Hoary Kickoff Meeting ---------------------- This meeting was held Monday 25 October 2004, at 16:00 UTC, in the #ubuntu-meeting channel on irc.freenode.net. Summarized by: Benjamin Mako Hill Full Log Available: Agenda ------- * Review release schedule * Discuss `Hoary Merge Process`_ * Review feature goal list - Identify "early breakage" items which should go into Hoary as soon as possible - Initial feature goal assignments .. _Hoary Merge Process: http://www.ubuntulinux.org/wiki/HoaryMergeProcess Discussion ----------- Release Schedule ================= The first agenda item was the release schedule. Jeff Waugh pointed people to the `GNOME release schedule`_. Jeff suggested that were not any changes significant enough to discuss as this point. Matt Zimmerman said he was happy to let Colin Watson, "tweak the CD milestone dates as you feel is appropriate; you might want to wait for Jeff Waugh's changes though." .. _GNOME release schedule: http://www.gnome.org/start/2.9/ Hoary Merge Process ==================== Merging in things between hoary is going to be, ergh, hairy. Scott James Remnant did some analysis on the number of packages that have changed and where. Merging is going to be complex because we now have three version we care about: the version in warty, the old version in Debian that we forked from, and the new version in Debian. Merging the two version is going to be tough and there's not a clear Correct Way to do it. James Troup announced that the merge infrastructure was mostly ready although he needed two things before things were ready to go: a) proper seed lists for hoary b) decision on what, if anything, we're doing about indices files for hoary? Matt Zimmerman pointed out that, "regarding the seed lists, let's use the Warty seed lists; we can update them later. We'll need to have a review of the proposed seed changes, and I expect we won't get to it during this meeting." Matt Zimmerman proposed a tool to automatically import unmodified packages and for modified packages (those with an ubuntu version number), send out a notification: probably a simple email to a mailing list to start saying that the package needs review. Fabio Massimo Di Nitto suggested that we should track these in Bugzilla. This seemed like it would be a bit too heavy for some people but Matt Zimmerman agreed to do the initial test that way. The plan is to ignore universe right now and focus on main. In terms of getting an idea of the merge, Scott James Remnant analysis showed 10,704 "rejects" and around 7,000 of those with same changes on each side: plus 3,596 left as different changes to both sides. Some 2,500 of those in .po files and debian/changelog or debian/control. Warty has some 295,004 hunks. Mark Shuttleworth pointed out that any magic Scott can bring to the next two weeks will make him friends for life. Scott pointed out that, "Colin Watson's hunch was right ... it actually is easier to apply the debian changes to warty than try to go back and apply Warty's changes to debian." The key question is "how much of this can we realistically automate." The team called in Dafydd Harries to ask him about how to deal with po files since many of the changes involve po files. They asked him how to deal with a single version and two sets of patches to it and how to merge po files. The trick seemed to be unpacking the branch point package and working from there. Another big issue was the changelog patches which will always conflict due to diff/patch being how they are. That's something we should be able to consistently resolve automatically with a special tool. There was some discussion on the virtues of having a separate Ubuntu changelogs which Mark Shuttleworth was advocating but that a number of other team members were not as thrilled about. Scott James Remnant pointed out that, "if we can work out a way of automating a given type of conflict, I can put that logic into hct so it can do it automatically later." Matt Zimmerman summarized the situation saying, "the initial merge strategy is to let Scott James Remnant lock himself in a room for a day and then create bugs on the remainder." Feature Goals ============== The group then moved into discussion the feature goals of the distribution. UTF-8 ++++++ Matt Zimmerman introduced the conversation saying, "we'll set UTF-8 by default early in the release cycle, and just fix whatever breakage comes up. It's really a bunch of unclassified bugs at this point, rather than a feature." Matt asked what we should do about existing installations. The suggestion was to leave them at non-UTF8, send out an announcement asking people to change. This seems the most sane option because as folks poitnedout in the meeting, "we can't change the encoding automatically; this would break _every_ text file the user created." The other suggestion is to provide a tool which the user can run which will Do The Right Thing. Unified Hardware Detection +++++++++++++++++++++++++++ The next topic was unified hardware detection. The idea is to have a single hardware detection system for the: - installer - installed system - live CD Currently, Ubuntu is using three (discover1, hotplug and Knoppix). Matt Zimmerman's position was that hotplug should be the target for unification and this seemed to have a good deal of support. Colin Watson pointed out that hotplug is one of the post-sarge goals for the Debian installer but that we could move forward on this ourselves. He also pointed out that hotplug doesn't do X stuff so discover was still needed. The topic then moved on to X and brainstorming about how to unify the Live CD and the install CD in terms of X. Fabio Massimo Di Nitto suggested that he could, "create a script that simulate an installation to detect the hardware as I do in the normal installation and create a live config." Mark Shuttleworth stated that some of this new work should be done in Python. That would mean that Python would become an essential package. A number of team members expressed some alarm at this but Mark Shuttleworth suggested that this was not something he was going to be willing to budge on. One issue is that python is very big and that how one would split it up into a python-base was very unclear. Since hotplug was chosen, most of the work left is Live CD work and that would left up to Andreas Mueller to do. The final issue will be work on the X system so that it can be done in a dynamic reconfiguration way which will bring the Live CD and install CD into line with each other. Daniel Stone pointed out that the frame buffer for X almost never goes wrong but that hardware detection dynamically sometimes does so pure unification may simply not be what we want. GUI Installer ++++++++++++++ The team briefly brought up graphical installers. Mark Shuttleworth stated that, "GUI installer is on the hoary list, but it has a question mark on it and I won't commit to having a GUI installer for hoary as it will back us into a corner." That said, it something that people will start working on. PowerPC64 ++++++++++ It was proposed that the team not attempt PPC64 for hoary. Matt pointed out that there is currently no real vacuum for it to fill. Mark said that the team won't attempt any further arch's unless a community team steps up. James Troup pointed out that we'll need to do enough of PPC64 for G5 support. Matt said he expected that they would build a kernel and pointed out that this was noted in Bugzilla and discussed with Herbert Xu who is working on the kernels. Matt Zimmerman pointed out that this was a multiarch-wanting arch too which was another problem we might not be able to solve in the Hoary Debconf Question Priority ++++++++++++++++++++++++++ The team started discussion the virtues of changing the priority of debconf questions that are asked during a ``dpkg --reconfigure``. Currently, it is set to low priority questions which asks lots of questions. Some people thought this was a good thing while others did not. Michael Vogt pointed out that this could be accessed in a graphical way through Synaptic. Mark Shuttleworth set the final answer saying, "let's go with medium, but then you guys are going to have to put up with a lot of bugs from me in that regard." Mark is interested in only seeing relevant and important questions. SE Linux +++++++++ The team tried to put off serious discussion of SE Linux. People pointed out that the part they liked most was Mandatory Access Control (MAC). Matt Zimmerman thought it sounded like the sort of thing that would happen best in a derivative. Thom May pointed out that Fedora has been having serious trouble making it usable. It will require very serious work on both glibc and dpkg and a lot of other places as well. Mad Phat Startup +++++++++++++++++ The proposed boot splash implementation is usplash which works in userpace using the kernel frame buffer rather than patching the kernel itself. The alternatives to using vesafb are fb and X. The group seemed to agree that this was an issue for a bounty and decided to delegate this to Paul Sladen if he wanted this. Matt Zimmerman volunteered to manage this bounty. NTP ++++ The question was whether we should NTP differently for Hoary; i.e., using ntpd rather than just running ntpdate. This comes from the fact that that gnome-system-tools integrates with ntpd and not ntpdate. The obvious course would be to change g-s-t rather than to use ntpd. That was added to the bounty list. Speeding up the Boot Process +++++++++++++++++++++++++++++ The first issue that was discussed was hotplug. Matt Zimmerman noted that he had written a version of hotplug in perl that sped things up quite a bit but reverted it because it uses /usr/. Scott James Remnant pointed out that he had written on in C that he would look for. GDM was the other issue -- this one was delegated to Daniel Stone. Kernel Grab-bag +++++++++++++++ Most of the items on the agenda had to do with fixing the various warts in how we load kernel modules. The fact that that IDE stuff doesn't Just Work is the big one. Another issue was figuring out the right strategy for drivers which are no longer autoloaded with udev. Matt suggested that unless anyone here has a strong interest and the domain knowledge for it, it might make a good bounty. Encrypted Home Directories +++++++++++++++++++++++++++ Martin Pitt suggested that encrypted home directories should work out of the box. Colin Watson pointed out that partman was designed with support for encrypted filesystems in mind but that it has not been implemented yet. Matt Zimmerman thought it was a lot of complexity for a default install. Martin Pitt said that it could be but needed not be cryptoloop and that the users need not really see any difference. He also said it would need installer support to work from the very beginning. There seemed to be some interest although there was also some concern about supportability. Mark Shuttleworth it might be another issue for a bounty. Kickstart ++++++++++ Kickstart is Redhat's automated semi-custom install system. The short version if that we want to have it. Our system should be compatible in terms of the format of the file but not in terms of the data that we're describing. Fancy Keyboard Selector ++++++++++++++++++++++++ The keyboard selector would deduce your keyboard layout by what you type. It will require broad knowledge of keyboard layouts and differences and it will need to seed everything that needs keyboard layout information. Another good bounty candidate. Proxy/Authentication before Attempting First apt-get update ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ This was relatively straightforward (the code is already there) but it's an additional question in the installer which fell to the "few questions" ax. Currently, we test to see if we can reach archive.ubuntu.com and then don't ask the question if we can. What we are losing here is caching proxies. This was left as a pending decision. CD-based Upgrade +++++++++++++++++ The idea is to have an autorun script that will prompt the user and will, if it is a warty system, offer to upgrade it to hoary using the new versions of packages on the CD. Michael Vogt said he would take this. Installer Bootable from CD/USB +++++++++++++++++++++++++++++++ Everyone agreed that these (especially the latter) would be great features. The suggestion was putting the netboot kernel onto the USB at the moment. Debian can do this with the business card image but since we don't have one and the full image is a bit much for most USB devices, it might be best to create a new image for this. Put this one down for Colin again. Laptop Support +++++++++++++++ Matthew Garrett talked about laptop support and suspect. Basically, the problem is with getting suspend and resume to work well with laptop lid closures and opens. The problem seems to be that not all notice when the lid has been open and can call software suspend at that button. It's a lot more straight forward with the power button. This one is in Matthew's court. ACPI/APM +++++++++ At the moment, if your ACPI doesn't load or is not supported, you will not get APM instead. Scott James Remnant suggested, "can't we load acpi, and then apm -- apm will not load if acpi was loaded and worked." Matthew Garrett thought, "the apm init script could check for a /proc/acpi, and load apm if it's not there." Thom May took control of this one. Language Support +++++++++++++++++ Matt Zimmerman said, "the big part of this bit is the backend infrastructure to pull things out of packages during the build cycle. Which is a generic facility we may use for several things. The basic idea is to create language packs by extracting locale-specific data from packages." The team was split on how to do this and where the correct time and place was. The spec as it stands is that we want to have language packs which include the localisation data across the distribution, and exclude that data from the packages. X.org ++++++ Fabio Massimo Di Nitto announced that the work on x.org was going slower than was initially expected. They thought it should be done by the end of November possibly. There was some discussion about the monolithic versus split nature of the upstream and the packages. Matt Zimmerman and Mark Shuttleworth were concerned by how long it was taking to get it done but it was unclear what could be done to make it go faster as Fabio Massimo Di Nitto is already working extremely long hours. It was suggested to work from monolithic source at first rather than splitting but Fabio said it was going to have to be done right from the beginning. Matt Zimmerman made the tough choice to not split the source right now because it wasn't essential for the two major goals: - managing the upgrade from xfree - testing the software itself Daniel Stone will be visiting Fabio to hack on things from the 1st of November through the 14th and the goal is to have something done before he leaves. Sound Daemons ++++++++++++++ The proposal was to replace ESD with polypaudio. This is going to be a following GNOME's lead thing so we want to do it. However, it will involve breakage -- which we know. Palm Syncing +++++++++++++ There was a good deal of discussion about replacing GNOME's palm only sync with Multisync which supports a wider number of devices. However, multisync does not support Evolution 2.0 (only 1.4) and has some other pretty critical shortcomings. There was a suggestion to use opensync instead of multisync. Sound Theme ++++++++++++ The team discussed the virtues and need for a sound theme. The feeling was that we definitely want one. Mark Shuttleworth saw it as two separate issues really: * creating a good sound theme * Making sure all apps which use sound are correctly integrated to the theme: So gossip and GAIM could use the same sounds -- thunderbird and evolution -- for new mail, new IM, etc. Basically, the suggestion was to file bugs on packages where there are sound theme inconsistencies or unusabilities. Documentation Goals ++++++++++++++++++++ On the agenda was the need for - Python development documentation love - Ubuntu in a nutshell style booklet There was support for the need of more Documentation goals for Hoary as well. Hardware Survey ++++++++++++++++ There was discussion about an idea thrown around at the last conference about a system of finding out what hardware people have and then integrating this data into scripts that the autoconf uses. The problem was broken down like this: a) the design of the database b) the app that collects the data c) figuring out what the data means d) integrating it with the scripts that autoconf the setup like x, sound, network, etc Sivan also suggested that it could be used for finding other people with similar hardware to help reproduce bugs. A hardware matchmaker as it were. Other Issues +++++++++++++ After about 3.5 hours of the meeting, people started to tired and brushed quickly some of the issues. I've included some of those here in a slightly abbreviated form. - There was talk about having a GUI way of choosing available languages rather than the current GDM way of choosing languages. - A python port of yelp was suggestion which was deferred as bounty issue. - The group discussed the need for testing infrastructure. Mark Shuttleworth felt strongly that this should done internally and not sent out on a bounty because we're going to have a to live with it for a long time. However, we still need a default specification and some candidates. - There was discussion about package authentication as well. James Troup, Matt Zimmerman, and Mark Shuttleworth have discussed this in the past and had some answers. Michael Vogt will have to be involved as well as it will need some interface magic with synaptic. - The group gave the node to bzip2 packages. It is clear that bzip2 packages will need to Pre-Depend: dpkg (>= 1.10.24) - There was some discussion about a tool that was more user-oriented program than synaptic. Something that would make installing packages easy.