FWN/Beats/Developments

Developments

In this section the people, personalities and debates on the @fedora-devel mailing list are summarized.

Contributing Writer: Oisin Feeley

Fedora 11 Alpha May Be Delayed

Jesse Keating reported that the Fedora 11 Alpha release date might slip due to some anaconda bugs which manifested themselves late in his testing on some architectures. A later post suggested that installation using NFS was broken and that "[t]his likely means a slip, perhaps only a two day slip, of Alpha." More info to come either later this weekend or early next week. A bugzilla comment from Warren Togami on a side-effect of trying to fix this problem by reverting to an earlier nfs-utils version warned "People should be aware that NFS as a server in F11 Alpha is broken. That is all." As of going to press on 2009-02-01 there was no further information available.

GCC: Default ISA Flags and Glibc

Jakub Jelinek asked whether the minimum CPU which would run code compiled by Fedora 11's GCC should be re-evaluated. A follow-on question was whether the minimum supported kernel version in glibc could be bumped to 2.6.29. Jakub held out the promise of potentially increased speed and decreased shared library sizes.

A problem raised by Kevin Kofler was that mock builds would no longer be able to run on older Fedora releases and that some VPSs would not be able to upgrade at all. Gerd Hoffman agreed: "We just can't make the huge jump from .9 to .29. We have to do it smaller steps, considering kernel versions at least in supported Fedora versions, maybe also latest RHEL."

Josh Boyer seemed to believe that the required mass rebuild with GCC-4.4 would be difficult but possible. Mike McGrath outlined the amount of work which would be needed.

See this same FWN#161 "Dropping Support for i586 Architecture" for a related discussion.

RPM Packagers: Too Many Unowned Directories

Michael Schwendt raised the problem of unowned directories installed as a result of packagers unfamiliar with "how to include files vs. directories in RPM package %files lists."

Colin Walters remembered discussions which had suggested that if RPM were able to reference count directories there could be a technological fix. Separately Richard W.M. Jones made a similar argument. Panu Matilainen seemed willing to move this task to the top of his queue if it were sufficiently important.

Lack of Update Information

A can of worms was opened by Rahul Sundaram when he noticed that the update information provided by package maintainers was often unhelpful. He cited generic messages of the form "Update foo to upstream x.y.z" as a common problem and wondered if guidelines could improve the situation.

Following some questions Rahul expanded on the problem pointing out that package maintainers had the knowledge to tersely explain what upstream changes implied for ordinary users. He emphasized that he was concerned with the "description that is part of bodhi update and not the changelog which can be very brief."

Chris Weyl put forward the counter-argument that package maintainers had a difficult enough life already.

Richard W.M. Jones wondered if rpm could be altered to allow it to reference upstream changelogs which could be pulled out by other tools. Panu Matilainen averred that while rpm was alterable Richard's proposed change would just dump the information into the rpm payload and it would thus not be available to users until after they had installed it. Further brainstorming seemed to run into various practical dead ends.

Subsequently Rahul published a draft guideline which fanned the flames back to life. ThorstenLeemhuis asked "Don't we have way [too] many guidelines and policies already? [...] Note that I don't disagree with the text that was proposed. My 2 cent: Put it as text into the wiki somewhere, write "best practices" on top of it (avoid the words "rules" and "guidelines") and add a link to the bodhi UI ("best practices for filling this box with information")." Rahul appeared to agree that this was the best course for the present and deferred to FESCo for the ultimate decision.

Electronic Design Automation Content Without Tools ?

Chitlesh Goorah redirected a debate on Electronic Design Automation (EDA) tools from FESCo to @fedora-devel. Chitlesh is the prime mover behind the Fedora Electronic Lab Spin. He was concerned that FESCo had decided that packages in the OVM format were barred from Fedora on the grounds that there was no FLOSS tool which could use them although they were licensed acceptably.

Jef Spaleta explained that there were subtle problems in the discussion as "[OVM] is code of some sort. The problem is we don't have a compiler or interpreter that can process the instructions. In the context of Fedora its code that can't be used." Kevin Kofler supplied the appropriate guideline.

Kevin Fenzi expressed appreciation for Chitlesh's work on the Fedora Electronics Lab and asked if there was any use for OVM besides hooking it up with a non-Free simulator? Manuel Wolfshant argued that OVM was i[...] interesting for a subset of the people interested in EDA" and that it should be provided for them. Horst von Brand disliked the idea of mirrors carrying such a little-used package around and suggested that Manuel could just set up his own repository.

Dropping Support for i586 Architecture ?

Following FESCo discussionsBill Nottingham reported that the supported architecture list was going to change. Important changes include building binaries only for i686 and above. There are concerns that older thin clients based on i586 hardware and the AMD Geode-based XO laptops may then be unsupported or unstable. Bill characterized the discussions as a follow-up to the compiler flag discussions (see this same FWN#161"GCC: Default ISA Flags and Glibc") and summarized the main points as:

- install x86.64 kernel on 32-bit OS where appropriate
- install PAE kernel on other 32-bit OS installs where appropriate
- build only i686 and above for Fedora

Jeremy Katz added anecdotal reassurance that the XO should probably be fine with the i686 kernel and glibc.

Robert Scheck wondered what the definition of "where appropriate" was and what mechanism would be used to make this determination.

Dominik 'Rathann' Mierzejewski predicted "[t]here's going to be some screaming from VIA C3 and AMD K6 users about this." His suggestion was true during an older similar discussion (see FWN#93) in 2007 which concerned plans to drop shipping an i586 kernel. Suggested attempts to compensate by making the i686 kernel bootable on i586 architectures were thwarted as rpm balked at installing a kernel which violated its architecture check. Alan Cox was one of the strongest objectors to the possibility of thus losing support for i586 as he had many thin clients using that architecture. Doubt was cast during that thread as to whether the smolt statistics were believable. However, Alan has recently become an Intel employee (following other ex-Red Hat luminaries David Woodhouse and Arjan van de Ven) and did not contribute to the thread. The smolt statistics listed on the feature page suggest that there are only 130 i586 users.

Josh Boyer clarified that no decision had yet been made by FESCo and that a vote would take place next week.

Blinking Cursor Wastes Power

Matthew Garrett asked for comments on the idea that the cursor should default to not blinking. The rationale was that several less Watts of power would be consumed.

The suggestion seemed generally popular but Dominik `Rathann' Mierzejewski wished to retain the blinking cursor and expressed a desire for more information on the methodology which Matthew had used. Bill Nottingham reminded that it would still be possible to turn the cursor back on from this new default. Matthew provided some of the requested details.

Matthias Clasen suggested changing a GTK setting which disables cursor blinking after a timeout. Josh Boyer worried about other desktop environments and vttys.