Ho hum, installation time again That's it: beige ⌣ I'll paint the ceiling beige… And before I start, for those of you unfortunately unable to get out from under Internet Explorer, Yes, I have added a banner extolling the virtues of Firefox, and No, no-one is paying me to do it (Google either can't or won't pay its $1 finder's fee to AdSense customers outside the USA). For those of you using normal browsers, you simply won't see anything different. So I decided to upgrade Z again, to OpenZaurus 3.5.4 RC (see sidepanel on left). As usual, the folks at OpenZaurus and Opie have done an excellent job against all the odds to get this out, and there are just a few minor wrinkles to iron out. But I said I'd log the process, as I did with FC4 a while ago, so here she is, a log of installing 3.5.4RC (58-6 size) to an SL-5500. Bugs, suggestions, etc are flagged. Under 3.5.3: download files, rename, copy to CF card. Remove SD card, flash, remove CF card, boot. Boot says something about not being able to synthesize pci hotplug events, but otherwise it boots and installs packages as expected. There still appears to be no way to grab a log or copy of this procedure, and it scrolls off the screen too fast to make notes. Is there any way the boot/install log can be tee'd into a file? Initialization looks fine. Time zone defaults to UTC which is good. My network card isn't recognized by pcmcia/config, so it has to be a manual setting first time round. Get time from network should be grayed-out when no network card is present. Switched Tokyo to Dublin, set time, changed Week Start from Monday to Sunday, changed NTP server to a local one. I've never understood what Predicted Time is: it doesn't seem to do anything meaningful except predict the wrong time ⌣ Usual long pause before Personal Details screen appears. The script is obviously doing something, but the delay would unnerve a novice. Add a Loading… popup or progress bar so users don't get worried. Skipped Personal Details because my businesscard.vcf is on my SD card and I'll copy it across later. Opie starts up. Nice font, big improvement. Tabs are still a bit clumsy by default though: not a serious problem, just æsthetics: they need to be coloured and round-cornered a bit to add to the eye-candy. No mail client, no FTP client, no browser. No problem for initiates; big, Big, BIG problem for newcomers. Zaurus is dismissed as useless unless it fires up Internet-ready first time, every time. We seem to be missing a big trick here. Yes I know there are problems, konq is unmaintained and dies on many sites, and opie-ftp fails to store site setups, but opie-mail works OK most of the time and it reads news. Add opie-mail and konqueror or opera at least, and add mail config to the startup script. Either fix konqueror (it's very close) or cut a deal with opera. Fix the dropping of site data from opie-ftp. Fix bug in opie-mail IMAP where it fails to delete mail from server. Inserted SD card. No popup to say it's inserted (although at boot time I noticed that sdmgr itself did start up). Check the applet and it says Empty Socket. Open File Manager and find that SD card actually is there. Fix sdmgr to notify tray applet that SD card is inserted. This has always worked until now, so maybe it's been broken by mistake. There was surely no need to break or disable it. Open a console, go to /media/card and run my script that copies my backups of addressbook.xml, datebook.xml, businesscard.vcf, todolist.xml, hosts, resolv.conf, and config.opts files from the SD card into the relevant places in /home/root/Applications or /etc. The script also HUPs cardmgr to enable CF cards. Re-insert the 64Mb CF card used for flashing, and it pops up a window saying it's not recognised and do I want to configure it. This is new: a really excellent idea, and it displays the recognition string from the card, which is otherwise impossible to discover. But DUUH this is the card you just flashed the system from, right? So you know you can read it already. It should have been added to config.opts automatically at load time. Ensure CF card used for flash gets added to pcmcia/config.opts if it's not already in pcmcia/config. And it's ignoring my pcmcia/config.opts which already identifies all my CF cards. Fix cardmgr so it behaves normally for cards found in pcmcia/config and pcmcia/config.opts and doesn't ask for them to be configured. And where FFS does the new pop-up card detector store the data it gathers? Certainly not in config.opts… Configured the card for activate on insertion and on boot. It's a Really Bad Idea to make it default to suspend: WTF wants card insertion to cause a Suspend? Fix card detector to default to Activate. Weirder and weirder. Having done the CF config, the applet now does identify my CF card with the very string I already had for it in my config.opts. So it seems to read the file, but ignores it when a card is inserted and asks for configuration anyway. Not good. Eject and reinsert to test…nope, it still asks for it to be configured all over again. Configure it and the applet displays the right string again, which it knew already…so why ask? Is there a problem with the config applet writing to the config.opts file? It's going to be a major problem if users have to configure CF cards afresh each time they insert them. Good menu for the card recognition applet though. Definitely the way to go, and a major improvement. Inserted Ethernet CF card, which is wired to my DSL router. Applet asks for configuration, same questions, same reply (activate), plus it wants the driver (pcnet_cs). Same error: the applet then displays the real name I had already stored in config.opts. OK, at least it's consistent. Open Network, yep, it grabbed an IP address from the DHCP server. Looks like we're online. Configure Light & Power, looks good. Configure Appearance, fix rotation to anticlockwise. Check Calendar, Contacts, ToDo List, and Today, all working with my copied data OK. Open console and set root password. ssh into Z from a Konsole on my laptop. Time to zap all trace of my previous 3.5.3 packages, even though 3.5.4 uses the same repo for them, as they had gotten messed up. Space is tight on an SL5500, so I've been keeping all the package-related stuff on the SD card as the root partition is at 96%. Remove the old /media/card/packages (actually move it out of the way until I understand/remember what creates it and how ⌣ Try an ipkg update. Oops. It can't download 3.5.3-snapshot-20051024/feed/opie/Packages.gz from http://openzaurus.org/official/unstable/ which is hardly surprising, looking at the name. Looks like it has tried to autoconfigure itself using the kernel label or something. Check /etc/ipkg.conf and sure enough there is the faulty URI. Use vi (gag, spit, wash mouth out with soap and water, no emacs on the Z) to get rid of the snapshot and date. Now ipkg update works. Fix /etc/ipkg.conf to use the right repo directory names. Try ipkg -d sd install opie-mail (assuming the GUI package manager is still unusable). Started off well, downloaded the IPKs, including the lib deps, but then hung on installing libcrypto first…maybe because there's no /media/card/packages to put them into? Hmm. No flicker from the hub lights, maybe nothing is happening. Ctrl-C is your friend. Damn, it wasn't that, it was just hung downloading. Note: the wget REALLY needs to display a progress indicator while downloading in ipkg. Added libcrypto separately via another ipkg install. Make wget in ipkg display progress. Restarted Opie, but no sign of an icon for mail. Bummer, I hate having to make .desktop files by hand. Same bug as always: ipkg fails to run ipkg-link add after installing. Do it by hand. Make ipkg produce a usable icon for installed binaries. Restart Opie and there's the mail icon…but Grrr same bug as in 3.5.3 and others: Could not locate application opiemail. This is getting tedious. An installer should not lose track of the binary it is installing. Running which opiemail says it's /opt/QtPalmtop/bin/opiemail, which is a symlink to /media/card/packages/opt/QtPalmtop/bin/opiemail, which is a symlink to quicklauncher. So presumably quicklauncher isn't smart enough to do a recursive basename on the lib deduced from the value of $0, and gags on the reference. Or something. Either way, both opiemail and libopiemail.so are linked in, but there ain't no way they'll run. I don't remember how I fixed this before, but it involved the wholesale moving around and softlinking of directories for days to get it set up, which ipkg-link is not doing correctly. Overall: Some big and much needed improvements, especially in the flash and install: congratulations and thanks to everyone who has worked on this. Bugs so far are small and should be relatively simple to fix. The card insertion popup will be fine when it works. There's clearly something radically broken in the way ipkg fails to leave the user with a working icon, though, which is a bigger problem (although probably a simple fix). And not including a mailer, ftp client, and a browser in the default distro may be a political problem which needs to be sorted out. Pride and Printability Jane Austen never had this problem User calls the HelpDesk: We've just been trying to print some Web pages and the text is running off the right-hand edge so the ends of all the lines are missing. HelpDesk: Uhuh… [long and involved explanation leading to an increasingly irate user and an increasingly frustrated HelpDesk]. Sounds familiar? If you're using Microsoft Word to prepare pages for the Web it should, and it should also make you want to scream. The problem is a fundamental flaw in the behaviour of Word so crass that it makes you gasp. Embarrassingly, it's at its worst with Microsoft's own Internet Explorer, although switching browser won't make it go away. It happens when you try to turn a Word document into a Web page by using the Save As…HTML menu item. Word is a business/office wordprocessor designed for creating printed pages, nothing more. Using it for preparing anything else only ends in tears. Microsoft's conversion policy to HTML is to preserve as closely as possible the exact format of the printed page, even down to the length of lines and line-breaks, even if that means making it unprintable, even in Microsoft's own browser. To that end, Save As…HTML deliberately and perversely goes out of its way to ignore the reformattable nature of paragraphs and list items, forcing them to the exact line-length of the original paper version and ignoring the auto-adjusting nature of web browsers which are designed specifically to fit the text to the width of the screen or paper. The reasoning is simple: Microsoft Word Marketing were apparently terrified to the point of insanity that users who had written documents in Word and then saved them as HTML for the Web would find that their documents had been reformatted to fit the screen, and complain. And another part of the same empire was insistent that exported documents should be re-editable in Word, so truckloads of superfluous markup get added to your document in order to make sure you can open the HTML file as if it were a .doc file and never notice the difference. List items are particularly subject to this kind of mangling because Word treats them as paragraphs with a number or letter at the start, so they fail to use the built-in hanging indentation that the default Web browser rendering provides. Your Web page also ends up with huge amounts of unnecessary vertical white-space because of Word's galvanic frog-leg knee-jerk use of empty paragraphs to create space between paragraphs—and of course the default in a browser is that paragraphs get a blank line between them anyway; but these so-called empty paragraphs end up with a non-breaking space, which forces them into a zombie existence, like dead men shouldering the coffin of your text. So what's the solution? Using Save As…XML doesn't help—it just formalises the error, rather like printing an scaled-up bitmap on a hi-res printer simply gives you very beautifully-rendered jagged edges. Using Named Styles from a template in your Word document won't help either: it ignores List Item and simply outputs a paragraph stuffed to the gills with attributes we never added to HTML, bloating the file with more redundant markup to add the the garbage already inserted by the desperate attempt to preserve round-trip editability mentioned above. You can just see the spastic twitch of the Marketing suits on being told you can round-trip files between Word and HTML, blindly unaware that Microsoft had an excellent product a decade ago which I posted about before, called SGML Author for Word which did the exact same thing using Named Styles. The moment XML arrived, and with it a whole new market, they killed it off, perhaps because because it threatened the dominance of Word. And this is really the core of the problem. This kind of apotheosis of the visual constitutes part of the textual hegemony which Microsoft have cleverly imposed on the world's organisations and which is going to cost them dear in the long run. This ethos has successfully conned people into believing that because it looks pretty, it must therefore also be right, which a moment's reflection will show you is wrong. The next version of Word has all kinds of bells and whistles including a floating ribbon menu, and it makes XML the default save format, relegating .doc files to the obscurity proper to an unloved proprietary format that has overstayed its welcome, like an elderly incontinent aunt quietly being tidied back upstairs into bed. But the next version of Word will undoubtedly also include a way to switch back on the classic behaviour, to satisfy the unretrainable, much like every Fortran compiler includes an -f66 option to make it compile DO-loops like God intended. So what's the solution? Go to Tools|Options and make the Style Width 2cm wide so the dog can see the rabbit. Go to each so-called Normal paragraph and use Named Styles from a template to relabel things properly. The default styles in the Style drop-down are fine provided you click Show All Styles, not Styles in Use. Save the file and then open it in OpenOffice, AbiWord, or pretty much any four-wheeled wordprocessor, and Save As…HTML. Or at least, that ought to be the solution. It certainly fixes the unhinged list items, but almost all the wordprocessors still insist on keeping empty paragraphs and still try to enforce the line-length. This is what it is like to be hegemonised—no wonder SlashDot's icon for Microsoft's activities shows Bill Gates as a semi-assimilated Borg. What's needed is two modes: an anal-retentive mode which performs as now, to preserve the appearance beyond the call of duty, and ensure that you can re-edit the file in Word or whatever ad nauseam; and an unclenched mode which goes out of its way to produce clean, flexible, dynamic HTML which will slip silently into any Web site and perform just like it says on the tin. Another possibility is that the printer margins are too wide to allow the page to fit in the remaining printable width. This happens when the stylesheet tries to enforce a page width (which designers love to do but which is a really silly idea). It's always worth reducing the left and right margins to the barest minimum (usually 3-4mm) just to see if it works; and then increasing them in steps until you find the optimal setting. Or calculate the required screen width from the CSS in pixels, then either divide by 96 to get inches and subtract from 8in for a Letter sheet, or divide by 3.78 to get millimeters and subtract from 210 for an A4 sheet. The result is the total allowable space for LH+RH margins, so allocate it as you see fit. There are of course Word-sanitisers, which claim to clean up the garbage markup inserted in anal-retentive mode, but they don't do all of it. DynaTag of course, if you can still get it (try Red Bridge) does a spectacularly good job of producing XML from Word, but it needs training, and is really only for bulk conversions. For the rest of us mortals, getting out from under the dead hand of proprietary format control is becoming a daily essential and one that Tony Bove is offering a solution to, in his forthcoming book Just Say No to Microsoft. Playing hard to find Digging out highly-specific information isn't always easy An FAQ for users with their first CD burner used to be how do I make an autorun CD? (maybe it still is). The answer was always pretty much along the same lines: Well, it depends on which version of Windows you're using… For reasons best known to Microsoft, there are about a dozen different, mutually incompatible ways to make a CD start a specific program on insertion into a Windows box. An entire cottage industry sprang up to supply autorun software which would test, guess, or bludgeon its way into finding out what would make the CD autorun under any flavour of Windows into which you put it. Meantime, Mac users sat there looking smug, until a widespread viral attack using autostart poisoned the Mac community against software started by hardware actions. Linux users just had to wait. The fuss has long since died down, but I recently had occasion to need to make a CD which would autorun on any of the three platforms, and it proved unbelievably difficult to find any information about how to combine the effects in a single ISO image. This wasn't helped by the fact that—as an infrequent Mac user—I was unaware that support for autostart ended with OS/9, so it simply doesn't exist on OS/X. I was in the fortunate position of not having to support obsolete versions of Windows (3.x/95/98/Me), but I was constrained to using only free software. This turned out to be a benefit, as usual, in that everything needed is indeed available in Open Source form. I've documented what I found in a short web page for the benefit of anyone else stuck for the information. It's restricted to the specific task I had in hand, so it won't solve everyone's problems, but it's been working just fine for a month or so, and given me a good appreciation of the hard work that was put into mkisofs. The $50 Million Giveaway What's changed for venture capitalists in the years since the dotcom bubble? This is fascinating. Slashdot this morning pointed me at a Business 2.0 article on 11 VCs saying what projects they would fund if the right proposal came their way right now. Having done DD for VCs during the dotcom bubble, I thought it would be interesting to compare what they want now with what they used to fund, and see if there's any difference. A great idea, but still impracticable. Text messages are notoriously subject to delay, and authentication must be real-time to succeed. Neither buyers nor sellers will buy into it if it means a long wait at the sales counter until the text message arrives. But crack that one (maybe with IM-style messaging via cellphone) are you're in. Oh, and large chunks of the US still have no GSM signal… Five years ago: What's a text message, Mommy? Not until Hell freezes over will banks accept using mission-critical financial software which has been written by a consortium who also supply their competitors, based on code they themselves originally wrote. Yes, it's partly illogical: they already feel happy placing customer-critical information into proprietary apps which are known to be unstable and even buggy (guess!), but when finance systems come in the door, logic sometimes has to go out of the window. Just think: learn to crack one bank's apps… Five years ago: Let's convert all this legacy COBOL… Improving the accuracy of targeted product recommendations is an excellent idea. But the suggestion to tackle it from the algorithm end is like treating the symptom instead of the disease. What's needed is better quality data and data identification. It's a bit like the Information Retrieval people some years back, doggedly insisting they can make semantic sense of unmarked plaintext corpora by sheer algorithmic power (I nearly got kicked out of an IR meeting in Glasgow some years back for suggesting they'd have better results if they marked the text up in XML first ⌣  Get the data model right, and everything else pretty much drops into place: get it wrong and your project is screwed from the start, no matter how smart your algorithms. Five years ago: Let's start an online clothing store… This one's a winner, but not at $30 a month—in a year you'd have spent enough to buy a cheap PC outright: seniors are not stupid, and they can do the arithmetic just as well as you can. At $5 or $10 you'd hit the spot, though. The usability problems are significant, but not insurmountable, and I'm tempted to stick my neck out and predict that this one will make it, although it will be like most of the other simplified systems aimed at the non-technical user (WebTV, AOL, Windows ME, etc), hopelessly crippled, grotesquely proprietary, and impenetrably hard to support. Five years ago: Let's invent a whole new bunch of tags for HTML… This one would beat the Mobile ID for Credit Card Purchases hands down if they can solve the security problems. Instead of back-channel auth, just do away with the credit card entirely, and go for a dialog-based ID and debit/credit service over the cellphone. But the banks can cripple this one at a stroke by using transaction charges, so get them on-side first. Plus the US banking system is still in the quill-pen era for anything involving foreign currency, so they'd need a nib sticking in their corporate butts before I'd risk travelling abroad with this device. Five years ago: Everyone would be doing all their transactions over the Internet by 2005 anyway…[close but no cigar] Dead in the water on cost. If price was a factor, companies would already be doing this. The cost of software, and even of support for it, simply isn't important: companies are making amply enough to cover it, and are prepared to keep paying to avoid the perceived pain of change, so shelling out to the big boys for proprietary software just doesn't matter—financially. The real IT costs are in hardware and skills: companies will just buy boxes, but they really hate spending on upping their skill residencies. And, uh, most of the apps required already exist, and lots of them are already in use, unapproved and unannounced, having been snuck into your datacentre under the cover of a new server. What'll bite your ass is when you downsize or outsource and lose all the skills you didn't know about that have been keeping you going for the last decade. So on crisis support this one might win. Five years ago: What's Linux, Mommy? The keys to this one are aggregation and bribery. The bribe of a small payment to popular posters is a dead cert, and aggregation software is improving (although what I had to say about algorithms vs data above still applies). The killer is the phrase keep it constantly updated. Unless their is some kind of editorial busybody pruning back the obsolete information into an archive, the system will gag on its own success. If I was running a local newspaper, I'd either be seriously worried or writing my proposal now. Five years ago: Fill up the shortfall with ad revenue, Mike, they'll love it… Cookies via VOIP. No big deal. When enough bandwidth becomes available: right now trying to run Skype and download stuff is a non-starter. Five years ago: Hey crackle, spit this Internet bzzzz, click Phone thang is snap, crackle, pop soooo kewl… This one's out of my field, but it sure sounds like today's networks don't require a separate software layer for every service…it's just been implemented like that because that way the vendors keep control of their IP and preserve their revenue streams. The VC is right, it's something the big infrastructure providers won't do and the wireless carriers and ISPs can't do. If providers want a single platform, there are plenty to choose from, and maybe they should just gang up on their current providers and force a change. Five years ago: Mobile phones? But in America we user *pagers*… This one's a real winner if, as the VC says, you can navigate the health-care systems. As the HL7 experience has shown, even what ought to be obvious technologies and methodologies rapidly get bogged down once they hit the organisational wall. Life-critical device production and database management are well-established areas, but life-critical systems require more than ordinary care on the communciations side, so I'd add messaging expertise to the VC's requirements. Five years ago: What's XML, Mommy? This links two good ideas: technology convergence, where any device in your home can handle any service (doing your email on your toaster being the bathetic instance); and application persistence or pervasive continuity, which is rather different (leaving your email half-written on your desktop, walking out to the car and editing the message on your PDA or cellphone, then arriving home and finishing it off on your Playstation). However, contrary to the suggestion, the second has actually been on the market for several years, but the first is still as elusive as the paperless office. If they can get it to work that would be great, but not at the expense of reinventing the wheel. Five years ago: I saw Mission Control demonstrated… Don't get me wrong, I think it's really good that these guys have given their time and energy to showing us where they think the future lies, and a lot of it corresponds broadly with the way most people I know are thinking. Maybe it also means they aren't getting as many of the truly ludicrous proposals we saw last time around, which may also save some of the pain of having to recommend against a proposal by someone you know well (knowing the principals was pretty much unavoidable in many of the very narrow vertical markets). But prospective proposers still need to remember that VCs budget on losing—what is it?—85% of everything they invest. They make enough back on the 15% to cover the 85%, so saving them from backing a few losers isn't interesting to them: the money's already dead. What VCs really need is a device which is IP-connected, which authenticates via cellphone, can recommend best-of-breed choices, is easy to use for OAPs, speaks IR and RFID, runs Open Source, connects to your local social network, is fully extensible to new services, can do your email, and acts as a truly reliable bullshit detector. A simple upgrade Upgrading an operating system is never exactly simple but it could be a lot worse. I've been running Red Hat 9 for several years, and it's definitely time for an upgrade. RHEL isn't an option on a personal machine; Debian is too much bare-bones work—perfect for the OS hacker, but inaccessible for an end-user; I hear lots of good things about Ubuntu, but I need something of proven solidity. After having used Fedora Core 2 and 3 on servers in the office, I felt FC4 would be the obvious path, and I resolved yesterday not to repeat the same mistakes I made when moving from an old laptop running RH 7.2 to a new one with Windows preinstalled on which I wanted to dual-boot RH9. These were: Miscalculating how much space I'd need to repartition for Linux. I could have left Windows in 5Gb easily, not 10. The question now is, do I even need to keep that Windows partition at all?—I haven't used it for over six months. Believing the error message that said the partition table couldn't be rewritten—Ranish Partition Manager is a wonderful program, and it did the job even though Dell's firmware tried its best to stop me. Not finding out beforehand what the display's horizontal scan and vertical refresh rates were, even though I knew that XFree86 would need them (I've never had a Linux installation correctly pick these values up from the hardware, ever). Not checking the Ethernet card type either—fortunately RH9 detected this OK. Allowing it to install and use the ghastly and appalling CUPS instead of LPRng. LPRng wasn't broken, so there was no need to reinvent this wheel, and CUPS is mind-bendingly hard to configure, and refuses point-blank to print from one RH9 system onto another. Bothering to install Gnome. It's cute but useless. I use KDE, and if it wasn't for the need for the occasional Gnome library in order to run a utility like Gkrellm I wouldn't bother again. Unfortunately there are still too many utilities written to operating system variant-specific APIs (exactly what the Open Source community complains about in respect of Another Operating System). Not learning IPtables from the ground up. I still haven't, so I'll have to rely on whatever default firewall thingy FC4 offers during installation. In this day and age, configuring a personal firewall and turning it on and off should be easy—Windows XP makes the turning on and off easy but fails to provide much configuration. My old Bastille script worked well, but provided no help. CLUG has a project to make a web interface to IPtables, but it's early days yet. So after a false start (one workstation garbled three of the CDs I was burning), it all fired up fine, except that it couldn't recognise my laptop monitor. It's a Dell Inspiron 4150, guys…maybe Dell just never supplied the spec. But otherwise it went well, although there are unnerving long pauses occasionally when the Next button won't click while it thinks, and there's no progress indicator. So first thing to do is configure the network, Firefox, and Thunderbird, and we're in business. Pretty painless, and it did indeed eventually ask me to configure the monitor. Next step is to rip out the FC-supplied , which I took the liberty of assuming would be just as out-of-date as all previous tetex RPMs, and install a fresh copy from the Collection DVD. This takes for ever, as I installed pretty much all of it, but it's well worth it for a robust and up-to-date formatter. The only flaky oddity is that pointer movement from the glidepad over the panel in KDE sometimes causes different screens to flutter past. Picking a screen from the four on display means clicking and waiting a second for it to take. I haven't tried this in Gnome, but if it's a glidepad driver bug then it's going to be annoying. The whole X thang seems slightly slower than earlier versions (x.org vs XFree, I assume, plus the laptop is getting on for two years old now, and doubtless FC4 is bigger than RH9). But overall a success, and I'm suitably impressed. Even CUPS, and that's saying something. Rail travel in the UK Come back Dr Beeching, where are you when you country needs you? In the early days of rail, different companies ran different services over different tracks. This was called competition, although travellers probably called it something else (one is reminded of the Punch cartoon of the stress of Changing at Crewe). Technology came and went, and by WWI, things had been settled down with a little governmental intervention to four big companies running a fairly comprehensive service. When it became expedient after WWII for the government to further nationalise them into British Railways, the downfall of civilisation was predicted. Indeed, the 50s and 60s saw the dereliction of the service, unfunded, grubby, uncared-for, unpopular, and increasingly out of date. With the road transport lobby buying up all the politicians, it wasn't surprising that Beeching acted as he did, and no amount of window-dressing could compensate for the loss that became the British Rail of the 70s and 80s. But wait…there's less: Maggie Thatcher had the coup de grâce to deliver. By selling off the ownership of the track and maintenance separately to the companies that ran the trains, she effectively paralysed the entire system in order to funnel the proceeds into the pockets of her party cronies. The British live with the results today: no synchronisation between timetables, incompetent or amateur management, grossly overcrowded trains, lots of glossy PR, and increasing accident rates because all the skilled engineers who knew how to run a railway left or were pushed out, leaving maintenance in the hands of the untrained. It took me an hour and a half to queue for an advance-purchase ticket from Oxford to Exeter (the normal ticket office only sells full-fare and travel-today tickets). The booking clerk was one of the old school: polite, diligent, and competent, struggling with a 80×40 character-cell interface. She finally found me a fare for only about twice what the same distance costs in any other country, and it restricted me to one specific train each way. When it arrived, the train was packed solid, and my reserved seat was already taken (I let the girl sleep, it looked like she needed it). Quite apart from the shambles over track maintenance, the numbskulls who run the trains (and I am disappointed that that includes you, Richard Branson) clearly don't grasp the attraction and simplicity of the previous system of buying from a small set of well-known tariffs, available for travel on almost any train, and not running competing services over the same pair of rails. You can run competing services in the air (Branson does it excellently), and on the water, and even on the road, where one vehicle can overtake another. But not on the permanent way. Competition in rail traffic is a well-known no-no—even the Americans have realised this, and they're usually better at taking advantage of competition than Europeans are. If you're visiting the UK, take care! School's out Something old and something new from the XML Summerschool, the Extreme Markup conference, and world of LaTeX Suspicions confirmed: no-one has a good word to say for W3C Schemas, I'm afraid. Everyone is prepared to admit they are useful in some circumstances, and essential in data applications when you need strong typing for element content or attribute values, but for most text documents they just get in the way, and everyone hates the syntax. Interesting how the programming community, all with Computer Science degrees, which means they presumably understand the importance of data modelling syntaxes, collectively screamed when XML came out that they couldn't possibly learn or use two syntaxes (declaration and document; and how hard declaration syntax was, poor little mites)—and then went blithely on to implement a schema language in a syntax designed for writing documents. However, they're here to stay, and none of the told-you-so about how much simpler it would have been to add datatypes and cross-checking to Declaration Syntax will stop them. The missed trick, though, was in not pushing for a TC to SGML to allow Document Type Declarations to specify a schema in the same way as they do a DTD. There is no good reason why an additional token of SCHEMA or DTD should not have been introduced after the SYSTEM (and, optionally, PUBLIC) keyword (and possibly a new Public Text Class of SCHEMA for FPIs). The lack of a formal binding of schema to document is an unnecessary hindrance to implementation, precisely the kind of thing that was avoided in other places in the design of XML. However, the discussions on editors are heating up. Almost every application you come across seems to have dissatisfied developers and users because the kind of editing they want to do simply isn't available. There were several presentations on the shortcomings of editor technology at this year's Extreme Markup conference in Montréal—the first time the topic has arisen. On the one hand you have those fluent in markup, who have no problem with plain-text editors and exposed pointy brackets, who have a wide choice of systems to pick from. On the other hand you have their customers, the end-users, for whom the developers demand a Word-like interface because it's believed (wrongly, in fact) to be all they will accept. Oh, and it must be easy to use; it must create 100% error-free documents regardless of user mistakes; and by the way, we're hiring an infinity of untrained monkeys to do the typing, so it must be able to guess at most of the markup…you get the picture. I've been working on this topic for a few years now, and I've started to put together a web site for some of the results. Seán McGrath had some interesting comments to make in Oxford about using tailored DTDs or Schemas for data collection, because the results would be fed into something else anyway, either for pipelining or storage, so expending a lot of effort on meticulous encoding up front is probably wasted—keep that for the repository copy. This is pretty much what we've been doing for several projects in recent years (including the Old Irish Lexicon, the Digital Dineen, and the eDIL), and only moving to TEI when the text is clean and checked, before the in-depth markup starts. But it's still typically done in Emacs… Since PDF arrived, there have been numerous packages for creating presentation slides, most of which make Powerpoint look very sick (not hard, I hear you say). I've been using pdfslide for some years, but I'm about to move to Beamer, having seen what it can do. Download their presentation on the Weak Cardinality Theorem (an example of a real-life talk in their Examples section) if you want to see what the future of screen teaching is about. Pick a card, any card Making the bold step of choosing a scripting language Apart from the wars between editors (Emacs vs vi, etc) probably nothing else raises a techie's hackles like a good argument over the choice of programming language. It gives you a great opportunity to extol the virtues of your favourite language and mock the inadequacies of all the competition, including the incompetencies of those who choose other software, your management, Marketing, Sales, project managers, tech support colleagues, and anyone else who wanders in range. For many years I've been using my own homebrew Bourne shell script for installing PostScript Type 1 fonts. Partly because I use a lot, and partly because I haven't found anything else that works as well. But it's time to rewrite it to handle many more font CDs, and to use the same standard fontfile naming conventions that Karl Berry defined for fontname. We'll leave aside the ghastly mess that is X fonts for the moment, as it's on the mend, and real-time font rendering with anti-aliasing is becoming workable for Unix-type systems Real Soon Now. We'll also leave aside the horrible blurry, jagged font rendition that seem to be endemic on Mac screens. That pretty much cuts out the competition, which is why I roll my own with and Type 1. The task is straightforward enough: Alan Jeffrey's fontinst program does a great job, but doesn't leave the font immediately usable in with a style file and FD files, and it's actually written in code, so while it's extremely portable, it ain't got no GUI apart from the editor you execute it in. I wanted to automate the whole shebang, so you just drop a AFM file on an icon and it does everything. That, too, is not hard, as I already have working code. But picking a scripting language was more interesting. I did once have a version written in a little MS-Windows utility language called Visual DisplayScript, but it's a bit limited, and won't execute anywhere else, so I've been poking through the portable scripting languages which can be wrapped for delivery on any platform…and here's where the war started: I have the code structure already, so I wanted rapid GUI development. Perl is right out: life is too short to learn how to write modem noise when my telco can generate it for free. Python was a very serious contender, especially since Seán McGrath does such a good sell on using it with wxwindows, but there's still too much noise in the syntax. I finally went for Tcl/Tk with vTcl for GUI development, and I must admit to having been influenced by the fact that when I downloaded and test-drove it, everything just worked first time and required virtually very little RTFM. (I should digress here to say that experimental CS languages were out, and C, C++, and Java were never contenders. I left programming in 3rd-generation languages behind me many years ago, and any language which kicks off by declaring its main routine to be null has to be brain-dead.) So I've been using the last fragments of what they call spare time for this, which is my excuse for not keeping these pages updated for a few months, and it's going well. I've found all sorts of gaps in the documentation and the code-generators, as well as in mky own knowledge, but nothing that is problematic so far. Tcl/Tk is a rich language with a simple syntax, and it does stuff in a way which matches the way my brain works, which is what I wanted. I have to thank the denizens of the vTcl help forum and the newsgroup for their help, especially my co-user Donald Arseneau (I had no idea he was a Tcl user). It's going to be called Gutta-Percha because of its associations, and it'll be Open Source (probably LPPL), and with luck I'll have it working for the Practical Conference next month.