Petter Häggholm’s blog

I’m feeling oddly adrift in my Linux taste, these days. I wasn’t a Linux evangelist before, mind, but I would like to be able to answer with a recommendation if somebody asked me what distribution I think they should run. For a long time I would have confidently replied Ubuntu!, but right now I should be unable to do so unless their æsthetic sense differed radically from mine, for starters. I started this year an Ubuntu user. Right now I’m a slightly disgruntled Fedora user in search of something better.

Ubuntu

My experience with Ubuntu started bright and turned better with the years. It was always intended to be easy, friendly, and ready out of the box. There was a time when I was too attached to the tweaking of my Gentoo days to appreciate it, but once I started worrying about two or three computers rather than just one desktop, Gentoo felt like too much work, and Ubuntu’s satisfactory out-of-the-box experience was a relief. Installing it is a snap: Always works, never causes trouble. Upgrades are smooth. Release updates were a bit of a jar from the rolling schedule of Gentoo, but they always went without a hitch, or at most very minor hitches. (Except when I chose to upgrade to beta versions, but if I choose a beta version I know I’m inviting potential trouble!)

The problem is that while Ubuntu has an exquisitely engineered distribution, what it actually distributes is less satisfying to me of late. In part I get annoyed by the tension between Ubuntu and the FOSS community—all the controversies over contributor agreements, playing poorly with upstream, demanding that other projects adhere to their schedules, and apparently picking their software stack based on political desire for control:

Since both init and Xorg are flexible enough to provide the sorts of improvements that Shuttleworth advocates, the suspicion is that such decisions are not technical, so much as political. That is, what concerns Ubuntu/ Canonical is not the technical merits of the applications, but its ability to dominate the projects that dominate its software stack.

The launch of Ubuntu One sort of cemented my generally suspicious attitude toward Canonical. Still, while I might not be wholeheartedly enthused by the company, the product still seemed good.

Until they launched Unity with no good fallback or alternative and in a fit of anger and disgust I left Ubuntu behind. People can claim all they like that it’s similar to GNOME 3. To me GNOME 3 is different than its predecessor, but looks sleek and polished and looks good on high resolution monitors. Unity expressly comes from a netbook project and a harebrained attempt to shove multimonitor, widescreen setups into a low-res netbook mold. Additionally, it looks like the OS X dock interface (which I heartily dislike), but redesigned and styled by ignoramuses armed with crayons rather than the expert UI designers at Apple.

Gentoo

For a brief while I played around with Gentoo again. I like it. I genuinely enjoy the fiddling I have to do to get a Gentoo system up and running, and I really don’t think there’s so much of it that it’s a serious burden. The emotional appeal of a system that I have customised is great; it’s the comfort of a carpenter whose tools have worn down by pressure and friction to fit his hand alone—I don’t pretend that my managing CFLAGS measurably helps performance for most applications, and even USE flags, though definitely useful, don’t affect me that much. But it’s comfortable and pleasing, as someone who cares about his tools. It’s also pleasantly familiar, as the distro on which I cut my teeth as a regular and moderately competent Linux user.

The problem is chiefly just that while I’m happy, nay, delighted to manage a Gentoo system, I’m not half as happy to manage three of them, and between work desktop, home desktop, and laptop, I would be. That’s too much repetitive work; too much time.

Minor problems include never quite being entirely satisfied either with stable (which is too far behind!) or unstable (which, though rarely, sometimes means a bunch of fudging and masking and version-specific flag management); and at the time when I last tried it, the fact that I was really kind of curious about GNOME 3 and Gentoo had no reasonable way of checking it out—it was faster to try Fedora.

So right now I’m not using Gentoo, but as always when I’m not using Gentoo, I sort of wish I were.

Fedora

My experience with Fedora is mixed. Once I get a Fedora system setup and running, I have no complaints. It’s solid and stable and easy to manage and keep updated, as I expect from a Linux system. They stay up to date with software versions and follow upstream rather than going off on silly, Ubuntu-esque digressions, both of which I appreciate. Running it, then, is a pleasure.

But setting up Fedora is another matter. I’ve done it a few times this year, and while it’s fine when it just works, it—wait, no, I don’t know what that’s like. I actually think setting up Gentoo is more straightforward: It’s a lot of work, but it bloody well works the way the guide tells you it will. Fedora is simple in theory, but never seems to work out of the box.

This is what I’m currently running because frustrating as setup can be, I only have to do it every six months or so at the most, and in between it’s pretty much sunshine. But ye gods! are those intermittent periods ever exasperating! Installing a release version of a distribution should not be this error prone, and the upgrades? Disgraceful.

It’s kind of the antithesis of my view of Ubuntu right now, really. If I could run an Ubuntu installer and end up with a Fedora setup, then I’d be happy. That’s not what happens, though. Instead, what I get when I try to install Fedora (I say try, but there is eventual success), is a series of tales of woe I place behind a cut for your comfort.

Now, I’ll freely acknowledge that I’m a bit of an early adopter if I upgrade to the newest version the same week that it’s released, but I find all this very disappointing in what is supposed to be a release version, having gone through formal alpha, beta, and RC stages, with the final release even pushed back (I think twice) to resolve blockers. This load of issues, on three separate systems, is the result? I’m typing this up on a computer that is now finally running a perfectly beautiful GNOME 3 on Fedora 16, but it really shouldn’t take this much drama to get here. I’d excuse it if I were running Gentoo/unstable (excuse it, but be surprised to find it—the quality of Gentoo’s unstable branch would have to go downhill for that to happen).

Still, having tried the two biggest distributions (Ubuntu and Fedora) and found each wanting in its own way, I’m not sure where to turn next. Linux Mint? It is Ubuntu-based but seems less willfull and control-freaky, and the next version (due any day now) will ship with GNOME 3. Maybe that’s worth a try. openSUSE? Something else entirely?

Do any of you guys know much about Apache and mod_rewrite? I could use some help.

Update: Chutz asked me the rather obvious question, had I tried turning off all the RewriteConds? The rather sad answer is that no, I’d missed that obvious debugging step. When I did, the RewriteRules worked… With a bit of help from a very high log level, it turned out that while mod_rewrite applies the RewriteBase to the URI (here, truncating the directory) when applying the rule, it does not apply the RewriteBase when matching against a RewriteCond. Thus, the solution is to write my rules as below, but to insert the directory name—the same directory name as the RewriteBase!—in the matching rules, e.g. ^/newsite/\w+.

I’m playing around with some stuff (on my local box, so far, though I’ll be replicating it at Webfaction…if I can get the damned thing to work) with a dynamic website that uses mod_rewrite to take extensionless URIs and turn them into script invocations (mod_wsgi, as it happens, moving away from the largely-deprecated mod_python). This works beautifully when I only have one site. Now, however, I want to have two sites in different <Directory> sections in the same <VirtualHost>, and things aren’t working so smoothly. In fact, as soon as I change my DocumentRoot to something other than the path of the <Directory> the RewriteRules seem to stop working, even without adding a second <Directory> section.

All I get for every request in the mod_rewrite log is a notification that it passed through:

(In the Apache error.log, of course, I get the expected error messages about requests for resources that can’t be found.) I’ve tried to add an appropriate RewriteBase, but so far to no avail. My current setup looks like this, and doesn’t work:

Listen 80
LogLevel info
LoadModule wsgi_module /usr/lib64/apache2/modules/mod_wsgi.so

WSGIPythonPath /home/petter/projects/newsite:/home/petter/projects

NameVirtualHost 127.0.0.1:80
<VirtualHost 127.0.0.1:80>
	ServerAdmin webmaster@localhost
	RewriteLog /tmp/rewrite.log
	RewriteLogLevel 2
	
	DocumentRoot /var/www/localhost/htdocs/wsgi
	<Directory "/var/www/localhost/htdocs/wsgi/newsite">
	        Options Indexes FollowSymLinks ExecCGI

		AddHandler wsgi-script .wsgi

		Order allow,deny
		allow from all

		RewriteEngine On
                RewriteBase /newsite

                # Really, really annoying; the trailing slash fixes don't seem
                # to work on the server's document root...
                RewriteCond %{REQUEST_URI} ^$
                RewriteRule ^.*$ test.wsgi?page=index [QSA]

                # Redirect .py files
                RewriteCond %{REQUEST_URI} ^\w+\.py$
                RewriteRule ^(\w+)\.py$ test.wsgi?page=$1 [QSA]

                # Redirect extensionless URLs, unless they're for directories
                RewriteCond %{REQUEST_URI} ^\w+$
                RewriteCond %{REQUEST_FILENAME} !-d
                RewriteRule ^(\w+)$ test.wsgi?page=$1 [QSA]

                <Files *.xml>
                    Order Deny,Allow
                    Deny from All
                </Files>
	</Directory>
</VirtualHost>

1. Find someone to borrow a screwdriver from.
2. Open computer.
3. Notice that harddrive blocks long PCI-E card.
4. Move harddrive.
5. Physically install PCI-E card.
6. Boot computer, note kernel segfault, wonder if it was caused by yesterday's update or the new card.
7. Reboot with the last kernel version, note kernel segfault, conclude it is indeed the card.
8. Realise that onboard video is still enabled.
9. Reboot and disable integrated video; boot and note kernel segfault.
10. Check BIOS and discover that if you set onboard video to disabled, the setup will ignore you. Briefly curse ASUS.
11. Remove PCI-E card, boot, and search for solutions.
12. Discover that the problem is that incompatible kernel modules are loaded (intel_agp and the NVidia module for the PCI-E card).
13. Briefly curse Ubuntu for providing no easy kernel module management.
14. Figure out that you can force udev to refuse a module, even when autodetected, by blacklisting it in /etc/modprobe.d.
15. …
16. Profit!

So. I'm not entirely happy with the aging hardware of Odin over at cs.ubishops.ca, and not only is money to replace and maintain it hard to come by, but let's face it, I graduated from Bishop's over a year and a half ago—eventually it seems proper to cut some of the ties. I have my own little server sitting in a closet in the form of castleblack, my old Athlon XP desktop, but desktop hardware in a closet is not the ultimate in reliability. Given how dependent I am on these things, I really ought to invest in more.

There are a few problems, though. One is, ironically, that there's so bloody much choice! Monopoly is bad, monoculture is bad, but search for webhosting and you'll find thousands of webhosts. Search for search tools to help you find webhosting, and…well, I'm just lost, dazed, and generally confused. I don't know where to start looking to find the features I need at the prices I want. And that's not even getting into what hosts are actually reliable and trustworthy!

Problem number two is that my needs are rather specialised. I need simple webhosting, yes; with PHP, SQL (preferably Postgres but I'm not picky), and CGI (at least Python). This is pretty basic. I also need email with IMAP access. Not too hard, either. However, what I'd really like to do is to replace not only my Odin account, but castleblack as well. Now, I use castleblack for a few things. I keep an IMAP-enabled mail server on it to centralise my email archives, for one. A host that offers IMAP access will do nicely, as long as they offer enough space—my archives are 327 MB and constantly growing, of course. Something I want to use castleblack for is a contact list in the form of LDAP, but OpenLDAP is, pardon my French, a bitch to configure, and setting up an SSL enabled LDAP server and getting it to work on the client side…well, no success so far. So I want a webhost that offers

• Standard webserver stuff (HTTP, PHP, CGI)
• Python CGI (not too bad but less common than perl)
• Email with IMAP access
• At least 500 MB of space right now; call it 1000 MB for a safe margin
• Configurable LDAP (give me phpldapadmin and I'm golden, just give me a properly configured openldap setup because I'm clearly incompetent to set one up myself)
• Bonus: A calendar server

Just to further complicate matters, I do use castleblack for one more rather notable thing, which is to host a subversion repository that contains my entire home directory. Although it could do with some cleaning, it presently weighs in at 1.9 GB, so I'd want no less than, say, 3 GB (again, I want plenty of room to grow!) for a grand total of 4 GB of space and subversion on top of the features I have already listed. The space is not so unrealistic—I actually found a host that offers 15,000 MB for only $4 a month or so (but not so many features). The feature list, however, especially LDAP and subversion…I don't know. I'd be all right with hosting the subversion repository on my own, though, if I could offload everything else. It's easier to back up, anyway, and I don't know what the bandwidth requirements are. I should probably look into running a bandwidth analyser on castleblack. I'd also be willing to consider other revision control systems.

I have a feeling this is asking rather a lot. On the other hand, in the enormous deluge of webhosts out there, maybe there is one that offers what I need? I don't know. I tend to lean in the direction of dedicated virtual servers or even colocation just because it's so bloody hard to find what I need, but the truth is, I'd probably be happier if I could find a proper host that could do the configuration for me. Yes, I like playing with computers, but I can do that at home. If I could offload all my important data (svn repository + email) and have someone else vouch for its safety, with regular backups and so forth, that'd give me some peace of mind (and make tinkering optional). Also it might give me LDAP that actually works. I'm not closed to these ideas, but I should consider all the options. Or, rather, at least all the major categories of options…

I don't know what the cost tradeoff would be. castleblack is free in a way since it's just old hardware I had more or less lying around, but honestly, I could probably sell it used even today and get a year's worth of hosting somewhere, and the added safety and reliability would be worth it.

So. Does anyone have any suggestions for me? HELP!

• After many months of waiting and several supposed and alleged deadlines missed, ATI finally released working drivers for Radeon cards on AMD64. It works just fine. Unreal Tournament 2004 runs beautifully, and once the numbskulls at CSN actually pay me my student loans, I'm very likely buying it.


• My printer works with CUPS. No idea what I did wrong before; I don't think I did anything really differently—but it works now.


• My scanner works! There was never a working SANE backend for it, and I had quite given up; I was searching eBay for a scanner that the compatibility list does claim has at least "good" support. Imagine my surprise when I cross-check with the compatibility list only to find that a newly written driver has not "basic", not "good", but "complete" support for the scanner I already own! And of course, this is how things should work. Knowing that I want the backend and frontend portions from the SANE project, I do the following:
  emerge sane-frontends sane-backends
  ... That's all it takes, and just like that, the Gimp picks up my scanner. Who said Linux wasn't user friendly?