>Something about Portage</description> <member is-lead="1"> <email>firstname.lastname@example.org</email> <name>Example Developer</name> <role>Lead</role> </member> <member> <email>email@example.com</email> <name>Another Developer</name> </member> <!-- members are not inherited, purely organizational hierarchy --> <subproject ref="firstname.lastname@example.org"/> </project> <project> <email>email@example.com</email> <name>Portage-related utilities</name> <url>https://wiki.gentoo.org/wiki/Project:Tools-Portage</url> <description>Maintainers of various common Portage tools</description> <member is-lead="1"> <email>firstname.lastname@example.org</email> <name>Another Developer</name> <role>Lead</role> </member> <member> <email>email@example.com</email> <name>Example Developer</name> </member> <!-- members are inherited --> <subproject ref="firstname.lastname@example.org" inherit-members="1"/> </project> <project> <email>email@example.com</email> <name>Portage-related utility of some kind</name> <url>https://wiki.gentoo.org/wiki/Project:Some-Portage-Tool</url> <description>My random Portage tool</description> <member is-lead="1"> <email>firstname.lastname@example.org</email> <name>Me!</name> <role>Lead</role> </member> </project> </projects>
The projects.xml file is placed inside the metadata directory inside the repository, and applies to the repository and all repositories specifying it as a master (either directly or indirectly). Appropriately, when a project lookup is performed for package, the projects.xml from the repository containing the package is scanned first, and then its masters are scanned recursively.
Each project must not be specified more than once in the effective set of projects.xml files applying to a repository. In particular, it is not possible to alter or redefine an inherited project in a sub-repository. It is recommended that each repository uses a separate namespace (such as the hostname part of an e-mail address) for its projects.
The package metadata description is fully self-sufficient for bug assignment. The order in which <maintainer/> elements occur (after applying restrictions) indicates the chain of responsibility. A bug is assigned to the first maintainer, while all the remaining maintainers are CC-ed.
For packages which have no maintainers, repository-specific bug assignment rules apply. In particular, ::gentoo packages with no maintainer are assigned to email@example.com.
In order to determine the effective list of maintainers, all project-type maintainers are expanded using projects.xml. Each project is matched by e-mail address, and replaced by one or more maintainer objects. Project members form person-type maintainers, with project lead (if any) having authority over remaining project members. Subproject form project-type maintainers which are expanded recursively.
The use of <herd/> element to indicate herd maintenance has been deprecated by the Council on 2015-10-25, as an extension of deprecating the concept of herds. As an alternative, introducing a <project/> element or modifying <maintainer/> element has been proposed.
The new <project/> element has been rejected as it meant reintroducing the same structure with a different name yet the same problems. The use of <maintainer/> element to indicate all maintainers has the following advantages:
- Clean database structure. Since both person- and project-type maintainers are in fact maintainers, they should be derived from a single element rather than two disjoint elements.
- Clean ordering for bug assignment. Before, the two elements were assigned weights which considered <maintainer/> more important than <herd/> against their usual ordering. Even if new element was introduced without such implicit weight, developers would mistakenly recall the old rules and keep applying them.
- More consistent record format. In the past, some herds/projects were described using the <herd/> element, some were using the <maintainer/> element and some even both. Using a single element avoids this inconsistency.
- Backwards compatibility. Re-using an existing, well-supported element means keeping backwards compatibility with existing tools. While their functionality will be limited until they are updated for the new project structure, they at least won't become completely broken.
There was a discussion whether projects should be identified by short identifiers (alike herds) or their e-mail addresses. The e-mail addresses were selected because of the following advantages:
- Re-use of existing identifiers. Since herds were deprecated and old project pages removed, there are no longer any official short project identifiers. The identifiers used on Wiki have forced case and certainly aren't short. Introducing additional identifier just for mapping metadata seems unnecessary.
- Stand-alone meaningfulness of metadata. Using e-mail address provides a meaningful information (useful e.g. for contact or bug assignment) directly in metadata. Using another kind of identifier implies the necessity of some transformation or mapping.
- Cross-project correctness. E-mail addresses are globally unique. This means that non-Gentoo projects can have their own repositories, and declare their own projects without risk of short name collision.
- Backwards compatibility. While current tools won't recognize the project-type maintainers as de-facto projects, they will still be able to correctly recognize their e-mail addresses.
In the previous system, firstname.lastname@example.org e-mail address was used to mark packages lacking active maintainer. This solution no longer fits the new system since maintainer-needed is neither a person, nor a project.
While purely technically, a new maintainer-needed project could be created, it wouldn't really fit the conventional project structure. Furthermore, it would still carry the special rules indicating that ownership by this project actually indicates no maintainer at all.
Instead, the case of no active maintainer is expressed by not listing any maintainers which is cleaner semantically. The bug assignment to email@example.com is carried through appropriate bug assignment rules.
The project structure is defined by GLEP 39 and therefore is outside the scope of this specification. The projects.xml mapping attempts to provide an off-line copy of the project information stored on Gentoo Wiki, in a format similar to the one used for herds.xml.
The basic goal for the format was to provide means for obtaining list of effective project members.
The subproject structure aims at defining collective projects where the members of a particular project include all members of subprojects. This used to be defined as <membersof/> in herds.xml and <subproject inheritmembers=""/> attribute in old project XML files.
It was pointed out that specifying type="" of a maintainer is redundant since the maintainer type can be determined by matching the maintainer's e-mail address against projects.xml.
This information was added explicitly to improve readability and avoid unnecessary project database lookups for non-project maintainers. Furthermore, mis-sync between the project database and metadata maintainer types is unlikely since people and projects are not inter-changeable, and we can't expect the person's e-mail address to be reused for a new project, or the other way around.
It was pointed out that specifying full names in metadata.xml is redundant since each maintainer has a single name that is commonly shared across all <maintainer/> occurrences. Instead, an additional database (dictionary) could be used to map maintainer e-mail addresses to real names — or real names could be dropped entirely.
The support for optional maintainer names was preserved from the old system. Specifying names is kept fully optional, and considered a convenience/matter of respect rather than technically important information. Furthermore, names change rarely unlike e-mail addresses. In case of proxied maintainers, it is not uncommon to reference real name when looking for the new maintainer's e-mail address.
While an external database of maintainer names would allow consistently assigning real names to maintainers, it seems like an overkill. Furthermore, it is quite likely that this database would be forced to reside outside the repository which would cause more synchronization issues and the proxy-maintainer workflow harder. In particular, currently proxied maintainers can add themselves to metadata.xml in a single commit to the repository. If external database was used, the database would have to be updated in addition to the repository commit.
The GLEP preserves almost full backwards compatibility to the current metadata.xml format, with the following changes:
- <herd/> element is removed. Since it was fully optional, no tools are broken.
- <maintainer/> is used to describe both projects and people. This was already the case sometimes, with the limitation of the tools being unable to expand project members. This limitation is extended to all projects in the existing tools, and can be removed through updating tools to support projects.xml.
- <maintainer/> is given new type="" attribute. No known tools refuse metadata.xml specifications that have extraneous attributes as long as updated DTD is provided.
The projects.xml file provides a replacement for herds.xml, fitting the new structure. Since a new file is used, the change is fully compatible to existing software. The herds.xml file must be preserved for a transition period until all <herd/> occurrences are removed.
Removing herds.xml should cause only very limited breakage. The Gentoo systems using e.g. CVS checkouts were already missing the file, and therefore the tools needed to handle that case gracefully. For improved compatibility, a herds.xml file listing no herds can be distributed for additional transition period.
The new projects.xml file format provides partial compatibility with herds.xml file format, aiming for reduced workload while migrating to the new system.
The migration to the new system will require two preparatory steps:
- all existing projects must be ensured to have unique e-mail addresses. Projects sharing the same e-mail address either need to be merged, or be given unique e-mail addresses.
- All herds need to be converted into projects, subprojects or disbanded (replaced by person-type maintainers).
Afterwards, projects.xml can be generated correctly from the Wiki and can replace herds.xml.
In order to make the current metadata.xml files compliant to the new format, a two-step conversion needs to be performed:
- all <herd/> elements need to be replaced with appropriate <maintainer/> elements, and the element order need to be adjusted correctly. In particular, new <maintainer/> elements must be placed after existing <maintainer/> elements, except when maintainer descriptions request otherwise. During a transition period, <herd/> elements may still be supported.
- All <maintainer/> elements need to be given appropriate type="". This could be done via matching <maintainer/> e-mail addresses to project addresses, and assuming project whenever there is a match, person otherwise.
The reference document type definition files for XML documents specified in this GLEP are stored in data/dtd.git repository . The DTD for projects.xml is stored as projects.dtd in master branch of the repository . The updated DTD for metadata.xml is stored as metadata.dtd in the glep67 branch of the repository .
The migration is done in four steps, using separate script for each step:
- preliminary cleanup (needed because lxml does not preserve original use of single vs double quotes),
- replacement of all <herd/> elements,
- removal of remaining firstname.lastname@example.org entries (now to be implicit empty maintainer list),
- setting of type= on all <maintainer/> items.
Each <herd/> will be replaced, based on herd maintainers' decision or lack of it, with:
- a project maintainer,
- individual inline list of current herd maintainers,
- no maintainers (effectively leaving the package to the other maintainers or dropping it to maintainer-needed).
Due to high backwards compatibility, no changes in Portage are required to use the new system. However, the glep67 branch of mgorny's fork of Portage  contains improvements for GLEP 67 support. In particular, the branch adds explicit maint_type attribute to _Maintainer objects, and removes herds.xml repoman checks (which would be inactive with removed herds.xml anyway).
The metadata.xml conformance with the new system would be checked implicitly once metadata.dtd is updated. Additional type-to-projects.xml checks can be added in the future.