Table of Contents
Each Debian Pure Blend has an own mailing list for discussion of specific development issues. Because there are several common issues between all Blends also a common mailing list was created. People who are interested in working on common issues like building metapackages, technical issues of menu systems or how to create CDs for Blends could subscribe to this list or read the list archive.
git clone https://salsa.debian.org/blends-team/blends.git
for anonymous users. Developers should check out via
gbp clone email@example.com:blends-team/blends.git
The current layout for the repository is as follows:
blends -+- blends (code of blends-dev and this documentation) | +- website (code to create the web sentinel + other tools) | +- med (Debian Med) | +- science (Debian Science) | +- ... (most other Blends) | ...
If a user installs Debian via official install CDs the first chance to do a package selection to customise the box is tasksel. The first Debian Pure Blend Debian Junior is mentioned in the task selection list and thus it is clearly visible to the user who installs Debian.
In bug #186085 a request was filed to include Debian Med in the same manner. The problem with the tasksel-approach is that all included packages should be on the first install CD. This would immediately have the consequence that the first install CD would run out of space if all Blends would be included in the task selection list.
How to enhance visibility of Debian Pure Blends for the user who installs Debian from scratch?
If the packages must be on the first CD feature of tasksel would be dropped all Blends could be listed under this topic in the task selection list.
Alternatively a new feature could be added to tasksel or in addition to tasksel in the installation procedure which presents a screen which gives some very short information about Debian Pure Blends (perhaps pointing to this document for further reference) and enables the user to select from a list of the available Blends.
By completely ignoring the installation of the official installation CD each Blend can offer a separate installation CD. This will be done anyway for certain practical reasons (see for instance the Debian Edu - SkoleLinux approach). But this is really no solution we could prefer because this does not work if the user wants to install more than one Blend on one computer.
This way is concerned to some ideas from Debian developers who took part in Open Source World Conference in Malaga and is explained in Detail in Section 9.6, “New way to distribute Debian”. This would save the problem of making Debian Pure Blends visible to users in a completely different way because in this case Debian would be released as its various flavours of Blends.
Whichever way Debian developers will decide to go it is our vital interest to support users and show our users the tools we invented to support them.
Some Blends maintain their own web space under http://www.debian.org/devel/BLEND-name to provide general information which will be translated by the Debian web team. This is a good way to inform users about the progress of a project. This page should link to the apropriate autogenerated pages as described in Section 6.2.4, “Web interfaces” to make sure that the content of the page remains up to date at any time.
Debian Package Tags is a work to add more metadata to Debian packages. At the beginning it could be seen as a way to allow to specify multiple sections (called "tags") per package where now only one can be used.
However, the system has evolved so that tags are organised in "facets", which are separate ontologies used to categorise the packages under different points of view.
This means that the new categorisation system supports tagging different facets of packages. There can be a set of tags for the "purpose" of a package (like "chatting", "searching", "editing"), a set of tags for the technologies used by a package (like "html", "http", "vorbis") and so on.
Besides being able to perform package selection more efficiently by being able to use a better categorisation, one of the first outcomes of Debian Package Tags for Blends is that every Blend could maintain its own set of tags organised under a "facet", providing categorisation data which could be used by its users and which automatically interrelates with the rest of the tags.
For example, Debian Edu could look for "edu::administration" packages and then select "use::configuring". The "edu::administration" classification would be managed by the Debian Edu people, while "use::configuring" would be managed by Debian. At the same time, non Debian Edu users looking for "use::configuring" could have a look at what packages in that category are suggested by the Debian Edu community.
It is not excluded that this could evolve in being able to create a Blend just by selecting all packages tagged by "edu::*" tags, plus dependencies; however, this option is still being investigated.
Please write to the
more information about Debian Package Tags or if you want to get
involved in Debian Package Tags development.
In section Section 6.2.5, “Future handling of metapackages” several issues where raised how handling of metapackages should be enhanced.
Currently there is no solution to address the special configuration issue has to be addressed. In general developers of metapackages should provide patches for dependent packages if they need a certain configuration option and the package in question does feature a debconf configuration for this case. Then the metapackage could provide the needed options by pre-seeding the debconf database while using very low priority questions which do not came to users notice.
If the maintainer of a package which is listed in a metapackage dependency and needs some specific configuration does not accept such kind of patch it would be possible to go with a cfengine script which just does the configuration work. According to the following arguing this is no policy violation: A local maintainer can change the configuration of any package and the installation scripts have to care for these changes and are not allowed to disturb these adaptations. In the case described above the cfengine script takes over the role of the local administrator: It just handles as an "automated-cfengine-driven-administrator-robot".
If there is some agreement to use cfengine scripts to change configuration - either according to debconf questions or even to adapt local configuration for Debian Pure Blend use in general - a common location for this kind of stuff should be found. Because these scripts are not configuration itself but substantial part of a metapackage the suggestion would be to store this stuff under
There was another suggestion at the Valencia workshop: Make use of ucf for the purpose mentioned above. This is a topic for discussion. At least currently Debian Edu seems to have good experiences with cfengine but perhaps it is worth comparing both.
A further option might be Config4GNU from freedesktop.org but it is not even yet packaged for Debian.
The first step to convince a user to switch to Debian is to show him how it works while leaving his running system untouched. Knoppix - the "mother" of all Debian-based live CDs - is a really great success and it is a fact that can not be ignored that Debian gains a certain amount of popularity because people want to know what distribution is working behind the scenes of Knoppix.
But Knoppix is a very common demonstration and its purpose is to work in everyday live. There is no room left for special applications and thus people started to adopt it for there special needs. In fact there exist so many Debian based Live CDs that it makes hardly sense to list them all here. The main problem is that most of them containing special applications and thus are interesting in the Blends scope are out of date because they way the usually were builded was a pain. One exception is perhaps Quantian which is quite regularly updated and is intended for scientists.
The good news is that the problem of orphaned or outdated Live CDs can easily solved by debian-live and the live-helper. This package turns all work to get an up to date ISO image for a Live CD into calling a single script. For the Blends tools this would simply mean that the tasks files have to be turned into a live-helper input file and the basic work is done. This will be done in a future blends-dev version.
This section is kind of "Request For Comments" in the sense that solid input and arguing is needed to find out whether it is worth implementing it or drop this idea in favour of a better solution.
At Open Source World Conference in Malaga 2004 there was a workshop of Debian Developers. Among other things the topic was raised how the distribution cycle or rather the method of distribution could be changed to increase release frequency and to better fit user interests.
There was a suggestion by Bdale Garbee
think about kind of sub-setting Debian in the following way: Debian
developers upload their packages to unstable. The normal
process which propagates packages to testing and releasing a
complete stable distribution also remains untouched. The new
thing is that the package pool could be enhanced to store more package
versions which belong to certain subsets alias Debian Pure Blends
which all have a set of tested inside the subset distribution
which leads to a stable subset release. The following graph
might clarify this:
DD -> unstable --> testing --> stable | +---> BLEND_A testing --> stable BLEND_A | +---> BLEND_B testing --> stable BLEND_B | +---> ...
where BLEND_A / BLEND_B might be something like debian-edu / debian-med. To implement this sub-setting the following things are needed:
There was a general agreement that technical implementation
of this idea in the package pool scripts / database is not too
hard. In fact at LinuxTag Chemnitz 2004 Martin Loschwitz
<firstname.lastname@example.org> announced exactly this as
"nearly implemented for testing purpose" which should solve
the problem of outdated software for desktop users as a goal
of the debian-desktop project. Unfortunately this
goal was not realised finally.
Once the promotion tools are able to work with sub-setting, reasonable subsets have to be defined and maintained. A decision has to be made (if this will be implemented at all) whether this sub-setting should be done according to the Blend layout or if there are better ways to find subsets.
The Bug Tracking System has to deal with different package versions or even version ranges to work nicely together with the sub-setting approach.
As a consequence of having more than only a single stable each Blend team has to form a security team to care for those package versions that are not identically with the "old" stable.
A not so drastically change would be to find a common set of packages which are interesting for all Debian Pure Blends which will obtained from the "releasable set" of testing (i.e. no RC-bugs). This would make the structure above a little bit more flat:
DD -> unstable --> testing --> releasable --> stable | +---> stable BLEND_A | +---> stable BLEND_B | +---> ...
A third suggestion was given at Congreso Software Libre Comunidad Valenciana:
testing_proposed_updated | | v DD -> unstable --> testing --> stable | +---> stable BLEND_A | +---> stable BLEND_B | +---> ...
The rationale behind these testing backports is that sometimes a Debian Pure Blend is able to reduce the set of releasable architectures. Thus some essential packages could be moved much faster to testing and these might be "backported" to testing for this special Blend. For instance this might make sense for Debian Edu where usually neither mainframes nor embedded devices are used.
All these different suggestions would lead to a modification of the package pool scripts which could end up in a new way to distribute Debian. This might result from the fact that some Debian Pure Blends need a defined release cycle. For instance the education related distributions might trigger their release by the start-end-cycle of the school year. Another reason to change the package pool system is the fact that some interested groups, who provide special service for a certain Blend, would take over support only for the subset of packages which is included in the metapackage dependencies or suggestions but they refuse to provide full support for the whole range of Debian packages. This would lead to a new layout of the file structures of the Debian mirrors:
debian/dists/stable/binary-i386 /binary-sparc /binary-... /testing/... /unstable/... debian-BLEND_A/dists/stable/binary-[supported_architecture1] /binary-[supported_architecture2] /... /testing/... debian-BLEND_B/dists/testing/... /stable/... ... pool/main /contrib /non-free
To avoid flooding the archive with unnecessarily many versions of packages for each single Debian Pure Blend a common base of all these Blends has to be defined. Here some LSB conformance statement comes into mind: The base system of all currently released (stable) Debian Pure Blends is compliant to LSB version x.y.
Regarding to security issues there are two ways: Either one Debian
Pure Blend goes with the current stable Debian and thus the
Packages.gz is just pointing to the very same versions
which are also in debian/stable. Then no extra effort regarding to
security issues is need. But if there would be a special support team
which takes over maintenance and security service for the packages in
a certain Blend they should be made reliable for this certain subset.
This reduced subset of Debian packages of a Debian Pure Blend would also make it easier to provide special install CDs at is it currently done by Debian Edu.