From Fedora Project Wiki

Developments

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

Contributing Writer: Oisin Feeley

Fedora 11 Will Support i586 Instruction Set

Last week (FWN#161[1]) we reported on a proposal to cease building Fedora 11 for the i586 CPU instruction set. FESCo had delayed its decision in order to discuss the matter further. The issue was addressed[2] on 2009-02-05 with the outcome that a proposal by Dennis Gilmore to continue supporting i586 for the duration of Fedora 11 but to transition to i686 for Fedora 12 was supported.

Prior to the meeting Warren Togami summed up[3] the advice of Jakub Jelínek as: "Jakub recommends i586.rpm for Fedora 11, because it doesn't gain us much of anything to go with i686 minimum. The benefits of i586 to i686 are simply not important because cmov is usually not a worthwhile optimization on ia32."

An interesting suggestion by Adam Jackson was[4][5] that if there is a committed user-base of i586 users they could probably support it in the Secondary Architecture (see FWN#92[6]) infrastructure.

Ulrich Drepper and Dominik Mierzejewski debated[7] whether the use of cmov can in some circumstances cause performance degradation.

It is unclear exactly what performance benefits could be obtained by passing various architecture-specific flag combinations to GCC but it does seem that the burden of building and maintenance will be eased significantly by these changes. As a related change[8] x86_64 kernels will be installed with a 32-bit userspace.

RFC: Power Management

Phil Knirsch initiated[1] a discussion of attempts to decrease power consumption especially in userland. A wiki page[2] reflects some of the research Phil has pulled together.

Richard Hughes pointed[3] out some interesting work on DeviceKit-power where he built on powertop. Olivier Galibert raised[4] a possible problem with Richard's use of D-Bus itself causing wakeups, but according to Colin Walters a patch existed[5] to fix this problem.

Many of the items suggested in Phil's page for documentation were suggested by Bill Nottingham as desiderata for defaults. While Phil agreed[6] in general he itemized some of the problems. These include problems with network interfaces and hard-disk spindowns which may be approachable as a result of a tuned daemon on which Phil is working.

An addendum of audio hardware power-saving was made by Eric Sandeen along with a list of bugs which led[7] Phil to wonder if a tracker bug to collate all the information would be useful.

Matthew Garrett expressed[8] some worries that hard-disk power-saving would cause physical wear and the relatime patches to work around over-aggressive deletion of content in /tmp would continue to be stalled.

The importance of separating out KDE and GNOME dependent features was noted[9] by Kevin Kofler.

Rawhide Report 2009-02-07

The last report[1] lists 14 new packages added, 57 modified and some broken dependencies. New packages include dissy, a graphical front-end to objdump and python-pygooglechart a Python wrapper for the Google Chart API.

Richard Hughes suggested[2] that the update to PolicyKit-gnome-0.9.2-1.fc11 might be useful: "If you're having problems with PackageKit and buttons "not working" you need this update."

Some of the x86_64 broken dependencies were due to to mono-2.4 being pushed to rawhide which led David Nielsen to suggest[3] that a heads up would have been useful. Alex Lancaster requested[4] that API/ABI breakage would be announced on @fedora-devel-announce instead of on the high-traffic @fedora-devel.

New module-init-tools Uses Binary modules.dep|alias|symbols

An update to module-init-tools-3.6 was pushed to rawhide by Jon Masters in order to speed up[1] boot time significantly. The files modules.dep, modules.alias and modules.symbols will have binary versions which are used in preference to their old text versions. Jon asked[2] if the need to run depmod -a after upgrades to module-init-tools would upset anyone. There seemed to be general approbation of his changes and they should land soon for Fedora 9 also.

New Georgian Fonts Packaged Rapidly

A call was put out[1] by Nicolas Mailhot for someone to package a completely new Georgian font pack created by Besarion Paata Gugushvili.

Nicolas was especially keen to get this done quickly as he had contacted Besarion and been rewarded with completely new fonts not shipped by any other distro, licensed with the FSF font exception to the GPL all within nine hours!

Tom Callaway responded[2] within mere hours.

Distro-agnostic /boot Metadata Standard ?

A negative review in German IT magazine "c't" led[1] Christoph Höger to ask if it was possible to preserve the ability to boot other GNU/Linux distros after installing Fedora. The most annoying point seemed to be that Windows installations are preserved.

A moderately long thread resulted and covered several ideas to allow the GRUB bootloader to identify other distributions. One such was[2] .that there should be an agreement among distributions to use a shared metadata standard on boot partitions.

GCC-4.4 Mass Rebuild Successful

Jakub Jelínek reported[1] that a mass rebuild of rawhide (snapshotted on 2009-01-26) of 6228 packages had produced only a few hundred failures. He listed these by type of failure.

Several of the packages listed failed to build for reasons other than GCC, for instance Java packages failed[2] due to maven being broken.

Thorsten Leemhuis provided[3] a list of packages and owners sorted by owner which was generally appreciated. He pointed out: "Finding all your packages in such a long list gets really hard as soon as you maintain 10 or 15 packages."

Problems reported due to a mismatch between the libstdc++ headers requirement of -march=i486 and Koji's default use of -march=i386 led[4] Jakub to whip up some fixes. He requested that CFLAGS were not altered in SPEC files.

Help Rel-eng Accelerate Updates Processing

One bottleneck in the processing of updates to packages is that they need to be signed. Work is ongoing to automate this (see FWN#147[1]) with a signing-server codenamed "sigul".

Christoph Wickert wondered[2] why it had taken over five days for an update to one of his packages to get to testing. When Josh Boyer responded that it was because one human (Jesse Keating) had to sign the packages and he had been also busy getting Fedora 11 Alpha released, Daniel P. Berrange suggestedhttp://www.redhat.com/archives/fedora-devel-list/2009-February/msg00515.html adding more humans to help. Jesse Keating suggested[3] that anyone who wished to help could take some of the load off the release-engineering team so that they had more time for package signing.