Title: Debian Package Maintenance Hub
DEP: 2
State: DRAFT
Date: 2012-01-13
Drivers: Raphael Hertzog <hertzog@debian.org>
URL: http://dep.debian.net/deps/dep2
Source: https://salsa.debian.org/dep-team/deps/-/blob/master/web/deps/dep2.mdwn
Abstract:
Debian maintainers rely on a multitude of services (DDPO, PTS,
DDPO-by-mail, BTS, etc), and information sources, in order to do
their job. The flow of information varies greatly from case to case.
.
This proposal is about creating a central infrastructure
that would consolidate several of those services and that would
standardize the information flow.
Rationale
This new package maintenance infrastructure is needed:
- to fix long standing problems;
- to provide a clean base to implement new features:
- that will help maintainers do a better job;
- that will help packaging teams to organize themselves;
- that will help the QA team to ensure that all Debian packages are well maintained.
Problems to solve
Maintainer vs Uploader
The flow of information is not the same depending on whether you're listed in the Maintainer field (in which case most services mail you directly) or not (in which case you're supposed to subscribe via the PTS or via a dedicated mailing list). But the opposite is true as well, some information is only available via the PTS and many maintainers have to subscribe to the PTS while excluding almost everything just to get the information they want.
See also #507288 for some more discussions on this topic.
This makes it very painful to change/switch the Maintainer field because people have to update their PTS subscriptions accordingly.
Duplication of work / inconsistency between the DDPO and the PTS
The DDPO and the PTS are completely separate services. This leads to duplication of work when a new information needs to be made available in their respective interface. It can also lead to inconsistencies between both services when bug occurs or when different choices are made.
Mailing lists as Maintainer
We often have mailing lists listed in the Maintainer field and it's not clear who are the real package maintainers and how many of them there are. The Uploaders field is often outdated, and/or is just a representation of who worked last on the package instead of who feels responsible for the package.
Goals
Provide a working replacement for DDPO/PTS
Since the service aims to merge the DDPO and the PTS, it must be a working replacement of both services and its set of features must englobe the features of the actual services.
Replace maintenance mailing lists
Packaging teams often separate the mailing list that gets the bug traffic and other notifications from their main discussion mailing list. This new infrastructure should entirely replace the former kind of mailing lists. Anybody receiving notifications and information directed to the package maintainer should get them via this new infrastructure.
It allows us to know how many people are notified for a given problem. If nobody is notified, the package is effectively orphaned. A more interesting case to detect is when several persons are being notified but all of them are MIA or marked as not being available for Debian (busy/in vacation).
Keep track of the list of maintainers
Packages using this infrastructure should automatically keep track of the list of their maintainers because anyone subscribing to a package must pick a "role":
- "maintainer" (only DD can select this role, or DM for their own packages)
- "contributor"
- "follower" (aka lurker)
Replace WNPP's RFH, RFA, O
Since the infrastructure already stores the information about who is maintaining what, it only makes sense to extend it to provide the list of orphaned packages (i.e. packages without maintainers).
RFH should be replaced by a system where the help request is better formalized so that we can better direct new contributors in places where their skills would be well used. Instead of just requesting help, you could request:
- a new maintainer (to replace you, i.e. RFA)
- a supplementary co-maintainer
- a bug triager
- a C/Python/Perl/… programmer (you should be able to choose the programming language)
Each request also documents whether there's an associated offer of "mentorship" associated to the help request. Of course, there would also be a free form description to give more details about what's expected.
Replace the LowThresholdNmu list
The LowThresholdNmu wiki page is a hack to let people know when NMU are welcome and not frowned upon. This information should be properly stored in the database and it should be associated to each package.
Provide better integration with the packaging VCS
A majority of packages are maintained in a VCS nowadays. It's thus important to extract the relevant information and make it easily accessible in the new interface.
Commits should appear in the usual activity stream of the package. We should be able to instantly see whether the VCS contains pending changes or not, or whether a new upstream version is in preparation there.
Enable new interactions with maintainers
The central role of this new "communication infrastructure" makes it possible to design new interactions with maintainers. Instead of being only a source of information, the infrastructure could be used to query package maintainers and/or let them provide supplementary information.
This could be used to improve the MIA tracking process.
This infrastructure would also be a more natural place to store the "available for Debian work" boolean flag ("vacation") that's currently never used because it's buried in db.debian.org and that it's not practical to update it.
This infrastructure could also be used to let maintainers document the responsibilities that they have agreed to endorse, and describe the associated commitments. That way it would be easier to detect packages that cannot be well maintained because the set of maintainers do not cover all the tasks that must be assumed to have a properly maintained package.
Provide new services to packaging teams
Given that this infrastructure would have native support for packaging teams, it would also be a good place to offer some standardized services for them.
For example, one of the central tool for teams are their VCS and it can be useful for teams to be able to monitor the state of their package in the VCS. The PET tool could be adapted, integrated and made available to all teams by default.
Support alternate notification systems
Email is the only official media used to communicate information to Debian package maintainers. If all the relevant mails are going through a central service, it's possible to store those emails and to forward the relevant information by other means (RSS, XMPP, IRC, etc.). Also new maintainers can then have access to some historic information that used to be private for no good reasons.
High-level design of the new infrastructure
Fixing the flow of information
In order to cleanly solve the problem of the information flow, and to get rid of the hacks made everywhere to send a copy of the mails to the PTS, packages would be (progressively) modified to indicate “Maintainer: <source>@pkgmaint.debian.org” in their control file. The current content of the field would be moved to the “Uploaders” field.
Until all packages have been converted, the PTS would forward copies of the mails to ensure that the new infrastucture can still be used for all packages (even those who have not been updated yet).
Using this intermediary address also solves the problem of maintainers who orphan their packages and are still listed as maintainers in many released packages.
QUESTION: It would be cleaner if DAK, the BTS, and all relevant services, could stop sending mails to the Maintainer field, and would instead always send the mail to the new infrastructure. That way we wouldn't need to change the Maintainer field and there would be no transition period.
The new infrastructure would then be configured with initial subscriptions for emails listed in Maintainer fields (except for mailing lists, since the infrastructure aims to replace them).
Leveraging UDD
At least the PTS has been parsing Sources/Packages files by itself, as well as a bunch of other source of information. But many of the most recent developments have piggy-backed on UDD to retrieve the information needed, leaving to UDD the responsibility of bringing all the information in a single place.
This principle should be generalized to avoid duplication of work and to make sure that all the important information are available in UDD.
But we must make sure that UDD won't become a bottleneck. Either because we have a local (live?) replicate of the database, or because we have ensured that our usage of UDD is limited to batch tasks that are not on the critical path for all the real-time user requests.
Using a modern framework for web development
DDPO is implemented in PHP. The PTS uses a mix of Perl, Python, XSLT and shell scripts. While both works very well and are reliable, the diversity of the tools and the fact that some are not widely known (e.g. XSLT) seriously limit the set of contributors who are able to hack on all the parts of the infrastructure.
With a modern framework for web development, we enlarge the set of people who are able to help us develop and maintain this infrastructure. It also offers us a proper separation between presentation and code, so that it's easier to let web designers integrate this service with the general look&feel of the various Debian websites. On top of this, we get a fully internationalized website for free.
API for data export
If the infrastructure is going to have a central role, there will be requests to extract data out of the system. We should cater for this by providing a public API (over HTTP) allowing to retrieve all the (public) information in some standardized manner.
JSON seems to be a good option for data export. It allows other services to reuse information from the DPMH, and it makes it easy for various web services to retrieve the information dynamically via Javascript.
Native support of packaging teams
Any Debian Developer must be able to create a "packaging team" in the system. Each packaging team has a set of packages that it maintains (or keeps an eye on). Anyone can "subscribe" to the team and gets (by default) all correspondance of all packages associated to that team.
The team subscription can be tuned (much like the current PTS subscription) to receive only a subset of the usual mails. A direct package subscription would take precedence over a team subscription, thus allowing the user to exclude some packages from its team subscription (or get more info for some specific packages where they are particularly interested).
Implementation details
Questions:
- How do we store emails? For how long? (we store all mails except the BTS mails)
- What language and web framework? (buxy's default choice: Python & Django)
- How do we authenticate users? And for DD super-powers?
- [To be completed]
Acronyms
- DPMH: Debian Package Maintenance Hub (this project)
- DDPO: Debian Developer's Packages Overview
- PTS: Package Tracking System
- BTS: Bug Tracking System
- UDD: Ultimate Debian Database
- DD: Debian Developer
- WNPP: Work Needing and Prospective Packages
- RFH: Request For Help
- RFA: Request For Adoption
- O: Orphaned
- NMU: Non-Maintainer Upload
Feedback
If you have comments about this proposal, please send them to debian-qa@lists.debian.org.
Changes
- 2011-01-13: Initial draft by Raphaël Hertzog.
- 2011-01-28: Integrate feedback from debian-qa@lists.debian.org.