[gentoo-releng] PPC release 2004.0 uploading

[Pieter Van den Abeele]

Hi all,

I'm pleased to announce the PPC release for 2004.0 is finally 
uploading. I had expected to be able to upload on friday, but 
discovered a few genkernel issues me and brad have been able to resolve 
in the meantime.
I will commit PPC specific release documents tonight. The cds should 
gradually become available on my internet server, I'll ping this list 
when everything is online and ready to be reviewed for quality.
Please note that uploading might take some time.

A short overview of this release:

Each bootable CD has four highly modularized 2.6.3 kernels on it who 
provide multi-processor and uni-processor G3, G4 and G5 support. So the 
same cd is able to boot your dual G5 and ibook G3, as well as your 
powerbook G4 (assuming you own those machines :-) ) Since I had to 
start hacking genkernel to fix some issues, I've also hacked in support 
for a few other things, such as:

- A configurable cd spin up time, (affected older scsi cdroms closing a 
very old bug)
- Initial support for initrd-level automatic module loading. (when 
accessing /dev stuff etc.)
- A cleaned up abstraction layer allowing bootstrapping live 
environments from devices other than the ide cdrom (live-boot gentoo on 
your ipod, firewire camera, or usb memory stick)
- Reduced number of dependencies for the genkernel tool (completely 
busybox based)
- Starting and killing devfsd on the initrd no longer a requirement. (I 
have never used devfsd in a livecd initrd. For some reason the catalyst 
built devfsd, not part of the busybox suite, locked on ppc. The new 
devfsd is tested on ppc and is no longer needed :-) )
- Per kernel based specified module loading. (to complement automatic 
module loading when accessing /dev)
- A trigerable debug initrd shell  that can be used to debug the cd 
bootstrapping process but also provides minimal tools to install gentoo 
(dhcpcd, vi, network modules, ...).
- Cleaned up output and commented my design choices for linuxrc (-ng?).

I hope some of these features can be integrated in genkernel for 
2004.1. I'll see to it the current genkernel maintainer can go over my 
code and toss out the stuff he doesn't like. I have clearly documented 
the parts which need to be changed to the current catalyst linuxrc to 
work on ppc (mostly included devfsd things, which imho isn't really 
needed, at least not on ppc) and which parts are just cleanups or new 
features.

Besides the 700M gcloop compressed livecd CD1 (featuring stages for 
generic ppc and G4 plus sources for everything stage1,stage2,stage3 
plus X), I also provide a mini-version of the cd, (+-80M) which doesn't 
have the distfiles, the stages or the snapshot. (It does have the 
handbook and the readme). The GRP cd is also 700M in size and features 
xfce4, kde3.2, gnome2.4, koffice, openoffice and so on (will provide 
details in the ppc release docs which are to be committed asap).

The cds feature tested thermal management for both G4 (windtunnel) and 
G5 (turbine). The user needs to activate that manually (modprobe 
therm_p72 on the G5 and modprobe therm_blahblah for the G4), support 
for radeon, rivafb, openfb, rage128, aty, ... framebuffers. Arguments 
to the different framebuffers are documented (Booting a mac into 
1600x1200 console looks much nicer than the default 640x480 we used 
until now).

I have also committed catalyst specs for a kde/gnome cd and have 
produced a kde/gnome PPC LiveDVD, which features both GRP, distfiles, 
stages and a kde/gnome live environment. The dvd version will not be 
uploaded. People interested in it should contact me, or can assemble it 
themselves using catalyst as component factory.

One thing that frustrated me was a feature I've come to like very much 
with my old ppc tool: a resuming capability. Right now, if an ebuild 
fails to emerge, catalyst restarts all over again: unpacks the stages, 
unpacks all previously emerged packages (pkgcache),... I must have 
spent hours unpacking things last week. Chances for failure increase if 
you're building 320 packages. This gets especially frustrating in 
combination with my number two frustration: the inability to predict 
the packages that are going to be installed. I've spend at least an 
hour looking for that one package which required xfree to be installed 
on an xfree-less cd :-) This is the "ah, damn, it emerges xfree again, 
lets ctrl-c, tweak something and wait 10 minutes to see portage 
emerging xfree again" type of frustration, which tends to get bigger on 
every iteration. The advantage of a resuming capability is that it is 
faster for debugging, while still allowing to restart all over again 
for cleanness. I have written down these experiences in REMARKS in the 
catalyst main directory, I hope they could be useful for future 
catalyst coding iterations. And I'm also hoping 2004.1 is going to be a 
just start the script and burn the cds operation :-)

Regards,

Pieter Van den Abeele


--
gentoo-releng@g.o mailing list