Ubuntu Traffic

Welcome to the third edition of Ubuntu Traffic. This issue covers the week of September 4 - 10 in 2004. Ubuntu Traffic summarizes the most important mailing list and IRC discussions involving the Ubuntu GNU/Linux distribution.

The big release announcement goes out next week (and will be out by the time you read this). There was a (predictable) flurry of activity in preparation for this.

Bits and pieces that didn't get a full story but are worth mentioning include:

Jeff Waugh announced that the Ubuntu team would be having a Tuesday Meeting. The meeting, predictably enough, centered around issues in the upcoming release. Many of the other threads summarized here touch on issues either first raised or discussed in th meeting.
Mary Gardiner pointed out that there is a real lack of dial-up tools in Warty. The group concluded that wvdial, which is universe, is less ideal but the best we have at the moment.
Jeff Waugh suggested that we reverse the sudo non-caching decision from last week as, after week of using it, it is very annoying. He seemed to get nothing but nods.
Jeff Waugh (again) suggested that we shift Abiword and Gnumeric back to the supported seed to avoid confusing users and the MIME system.
Jeff Waugh (yet again) suggested that we move Planner back to the supported since it did not fit the "greatest common factor" criteria. Mark Shuttleworth suggested that we leave it in as it's a good example of a best-of-breed application that gives the platform a richer, more comprehensive feel.

This week, the Sounder 8 CD was released (described in a story below) and there were a rash of new installs and install reports to go with them. Here's a very brief overview of the reviews and reports we saw on the list.

Sivan's Report

Sivan did an install onto a machine with a D865PERL mother board, ICH5 chipset, an an nVidia GeForce2 GTS 32MB Video Ram video card. Siva thought, Installation was a nice new breeze compared to my first experience with bf2.4 and sarge d-i, everything was seem to be detected correctly, and set up accordingly. Was also blazing fast (i wonder if it's my hardware that's to blame for this solely :) ) also and pretty straight forward

Sivan had the following errors to report:

Trying to choose my language yielded "Installation cannot continue" red curses window.
On the package deployment and installation phase, there were countless packages that reported to be missing dependencies and thus couldn't be installed at the time.
I don't know how important that is, but as someone used to hear music on the works, I believe the layman would not know what would be the right CD /dev entry for him to tell the CDPLayer about it.
Menu layout seems nice and appealing, however I'd suggest adding some more preset stuff to the desktop, like the Home folder link and maybe some other User specific presets.

James Gregory's Install

James had the following issues or suggestions:

I want it to install on /dev/hdg1. It has booted up and presented me with a menu with "Erase entire disk: ..." selected as the default. It scared the shit outta me, I imagine it would worry a newbie too (and how often do people want to obliterate their whole drive? It seems like it's only a sensible default for an empty hard drive).

James had issues with dual booting and having everything work: Lessing both /dev/hde1 and /dev/hdg1 shows an NT boot loader on hde1 and GRUB on hdg1. I have the sinking suspicion at this point that installing GRUB has destroyed the XFS partition. James continued and suggested we might be able to learn some things from the dual boot support in Mandrake.

Matt Zimmerman followed up with some useful advice:

Most Linux filesystems leave the space at the very start of the device empty, because it is traditionally used for things like boot loaders. XFS, on the other hand, places its superblock, containing precious filesystem metadata, in the very first block.

Some tools, such as lilo, attempt to implement a safeguard for this by testing whether the device appears to contain an XFS filesystem, and if so, refusing to proceed. The grub installer does not seem to do this at the present time. This check is simple to implement, but it hasn't been done (yet).

Because it is so important, both XFS and ext3 maintain backup copies of the superblock data. If your XFS filesystem has not been further damaged, it may be recoverable. I would look first to xfs_repair, but I personally do not use XFS at all, so take my advice with a grain of salt.

The two, Matt and James, tried to walk through the problems on the list in a good deal of detail that may be useful to others installing on a similar setup (RAID + XFS + DualBoot).

Jeff Waugh Install On His New Laptop

Jeff Waugh did an install onto his laptop: a Dell X300, very Centrino but with a Broadcom (tg3) wired NIC and a CD-ROM attached via the 'mediabase' base station:

Language = English
Country = Australia
Big red background, "Detect and mount CD-ROM: An installation step failed." Hitting continue drops me back to the installer main menu, so I try that step again. It dives straight into the next step, loading the installer components as if nothing happened! Wacky. Colin's gone to bed, but I can run through this again with live coverage. :-)
Network autodetection ran fine, it correctly chose the tg3 driver for my Broadcom wired NIC.
Disk partitioning, manual because I want to have a separate /home.
Big red background, "Copy packages to hard disk: An installation step failed." Trying this again did not work.
Chose to install the base system from the main menu, "Debootstrap Error: Failed getting Release file /cdrom/dists//Release." Weird, it doesn't know the release name.

Matt Zimmerman replied saying, I have a strong hunch that it's related to hdparm. Edit /var/lib/dpkg/info/cdrom-detect.postinst, comment out the hdparm stuff, and retry. Jeff Waugh followed up with, Hey hey hey, good hunch! No initial error... Loads the installer components straight off... Copying packages to disk! Hooray! A bug was filed for removal of the offending code.

David Miller Sounder 8 Install Notes

David Miller had this to report about his latest stab at an install with the Sounder 8 CD:

When detecting network settings, it again choked because my wireless has WEP enabled. It asked me for an ESSID, but never asked me for the WEP key, so it had no way to connect.

After manually entering network settings, there was a LONG pause with nothing but a blue screen before it prompted me for a hostname. I'm assuming it was probably trying to reverse-dns the IP I gave it to find out if it already had a name, but since it never asked for a WEP key, the DNS didn't work. Bug 1113 filed.

I backed up, dropped to shell, did iwconfig eth1 key xxxxx and then had it try again, and as expected, no pause, and it came up with the hostname from my DNS pre-filled in the box.

Another long pause with nothing on the screen after writing the ext3 partition before it starts checking the swap space on the swap partition.

After that, everything went really smoothly, up until I logged into gdm...

The /dev/pmu permissions error is still there, and the "Error activating XKB configuration" is still there (I'm assuming that hasn't changed since the last log I mailed to daniels, since that was only a day or two ago. so I won't mail another one unless asked).

Colin Watson replied that 1113 was fixed.

Cef Logs in Another Install Report

Cef did another install and had some great feedback. This included:

At the boot prompt:

F2 & F8 points people at the Installation Manual and the FAQ on the Debian web site. We need Ubuntu versions of these documents, and this stuff updated.
F9 attributes the Debian project (and implies it is in fact a Debian install), but doesn't attribute or list any info for Ubuntu at all. Even a placeholder is better than leaving it as is, otherwise it'll be missed.
F10 only lists copyrights and warranties for Debian, and nothing at all regarding Ubuntu. Same applies as per F9.

The after install unpack installed 1241 MB of files. I'm guessing that it decided that since I have an 80 GB drive I can afford to install all this stuff, which I notice includes gcc/make/etc. This is a LOT more than what the installation requirements available at the boot prompt suggest (which says 256 MB of disk).

All installed and gdm comes up. Logged in, and it all seems fine, except that the virtual desktop is 1280x1024, and the real desktop is 1024x768. I didn't even notice at first it wahttp://www1.optusnet.com.au/memberservices/servicestatus/?brand=OCABs virtual, simply because I thought the background image wasn't working properly.

Fabio Massimo Di Nitto followed up to say the issue in the final paragraph was fixed.

The thread turned blossomed into a discussion on keyboard layouts and locales. At the moment, the keyboard layout is being chosen based on inferring that can be done from the locales. While this can cause some confusion, it's being done to streamline the process and will be easier for most users who neither know about nor care about keyboard locales.

Later, Colin Watson followed up to Cef's first set of critiques saying that the branding in the bootloader had been updated.

Martin Pitt, Sounder 8, and Martin's PowerPC

Martin Pitt had the following to report about his install attempt of Sounder 8 onto his PowerPC:

base-config: the password creation dialog is not translated to German. Colin, may I upload a new package with the translation or send you a diff or so?
when it comes to installing the packages, there is a loooong pause after installing laptop-detect. A small line like "Preparing installation, this may take a while" would be nice here.
When booting, I still get a ton of "device-manager: dm-linear: Device lookup failed" messages.
Whoah, my snd-powermac sound card finally works out of the box! However, rhythmbox still does not play anything because snd-pcm-oss and snd-mixer-oss are not loaded. Does this happen only on ppc or doesn't it happen on i386 as well?

David Miller started a thread with a long list of usability critiques of Ubuntu. Dave suggested that we should have the default behavior in Nautilus be browse mode. Mark Shuttleworth followed up to ask:

We should optimise for the common case. 90% of the time when I'm double-clicking on a folder inside nautilus, I want to open that window and close the current one. Is that typical for everyone? If so, should we not make it so that double-clicking opens the new folder (in spatial mode, remembering where it was on the desktop last time etc) and closes the existing folder, and holding shift or using the middle button leaves the current window open while opening the new one?

Straw poll: when you navigate through folder hierarchies in spatial mode, do you usually want to leave the old folder open, or have it closed when you move to the next folder?

Jeff Waugh followed up to say that, Both, depending on what I'm doing... and when I need to *navigate* through a lot of folders, then I use navigational mode ('Browse'). Note that a straw poll on a list of very highly technical users is not a good way to figure out what to do. :-) Jeff also made a number of suggestions of other ways we could take this. Scott James Remnant answered Mark's question saying, The really interesting thing about Spatial mode is how it does optimise for the common case for ordinary users, but power users such as ourselves react (initially) violently to it because it doesn't optimise for us.

There was also some discussion on the list of having a composting manager that will fade the old unused open windows of Nautilus out from behind in a very useful and graphically appeal way.

Fabio Massimo Di Nitto put another another call for testing a a series of X packages. He was interested in feedback on hardware detect tools and autoconfiguration, on reconfiguration, and on configuration without hardware detection tools.

The thread got a good amount of feedback that was almost entirely positive in nature. Good work Fabio!

Sebastien Bacher posted a long list of issues he'd been considering with the GNOME system we're shipping. Since he listed far to many issues to bring up here, I'm summarizing only the issues that generated traffic on the list. One issues Seb brought up was, The wireless and battery applets in the default desktop displays errors if you don't have a wireless device or a battery. What's the decision about this ? (#945)

Mark Shuttleworth replied asking Seb if it would, be possible for these not to display an error dialog, but to display an image which suggests that the applet is not needed, such as greying it out?

A solution, using the laptop-detect was proposed by Mark Shuttleworth:

For warty: we will just have a nice "disabled" icon that is less scary, and the user can remove the applet from the panel.

For future, our best proposal is an install-time check that decides what the default panel should look like. If you want that to happen each time the machine boots, then remember we have to figure out if the user has already decided to remove that panel. So for example, if I don't have a wifi card, it shouldn't show up. Then I boot with a wifi card, and it should show. I decide I don't need it and remove it. Now I boot again - it should NOT show... because I already decided I didn't want it.

Elsewhere in the thread, Scott James Remnant responded about the icons that are actually on the desktop itself: One thing you're missing (I think) is a "Documents"/"Home" icon on the desktop. I created a new user today and this didn't appear, which means there's no obvious way for someone to get at their files. We either need to restore the "Home" icon, or put a "Documents" icon there. There was some contention about the usefulness of these things on the desktop.

Matt Zimmerman sent a message to the list saying, I just noticed that build-essential and its dependencies only add about 15M on top of the existing desktop install. How about we promote build-essential to desktop? Build-essential is a virtual package in Debian that includes compilers and most of the common tools that users need to build packages on a Debian system. Jeff Waugh replied to Matt saying saying, I would strongly suggest that 'secure by default' means 'no compilers for the script kiddies'. Let's not do this, it freaks me out. There was heated conversations about the benefits and drawbacks of shipping a compiler.

Mark Shuttleworth jumped in to say that, Agree with Jeff here - they don't belong half in Desktop, they belong in Ship.seed, and it should be possible to get the WHOLE bang shoot (perhaps even -dev packages for all Desktop libraries etc?) with a single command.

Martin Pitt brought people up to speed on the current way of unmounting devices and asked for suggestions on a better way of doing this saying:

To unmount, the user currently has to open the Computer window, right-click on the device and select "Unmount" / "Eject". This option should be always present, but it is too obscure. The user did not use the Computer window to access the device, and before he finds the Computer window and the context menu, he will have ripped out the USB stick (or, even more dangerous, the CD from the locked drive).

So how do we unmount in a more obvious way? During the discussion, several options were proposed:

Try to unmount the device if the device Nautilus window is closed; pop up a warning if it isn't.
Unmount the device as soon as no process accesses it any more. (Beware that an open Nautilus window counts as accessing the device).
Just let the user rip out the device; hal will detect this and gvm can unmount it afterwards. Devices are mounted with 'sync' anyway, so this should not damage VFAT devices too badly.
Use automount/autofs/subfs or similar
Add a small icon to the panel if a device is mounted. Clicking would attempt to unmount and pop up an error box if it fails (perhaps with the applications that still use the device).
Add a small icon to the desktop if a device is mounted. Left-click could bring up a Nautilus window with the content, right-click the usual context menu with the unmount option. Same error box here if device is busy.

In a follow up thread, Martin Pitt announced the decision that said we should plan for 3 because it's going to happen anyway and then try to focus on 5 and 6. Speaking on the last two, Martin said: These two seem to have general acceptance. However, according to Mark we should aim to implement 5. Volunteers? Since 6 should already be implemented, how long it would take to undo the modifications for suppressing the icons? I'd like to know how long our fallback solution will take. At release time minus this time we should drop the efforts of implementing 5 if it does not work by this time.

Thom May said he would try to minimize the damage being done with #3 while Sebastien Bacher said he could make the configuration change for the last one.

Colin Watson released the next Sounder CD -- numero 8.

Sounder CD 8 is ready:

http://ftp.no-name-yet.com/cdimage/sounder-test/8/
rsync://ftp.no-name-yet.com/cdimage/sounder-test/8/

See http://wiki.no-name-yet.com/WartyWarthog/Archive for access instructions. I recommend rsync if possible, as you can then download future images based on this one to save bandwidth. Installer improvements in this release:

The Back button no longer causes pathological debconf priority behaviour. There have been a number of other related usability improvements.
Problems with CD-ROM drives seeking about a lot during install should have been largely alleviated.
FireWire Ethernet support is gone, following consensus that it was good for little more than causing confusion.
The network configurator behaves better when DHCP is unavailable.
The "Copy packages to hard disk" feature is now enabled by default. Booting with archive-copier/copy=false (or answering no to the relevant question in expert mode) was supposed to turn it off, but a late-discovered bug means that that doesn't yet work.
If "Copy packages to hard disk" is enabled, then you no longer need the installation CD inserted after the first reboot.
If package selection fails, you'll be dropped into aptitude rather than stranded in a half-configured system.
gdm is now localized after the second and subsequent reboots as well as the first.
The snd-powermac module is loaded automatically on modern PowerMac systems.
amd64 images are now built entirely from the Warty archive. (Until recently, the installer's initrd was externally built.)

I haven't been able to keep track of all the other changes in Warty, but among other things, we now have our own kernels based on 2.6.8.1, pmount and friends have been polished up a lot, we have new prettier init scripts, most launchers should now use gksudo rather than gksu to acquire root privileges, and the desktop is settling down towards its final form.

If you're upgrading, you may want to try with a freshly created user account to pick up changes to the default desktop.

The main errata I know of so far are that powerpc installations will have trouble with some modules from /etc/modules not being loaded (bug #1088) and with XKB errors when GNOME is started (bug #1079). These should both be fixed soon.

Bug reports should go here: https://bugzilla.no-name-yet.com/