Bobulate

Oracle has released Solaris Express 11. That’s a binary-only preview of Solaris 11, which picks up somewhere after OpenSolaris was killed. The kernel identifies itself as snv_151a. As an exercise in futility, I decided to install this on my Sun Ultra 45 workstation — a SPARC machine. Of course I’ll be doing so only to evaluate the software I am creating (or packaging) for the Operating System, not putting the machine into production use — which is what the onerous license agreement (it’s even difficult to find the URL to link to!) demands.

I have a SPARC-based workstation under my desk, from the time that we in the KDE4-Solaris project actively targeted SPARC; that’s some time ago. The machine is still there and I like the architecture, so I try to use it every now and then. With OSOL this was always problematic. Solaris 11 seems to change that a little.

Solaris 11 comes with a text-based installer for SPARC, which is good enough. It doesn’t seem to come with a graphics driver for the XVR-2500 card that’s in the workstation, though, as the boot process tells me that the pciegfx domain will not be used.

For me it was a little confusing that some GNOME packages were installed — like gnome-audio — but no X server or display manager. Text logins are fine, but in order to test my software at some point I’m going to have to actually run X on it. Given the speed at which this machine compiles (random hardware begging: anyone have 2x1GB DDR PC2700 ECC DIMMs left over? Or four of them, for that matter? How about an UltraSPARC IIIi CPU @1.6GHz?) I’ll be happy to get through to Qt sometime in the next seven days. And let’s face it, Qt 4.7 keeps trying to outsmart the KDE4-OpenSolaris team by saying that it’s partly incompatible with Sun compilers — but in the case of Sun Studio 12 on SPARC it might be right 🙂

Turning the SPARC machine into a usable GNOME box took steps like these:

pkg install xorg evince gnome-terminal gdm
pkg install xorg-video kfb xorg-kfb fbconfig-kfb
pkg uninstall xorg-video xorg-video-wsfb

That last uninstall was to get rid of the wsfb driver which crashes Xorg. After mucking about a bit I got some advice on IRC, from the indomitable Alan Coopersmith in particular. You can find some of his work (newer Xorg) on the OpenSolaris forums. So two essential things: "fbconfig -dev kfb0 -defaults" and "pkg install slim_install", then reboot. So thanks, Alan.

The machine will come back but still be in text mode. SSH is enabled by default, so you can SSH in from another machine on the network and start the gdm service to get a login prompt. with “svcadm enable svc:/application/graphical-login/gdm:default“. If you watch the console, after about 15 seconds you get a gdm login screen. On the one hand, it’s tastefully done with some transparency applied around the session and language selectors; on the other hand there’s this giant Oracle logo on there like “abandon all hope ye who enter here.” And a kind of icky hex grid background and default wallpaper, as if drone-like creatures are going to lay their proprietary eggs in your brain and cause you to surrender your will. Ugh.

Why does the Oracle logo look so out of place on a GNOME desktop where the OpenSolaris one didn’t? Seems strange to me.

Anyway, it seems that Oracle have delivered a technically sufficient OS release for SPARC workstations. I can start compiling stuff on it now (some setup scripts will surely be added to the -460 repo as I carry on with that). Making a SPARC machine a usable development workstation takes another whole bunch of packages, just some of which are the Mercurial and the C++ compiler (you could choose gcc instead, but that’s not the mandate of the KDE4-OpenSolaris project). The Sun compiler is not available from the Solaris 11 package repository right now, so get it from the older OSOL repo:

pkg install mercurial
pkg set-publisher --non-sticky \
  -g http://pkg.opensolaris.org/release/ opensolaris.org
pkg install sunstudio12u1

The old adage "if it ain’t broke, don’t fix it" applies, but nonetheless I felt the need to fix my laptop. It was cluttered with FSFE materials that I shouldn’t be carrying around, for one thing, and the Kubuntu 9.04 on it was decidedly long in the tooth. As prepwork for Akademy (o harbinger of doom!) I decided to clean it up: one Linux install, one OpenSolaris. This is an MSI 620GX laptop, which is a Centrino 2 based machine. Hardly exotic stuff.

For social reasons — as in, Sebas had recently written glowingly about it — I started off with OpenSUSE 11.2. Installs nicely (but with GPT by default, it seems?) and delivers a good-looking KDE4 desktop, plenty of apps. Compositing was enabled (GeForce9600M GT). Setting up a devel environment was a mild challenge. For various projects I use svn, git, mercurial and darcs, so getting those is a first priority. Darcs was a little harder to get, but there is a package available, which I downloaded and installed manually. It was in one of the repositories, but I didn’t feel like setting that up for one package.

It’s when I tried to suspend to RAM or to disk that issues started showing up. Suspend to RAM fails, saying that the machine is unknown and not whitelisted. s2ram -f puts the machine to sleep, but it doesn’t resume. Similarly, hibernate (suspend to disk) works but doesn’t resume. I still need to send in the info for that, but after a half hour of fiddling with it — and knowing that Kubuntu 9.04 could suspend and resume on the same laptop, I gave up. Since I’m not particularly attached to whatever Linux I’m using, time to try something else.

Kubuntu 10.04 is what I’m running on my desktop — which has ATI graphics — and I appreciate that it starts up really quickly, etc. Vaguely annoyed at the microblogging thing it puts on the desktop by default, but that’s terribly minor. Installing all the dev tools was easy on the desktop. On the laptop, though, I didn’t even get that far. The nouveau driver included on the install CD doesn’t like the video card — and so the installation process bails out to a text screen. Folks in #kubuntu were helpful and ready with some suggestions, like nomodeset and using the vesa driver (hung the machine on boot). Running X -configure from the text login hung the machine too.

Fedora 13 up next. No compositing with the nouveau that is included — that’s in the mesa-experimental package, it turns out. Devel environment is easy to get, with all the version control systems one install command away. Of course, the first thing I tried this time was suspend and hibernate: both flawless. External monitor — important for presentations at Akademy — pops up a dialog with simple configuration. There’s one third-party application that I use that requires 32-bit libraries. Getting those was straightforward after finding out the package names with "yum provides ‘*/libraryname’". I see that they’ve also customized Konversation to go to multiple useful channels, rather that just the distro-channel.

So, it seems I’ll be presenting at Akademy from a Fedora-based laptop (Rex, Kevin, a beer is on me). All I need now is the latest Air-themed LaTeX templates and I’m good to go.

One thing I’m left with is why three different Linux distro’s, all relatively recent, behave so differently on a fairly conventional platform like this one. The technology is there; it was even there last year. Where do these regressions come from?