Here's how.
First, it's assumed that you're running Firefox. AFAIK, there is no analogous functionality for other browsers (please comment below if you know otherwise).
Now:
Now, when you load a page on ScienceBlogs (and any of several other blogs) each post will have two new bits of hot text in the Posted by: line: [kill], which hides all comments by this poster in every thread; and [hide comment], which hides only the single comment.
In place of a hidden comment, a one-line notice appears saying “Comment by User Foo blocked [unkill] [show comment]”. Clicking on [unkill] reveals all of the posts by that user; clicking on [show comment] reveals just the single comment.
]]>In all cases, you use HTML (or HTML-like) tags, which (with a few exceptions) occur in opening/closing pairs with your text in between, like this (for an imaginary tag “foo”): <foo>some text</foo>. You have to be very careful to get the syntax right or you will get unexpected results. Preview is your friend.
In order to quote a previous text, use <blockquote>. If you write
“<blockquote>John McCain is a tool.</blockquote>”, it comes out looking like this (indented and possibly with other decoration, such as a bar down the left):
John McCain is a tool.
You can get italics with <em> or <i>, like this: “John McCain is <em>definitely</em> a tool” produces “John McCain definitely a tool”.
You can get bold with <strong> or <b>, like this: “John McCain is <strong>definitely</strong> a tool” produces “John McCain definitely a tool”.
You can strike text with <s>, like this: “John McCain is <s>not</s> a tool” produces “John McCain is not a tool”.
You can underline text with <u>, like this: “John McCain is <u>such</u> a tool” produces “John McCain is such a tool”.
You can produce links with <a>. The text to be linked appears between the elements, and the URL appears in the href attribute. “<a href="http://www.google.com/">Google</a>” comes out as “Google”. Note that the quotes around the URL and the http:// are not optional. Also, including more than a certain number (2 or 3) links may result in your comment being held for moderation as an anti-spam measure. Preview is your friend.
Not that it makes much difference in the tag soup that is the blogosphere, but to satisfy the pedants, I should note that the <em> element is preferred for emphasis and <i> for non-emphatic italics. Similarly, <strong> is preferred for strong emphasis and <b> for non-emphatic emboldening.
]]>The first problem is to find the stale diversions. If you type something like:
$ dpkg-divert --listYou'll get several lines like this:
diversion of /usr/lib/libGL.so.1.2 to /usr/lib/fglrx/libGL.so.1.2.xlibmesa by xorg-driver-fglrx
Since you want to filter out the ones that apply to the ATI drivers, and all of them contain fglrx, you can do this:
$ dpkg-divert --list | grep fglrx
The bit you want is the first path. You can extract that with either:
$ dpkg-divert --list | grep fglrx | cut -d' ' -f3
or
$ dpkg-divert --list | awk '/fglrx/ {print $3}'
You should get a list like this:
/usr/lib/libGL.so.1.2 /usr/X11R6/lib/libGL.so.1.2 /usr/X11R6/lib32/libGL.so.1.2 /usr/X11R6/lib32/libGL.so.1 /usr/lib32/libGL.so.1.2 /usr/lib32/libGL.so.1
You can then manually go through the list, removing the diversions one-by-one:
$ sudo dpkg-divert --remove /usr/lib/libGL.so.1.2
Alternatively, if you're highly confident in your own bash-fu skills:
$ dpkg-divert --list | awk '/fglrx/ {print $3}' | \
> while read; do \
> sudo dpkg-divert --remove $REPLY; \
> done
The above is shown broken over four lines just so that it isn't too wide for the web-page: in practice, I would type it all on one line. This is what is “really looks like” completely: the backslashes (\) are typed in and escape the newline which must follow immediately. Bash supplies the > character at the start of each line, which is the secondary shell prompt.
Job done!
]]>The reason for this post is to inform the staggering number of Catholics who are indignantly posting on PZ's Pharyngula blog without understanding the background. It's as if they're assuming that PZ woke up one morning and decided to piss off Catholics for no reason.
A 19 year-old Catholic student at UCF went to a mass organised on campus. He brought a non-Catholic friend, who was interested in Catholicism, to the mass. During the mass, he explained various aspect of the proceedings to his friend. His friend was curious to see the Eucharist, so at communion, Webster tried to bring the Eucharist back to his seat before consuming it. In Webster's church, this wouldn't have been a big deal as, apparently, in some Catholic traditions, it is quite normal to pray back in the pew before consuming the Eucharist.
Another student, with whom Webster had had some previous disagreements, spotted him walking away from the altar without having immediately consumed the wafer, apparently the norm in most Catholic congregations, and grabbed hold of him, creating a scene. Webster shook off his assailant and went back to his seat. She pursued him, trying to take the Host from his hands. Webster tried to fend her off, putting the Host in his pocket, but ultimately fled the church.
Webster demanded an apology from the other student, holding on to the Host in his student dorm for almost a week before returning it.
Now, there's no doubt that Webster had a rush of blood to the head and behaved unwisely in his anger, but who can say they never made a mistake?
In the mean-time, the offended student made complaints to the university authorities, the student council, and the Catholic League. Bill Donohue weighed in with a fatwa condemning Webster Cook. As a result, Webster received 1000's of condemnatory emails including threats of violence and a couple of death-threats from the lunatic fringe of the Catholic church. Ultimately, Webster believed that he was in danger, which contributed to his decision to return the Eucharistic wafer.
A couple of media outlets picked up on the story.
PZ blogged about this crazy storm-in-a-teacup. Now, one may think that what Webster did was wrong, but it was a religious wrong and the penalty should be religious. If his priest had given him a dressing-down for his behaviour, no problem, but Webster did not deserve threats of violence and death, nor did he deserve to be subject to a witch-hunt in his secular life, with people demanding that he be dismissed from the student senate at UCF and expelled from the university. Separation of church and state: remember?
PZ was, I guess, pretty angry. In defence of this kid, he said “You think that's desecration? I'll show you desecration”. Not being a Catholic, he said “if anyone can score me some crackers, I'll desecrate them”, which was interpreted by some as inciting disruption of religious services, but the point is that ruining a kids life over a magic cracker is grossly excessive.
Now all hell breaks loose as Bill Donohue issues a second fatwa against PZ. PZ receives tens of thousands of condemnatory emails including a number of threats of violence and even a few death threats.
A fairly common feature in the tens of thousands of posts at Pharyngula was “you wouldn't dare desecrate the Qur'an, you pussy” from Catholic posters. PZ received several communion wafers and two Qur'ans in the mail.
Today, PZ posted a photograph of some communion wafers, some pages from the Qur'an, and some pages from Richard Dawkins's The God Delusion in the bin. He held off on posting the photograph until after the garbage collection, presumably so there would be no question of the religious artifacts being anywhere other than the city landfill.
]]>I see a huge, if not the, important issue in this election being the balance of the SCOTUS. The new President is going to appoint at least one liberal judge almost immediately (Stevens, I think, is hanging on until there's a Democrat in the Whitehouse), and maybe two (+Ginsburg?) in his first term. If McCain gets in, I see one or two more Nino Scalias being appointed, tipping the balance firmly to the right for a long time (Scalia won't be the same age as Stevens is now for almost 20 years). The consequences could be church/state separation being compromised, Roe vs. Wade overturned, gay marriage being torpedoed, and even successful appeals to recent judgements disallowing educational crapola in schools. It could be devastating.
As an international spectator, I selfishly want Obama because I think he'll be able to use his mixed-race, mixed-heritage, international background, foreign policy education, and civil rights record to help build peace and rehabilitate America's overseas reputation in a way that another old militarist Republican WASP, up to his oxters in political debt to the hawks and the religious right, simply couldn't do. From an international perspective, Obama is damn-near perfect: realistically, the international community couldn't ask for a better prospective President.
But it strikes me that, from an American and, ultimately, even an international perspective, the SCOTUS issue could be even bigger in the long term: I really don't want American core values (1st Amendment, particularly) and education compromised for a generation.
Thoughts?
]]>Sounds easy enough, but it's both surprisingly difficult and surprisingly easy to achieve.
The problem is that the ways of achieving this I've found online embed something in the distribution files that require that the person building the software for installation has to have subversion or the Autotools or both installed in order to get a meaningful version number. That's a pain.
This solution requires only that the developer is working on a Unix-like system with sh, tr and a recent version of subversion installed.
The solution is really this easy. Put the following in configure.ac:
define([svnversion], esyscmd([sh -c "svnversion|tr -d '\n'"]))dnl AC_INIT([myprog], [0.1.svnversion], [myprog@example.com])
I also include a line like this so I can get the SVN version seperately:
AC_DEFINE(SVN_REVISION, "svnversion", [SVN Revision])
This assumes that you have a recent version of subversion installed with the svnversion program and you're developing on a Unix-like platform with sh and tr (which includes pretty well every flavour of Unix/Linux).
What this does is use m4, the macro language autoconf is built on, to define a macro, svnversion which uses the m4 function esyscmd (which calls a program and captures its stdout) to call the svnversion program and trim the bothersome trailing newline from its output using tr.
Now, one apparent shortcoming turns out to be a feature, not a bug. If you run autoreconf in a directory that's been modified from the repository version, the subminor revision number obtained from svnversion is non-numeric — it embeds information about the revisions used, not just the topmost revision — so you get a subminor revision like “42:43M”. This causes some difficulty building because the SVN_VERSION macro (if you include it), which should be be a valid C literal, is not an integer literal. So why is this a feature and not a bug? Because you shouldn't be building the software for distribution from a stale repository and this ensures that you're not doing that. In practice, it turns out to be very useful, rather than a pain, since it helps makes sure that make distcheck will fail if you've modified files and not done a svn commit and/or svn update.
]]>The problem is an error in the preinstall script for the package. In a Debian package, one of the things that a package script sometimes does is “divert” files. The idea is that a package can specify that any attempt by a future package to install a file having a particular name is to be handled by renaming the file to something else, or diverting it. In this way, a package can protect files from being overwritten without causing another package to fail to install. If you have customised a package-supplied file, which might be overwritten by a package upgrade, you can use dpkg-divert to protect your version.
The ATI xorg driver package uses this facility to stop MesaGL, a software Open GL renderer, from overwriting the ATI hardware drivers. That's a good thing, but the problem is that the preinstall script diverts two files to the same target. In other words, it says “instead of installing /foo/bar, rename it /foo/quux” then a second later, “instead of installing /foo/baz, rename it /foo/quux as well”. That's not allowed (because it's obviously stupid).
This accounts for the seemingly bizarre error-message that xorg-driver-fglrx_8.476 is trying to divert to a file target that it, itself, has already diverted another file to!
I suspect that the person who wrote the script copy'n'pasted the line, remembered to edit the filename to divert, but forgot to change where it was diverted to. An easy, if silly, mistake to make. ATI have made a lot of progress in providing decent Linux driver, maybe we shouldn't be too hard on them for lousy QA.
The fix follows below the fold…
]]> The FixIF THIS BREAKS, YOU GET TO KEEP BOTH PARTS
In principle, the fix is very easy:
In practice, it's a little tricky to do manually, so I've provided a shell script to help make it easier: fix-cat84.sh.
It seems that the postrm script is similarly "challenged", but I didn't know this until Chris (below) brought it to my attention. I've done the best I can to fix the postrm script too (it's straightforward), but I would have to remove the existing package, install the new one, and then remove it in order to test the new postrm script, and that's a bit too much like hard work, so I haven't tested it. When an update to the driver appears, I'm going to have a problem removing the old package and I'll post my solution when that happens.
If you decide to go ahead, here's the instructions:
For example (if you like to live a little dangerously):
$ mkdir tmp $ cd tmp $ wget http://www2.ati.com/drivers/linux/ati-driver-installer-8-4-x86.x86_64.run $ sh ati-driver-installer-8-4-x86.x86_64.run --buildpkg Ubuntu/hardy $ wget http://caulfield.info/emmet/src/fix-cat84.sh $ sh fix-cat84.sh
You should now be able to install the packages in the usual way.
]]>Now, the velocity of propogation of light in a medium is determined by the permittivity and permeability of that medium. From these quantities, we can also calculate the characteristic impedance of the medium. We generally think of media as linear, isotropic, and homogeneous, but this is not the case in general. We often think of monochromatic light but, again, this is not the case in general. When light passes from one medium to another, a difference in the characteristic impedance of the media leads to reflections at the boundary.
What Trefethen started me thinking was, is there a kind of “permittivity” of a grid and a “permeability” for a method which would give a wavenumber-dependent “characteristic impedance” for the grid/method combination and a different characteristic impedance for boundaries, owing to either imposed or numerically necessary boundary conditions, and allowing instability to be determined easily?
Generalising to an irregular mesh, the mesh analogue of permittivity would be a wavenumber-dependent discrete tensor. However intimidating it might sound, and however intractable it might be for a human to calculate, it would surely allow a computer to predict, given a mesh and a method, what Fourier modes would cause instability and, perhaps more importantly, where in the mesh this might arise. A potential end-product might be a software tool which takes a mesh from, say, COMSOL Multiphysics, and identifies “hot spots” (analogous to “shiny lumps” in an optical medium) where particular Fourier modes would reflect, giving rise to instability. Given the Fourier transform of a particular initial condition or forcing function, instability might be predicted and avoided by mesh refinement.
Now, it may be that this analysis would end up being equivalent to solving the problem on the mesh in full generality, or be obscenely computationally costly for some other reason, in which case, I admit, the entire idea is utterly worthless, but there's no harm in asking, right?
]]>I intend to annotate this with hyperlinks to relevant online documents and other media. If you can suggest links, please do. I'm particularly interesting in a video of one of Stephen Meyer's lectures: in the Q&A at the end, an evolutionary biologist makes a fool of Meyer by describing a half-dozen precursor structures to the rotary bacterial flagellum beloved by cdesign proponentsists.
This post is dedicated to the public domain. Do with it as you wish.
]]> The List| Republican | Democrat | European Term |
|---|---|---|
| socialist | moderate | conservative |
| ordinary person | wingnut | Republican |
| Godless | atheist | secular |
| moonbat | rational person | Democrat |
| democracy | majority rules | corporate puppetry |
| Libertarian | Libertarian | demented fruitcake |
| pinko | socialist | centre/moderate |
| Godless commie bastard | communist | socialist |
| fair and balanced | right-biased | blatantly bigoted |
| commie propaganda | left-leaning media | media |
| leftist grant whore | college professor | university lecturer |
| God-given right | war | illegal invasion |
| President | asshole | war criminal |
| fair election victory | narrow loss | bloodless coup d'etat |
| communism | socialism | social democracy |
| Marxist pipe-dream | universal healthcare | healthcare |
| fellow American | religious right | Christian Taliban |
| small government | libertarianism | anarchy |
| help ally | foreign policy initiative | bankroll tinpot dictator |
| Ann Coulter | that Coulter bitch | fascist fundy bigot (female) |
| Bill O'Reilly | douchemonkey | fascist fundy bigot (male) |
Eventually I will tart it up, customise it, and move stuff over from emmetcaulfield.net, but until I find my way around Movable Type, this default skin will have to do.
Jag bor i Uppsala, men efter halvannan år, min svenska är ännu jätte dalig. Ibland ska jag skriva nagonting på svenska också och jag hoppas att det skulle hjälpa mig lite när jag försöka att förbättra min svenska. Just nu kan jag läsa ganska bra, men min riktig problemerna med språket är att förstår talat svenska och att skriva. Om du ar svensk, inte vara buskablyg, och snälla hjälp mig: korrigera felen i texten!
De temps en temps, c'est possible que je vais écrire quelquechose en Français. J'espére que ça m'assisterai pendant que j'améliore mon Français, mais maintenant, j'habite en Suede et je veux apprendre la Suedois surtout. Si vous êtes un francophone, s'il vous plaît corrigez mon Français.
Anois is arís, b'fhéidir go scríofaidh mé cúpla focail as Gaeilge. Chaith mé trí mhí sa ghaeltacht Chonamara nuair a bhí mé óg, ach rinne mé dearmad ar an cuid is mó den ghaeilge a bhí agam san am sin. Tá fhios agam nach bhfuil mo gramadach lán ceart. Tá spéis agam mo chuid Ghaeilge a chur ar fheabhas, ach mar tá cónaí orm i tSualainn anois, táim chun an Sualainís a fhoglaim mar tosaíocht. Anseo, do chuala siad faoi teanga "Éireannach", ach ceapann an tromlach dóibh gur saghas Béarla é: bíonn iontas oraibh nach bhfuil.
Ubuntu has abandoned inittab in favour of their homegrown upstart, not in itself a bad thing. inittab was cryptic, nobody understood it, and everyone who ever edited their inittab did so with fear and trepidation, but kernel boot options should override defaults, not the other way around.
The “sane runlevels” solution presented herein actually works and has been tested, unlike a lot of what I read, and tried, in “help” forums.
]]> SolutionDON'T DO THIS UNLESS YOU KNOW WHAT YOU'RE DOING AND HAVE ANOTHER WAY OF BOOTING YOUR BOX. DON'T BLAME ME IF IT GOES WRONG: IF IT BREAKS, YOU GET TO KEEP BOTH PARTS.
I'm assuming that you already know what runlevels are and how to configure them on something that has ordinary inittab behaviour, like RedHat, CentOS or, it seems, any non-Ubuntu distro.
With that assumption, the solution is easy, you just do the following:
Stash a copy of /etc/event.d/rc-default away somewhere and replace it with my version of rc-default.
The version presented here scans the kernel command line for a numerical runlevel or "S" after the word init, not just for an instruction to enter single-user mode like the Ubuntu version.
In the Ubuntu shipping version, unfortunately, the presence of an inittab file (only one line of which counts) actually overrides anything on the kernel command-line, which is almost certainly not a good idea.
Edit /boot/grub/menu.lst in your favourite editor.
If you're using the Ubuntu way of doing things, just add init 5 to the end of the # defoptions line, so that your default runlevel becomes 5:
# defoptions=quiet splash init 5
Now, find the line altoptions line that configures single-user mode (there's only one by default), change it to say init S (capital letter “S” for sierra), rather than single, and add a new line for text mode, like this:
# altoptions=(text mode) init 3 # altoptions=(recovery mode) init S
Now, make sure to run update-grub as root, so that the grub configuration really changes (the altoptions lines are comments ignored by grub, but are recognised by update-grub which updates the non-comment lines that grub actually uses).
Now is a good time to test. Reboot the machine: it will go into GUI mode as usual. Type runlevel at the shell-prompt: it should tell you that the runlevel is 5, not 2 (Ubuntu's default).
Now, reboot again. This time, hit Esc during the grub countdown at boot-time, and pick “text mode”. It will still boot into the GUI, but runlevel should tell you that you're now in runlevel 3, not 2 or 5. Remember, all the runlevels still behave the same because you haven't changed the symlinks in /etc/rcX.d yet.
Install sysv-rc-conf (with sudo apt-get install sysv-rc-conf). It's a natty little tool which alleviates the tedium of manually renaming the symlinks in the /etc/rcX.d directories with an intuitive curses-based interface. Run it, and you can disable all the GUI-related stuff in runlevel 3. You can configure runlevels 2 and 4 any way you want, but don't touch 0, 1, S, or 6 (unless you really know what you're doing!)
In text mode, the gettys (terminal emulators on each virtual terminal) are started pretty early. So, some startup console log messages come after the login prompt. This might give the impression that the boot procedure has hung at the last log message (from rc.local), but it hasn't: you just have to hit Enter and the login prompt will appear.
In fact, if you watch the log messages streaming down the console carefully, you'll see the login prompt being presented and scrolling off the top of the screen.
From a virtual terminal, you can start a GUI session under the current user account with:
$ startx
Alternatively, you can get a graphical login by starting GDM
$ sudo /etc/init.d/gdm start]]>
Until Steve Ketchem sent me a screenshot showing a '631 running in a SC420 (I blurred details that named the machine, etc., or that I thought might otherwise leak information), it seemed very, very unlikely indeed: there is only one report online of a SC420 being upgraded with a '631 and the preponderance of evidence suggested otherwise. It appeared that the Intel® E7221 northbridge used in the SC420 simply doesn't support the Cedar Mill processors, making BIOS support a moot point, but Steve's email complicates the picture.
]]> Why Would I Want To?Simply: value for money and not needing a heatsink/fan assembly the size of Texas.
The LGA775-packaged Intel Pentium 4 family can be divided into those with the 65nm Cedar Mill core, which include all of the 6x1 (e.g. 631, 641, 651, and 661) sequence; those with 90nm Prescott core, including the 5xx sequence (e.g. 521, 522, 524, 531, 541, and 551); and those with the 90nm Prescott 2M core, which includes all of 6x0 sequence (e.g. 630, 640, 650, 660, and 670). The ancient Williamette and Northwood cores predate LGA775 and Intel's current 3-digit numbering scheme.
The problem is that the 90nm Prescott cores gobbled power and, infamously, ran as hot as the hob of hell. They were sold at a time when Intel were having their asses handed to them by AMD™ and the only thing Intel could do, in advance of releasing the 65nm dual-core processors, was ramp up the clock-speed on their aging NetBurst™ processor architecture and put giant cooling assemblies on top. The TDP (thermal design power) of Prescott cores is 115W compared to 86W for a comparable Cedar Mill core, representing at least one third more energy having to be dissipated as heat. If you think 29W doesn't sound like much, try gripping a 25W lightbulb.
Now that both Prescott cores are obsolete, the Cedar Mill processors are, cycle for cycle and cache for cache, much better value for money, use less energy, and need less cooling. Only a fool would use either Prescott when he could use Cedar Mill.
The Dell PowerEdge SC420 stands accused of failure to support Cedar Mill processors.
The SC420 uses the Intel E7221 northbridge, according to the published specifications (PDF). This is confirmed by CPU-Z which reports a chip ID of 2588 consistent with the E7221.
Both the Intel E7221 northbridge, and the Dell SC420 which uses it, were developed long before the advent of the 65nm Cedar Mill core. The E7221 definitely supports the Prescott and Prescott 2M 90nm cores, since they are listed on Intel's Supported Processors page for their SE7221BK1 server platform, which uses the E7221 northbridge.
The list of supported processors for the Intel SE7221BK1-E motherboard does not include any 65nm/6x1 series processors.
The online product information for the SuperMicro P8SC8 and its siblings, the P8SCi and P8SCT motherboards, also based on the same chipset, explicitly state: “Supports an Intel® Pentium® 4 processor 6x0 or 5x0 sequence on 90nm process ONLY...” (their capitals).
Although the PowerEdge forums contain conflicting and confusing suggestions and opinions, there is not one single report of a working 6x1 processor. On the contrary there are at least two posts which report failure: this post and this one in which the poster also says that multiple similar failures are reported on a Japanese SC420 site.
Dell says that you can upgrade the processor in the SC420 and stipulates the processor requirements as “Intel® Pentium® 4 processor with a minimum clock speed of at least 2.8 GHz, and front-side bus speed of at least 800 MHz and at least 1 MB of internal cache”. The Cedar Mill processors meet these requirements.
Dell released BIOS Update A02 for the SC420 on the 13th of February 2006, which “Added support for newer CPUs”, this was after the release of the Cedar Mill Pentium 4 on the 3rd of January.
The customer reviews of the 631 processor on newegg.com include one which states “Workes great in my Dell Poweredge sc420”.
The quotes from the Dell literature are accurate and were correct when it was written. They did not anticipate Intel developing and marketing a processor as a “Pentium 4” when it was not compatible with the E7221. How could they? You cannot expect that either Dell or Intel would plaster this incompatibility all over the place: their silence is deafening.
Yes, Dell did indeed produce that BIOS update and it was, indeed, after the release of Cedar Mill. However, it was also after the release of the Prescott 2M cores which are supported. They say “added support for newer CPUs”, not “added support for all of the latest 65nm CPUs”, now that would be convincing!
Now, let's examine that customer review again. It says “Workes (sic) great in my Dell Poweredge (sic) sc420 (sic)”. Could someone who made all those spelling and capitalisation errors not have typed 420 where he meant 430 ? Or, perhaps, posted a review to the 631 thread when he actually bought a 630, just going on the speed and other parts of the description?
In light of Steve's email, I might try a SL9KG Cedar Mill processor (Steve's screenshot shows a 631 with D0 stepping, which can only be the SL9KG) in one of my four SC420s, but it's still pretty clear that you can't sling in any Cedar Mill processor and expect it to work.
The best “definitely” SC420 compatible processor, as far as I can tell, is one from the a 6x0 sequence. The 6x0s have two advantages over the 5xx sequences: having the “enhanced Intel SpeedStep technology”, so the processor can be clocked down when under lighter load; and having 2MiB of L2 cache.
If you have used a Cedar Mill processor in an SC420 or any other board with the E7221 northbridge, please comment below.
]]>The previous non-Catalyst version, 8.42.3, solved a long-standing problem with the ATI proprietary drivers for the first time: not being able to run Compiz or Beryl easily because of lack of support for compositing.
The driver packages, for all versions, allow you to build .deb packages for a variety of distributions, including Ubuntu/gutsy, like this:
$ sh ./ati-driver-installer-8.42.3-x86.x86_64.run --buildpkg Ubuntu/gutsy
Unfortunately, on x86_64 (aka amd64), building the 8.42.3 package failed with an error message about missing files in X11R6/lib:
# amd64 needs some library redirection
[snip]
dh_install -pxorg-driver-fglrx "usr/X11R6/lib/modules/dri" "usr/lib32" cp: cannot stat `./usr/X11R6/lib/modules/dri': No such file or directory dh_install: command returned error code 256 make: *** [binary] Error 1 Removing temporary directory: fglrx-install.E10832
The problem was that ATI's build script looked for some 32-bit libraries, common to both the 64bit and 32bit drivers, in the wrong place.
The solution is very easy, you just do the following:
Assuming that you've downloaded the package into the “current” directory, all you do is:
$ bash ati-driver-installer-8.42.3-x86.x86_64.run --extract ati-driver $ cd ati-driver $ cp -r arch/x86/usr/X11R6/lib arch/x86_64/usr/X11R6/ $ ./packages/Ubuntu/ati-packager.sh --buildpkg gutsy
The packages should build correctly.
After doing this, I was able to install the packages and follow this guide to get Compiz working on Ubuntu/gutsy on an Athlon64 with an ATI X1550.
]]>Ubuntu has abandoned inittab in favour of their homegrown upstart, not in itself a bad thing, but as shipped it is not configured in such a way that some very useful behaviour of inittab is replicated.
]]> The configuration files for upstart are in /etc/events.d. The most important of these is rc-default which determines what runlevel is entered at boot-time. Here's the version that ships with Ubuntu/gutsy with line numbers added for reference.
01|# rc - runlevel compatibility
02|#
03|# This task guesses what the "default runlevel" should be and starts the
04|# appropriate script.
05|
06|start on stopped rcS
07|
08|script
09| runlevel --reboot || true
10|
11| if grep -q -w -- "-s\|single\|S" /proc/cmdline; then
12| telinit S
13| elif [ -r /etc/inittab ]; then
14| RL="$(sed -n -e "/^id:[0-9]*:initdefault:/{s/^id://;s/:.*//;p}" \
--| /etc/inittab || true)"
15| if [ -n "$RL" ]; then
16| telinit $RL
17| else
18| telinit 2
19| fi
20| else
21| telinit 2
22| fi
23|end script
Now, the content of this file means that if single-mode is specified on the kernel command line with -s, single, or S, then the system goes into single-user mode [11-12].
Otherwise, if a runlevel is specified in the usual way in /etc/inittab, then that runlevel is used [13-16]. only the line that sets the default runlevel is important, the rest of the inittab is ignored.
If neither of the above work, the runlevel is 2. [17-22]
Line [09] just records the boot time in utmp, it does not affect what runlevel is chosen.
In summary, the choices are (in priority order):
What this means is that there is no way whatsoever of choosing amongst alternative runlevels at boot-time. Even by editing the kernel command-line at the boot prompt, your choices are single-user mode or whatever has been “hard-wired” into the configuration files.
Furthermore, runlevels 2 to 5 are identical, and all of the tools that ship with Ubuntu by default, change all of them.
Well the “graphics mode or bust” model is broken. I get to see the break because I use my desktop in a number of different ways:
In addition, I use proprietary drivers for my graphics card. Every time I update the drivers, it's advised that I remove the previous version, and I have to recompile and install the kernel module interface. It's much cleaner to do all of this in text-only mode, and very convenient for the network to be operational while I'm doing it so I can fetch any new packages that I might need. It's also a very good idea to be able to fall back to text-mode with networking if the new driver doesn't work.
In short, I want to be able to choose a text-only runlevel at boot-time that is not the crippled single-user mode, and it takes far too much effort to get that working under Ubuntu.
There are far too many flies in that ointment!
In 11 years of using Linux, I have never known telinit to switch runlevels cleanly and without something whacky happening. The notion that you can seamlessly switch runlevels on-the-fly by typing telinit X is only proposed by people who've never actually done it.
Secondly, many graphics cards have problems switching from graphics-mode to text-mode.
Thirdly, when booting for the first time with a new graphics driver, you don't want to be inexorably committed to graphics mode.
]]>