I needed a brand-new service, as there was no existing POTS or DSL line – only Telstra cable and CATV. There was a lengthy waiting period, and after the installation date, I called an electrician to wire up a socket. It turned out the MDF hadn’t been tagged. After much arguing, Internode sent someone out to tag it properly. However, I had to call out (and pay) an electrician to jumper it. So if your ISP tells you your MDF or boundary point is tagged, don’t believe them – check for yourself before you call out an electrician.

At this point, I had a socket connected to the correct cable and pair, but still no DSL. Internode insisted that I find an analog telephone to listen to the line. I want naked DSL – why should I need an analog telephone? Anyway, I discovered that I had a POTS service of some kind, and even found out what its number was, and told Internode. They informed me that they needed a technician to come and “perform tests”. It took another week for the guy to come out, and he didn’t arrive on time. He just confirmed what I’d told them: my socket was connected to the correct cable and pair, but had POTS service. Apparently they don’t believe their customers.

After this, it took another day for the exchange to be patched correctly. I now have a working Internet connection, but my high-speed ADSL2+ here is barely faster than my plain ADSL1 in Melbourne, and I now have to fight for a refund for the period when I was being billed for a service that didn’t work. If you’re thinking of getting naked ADSL, save yourself the trouble and get something where a single vendor is responsible for the whole solution. Cable Internet or ADSL with a Telstra DSLAM would be a whole lot less trouble.

Now with WordPress 2.3, all that’s in the past. WordPress now seems to be able to do the right thing with text encodings. It’s just too bad the upgrade script can’t clean up the rot left from previous versions. After running the upgrade script, I found that every piece of Japanese text, every typographical quote, every accented character, in fact everything outside 7-bit ASCII, was horribly mangled. Now I had a number of options for going forward:

• Go back to a previous version of WordPress
• Leave it and hope no-one minds.
• Delete every post that got mangled.
• Manually fix every affected post.
• Come up with a l33t way to solve it without manual effort.

Now the first option would have been easy. I had a complete backup (like you should, too), and I could have rolled it back in a matter of minutes. But I like to be on the curve, and I like to have all the newest features, even if I never use them. Also, having the latest security updates is nice.

The second option wouldn’t fly, because even if the readers wouldn’t mind, I’d mind. The third option would probably mean deleting every post, since I’m in the habit of using typographical quotes, non-breaking spaces and dashes (as opposed to hyphens). Deleting all my posts after an upgrade would defeat the purpose of keeping a blog. The fourth option would be excessively time-consuming, and I’d have to play fill the blanks, which may not even be possible if important things were mangled.

So the only way to fix it would be to call on my inner geek. I had a quick look at the database contents in phpMyAdmin, and had a look at the database code in WordPress. I noted that the new tables created by WordPress 2.3 had the collation utf8_general_ci while the upgraded tables had the collation latin1_swedish_ci, and WordPress was asking MySQL to communicate in UTF-8. Armed with this, I downloaded a UTF-8 SQL dump of the database.

The rest of it was actually reasonably simple: I opened the SQL dump in TextWrangler, which correctly identified it as being UTF-8 with no “byte order mark”. I then found all references to the latin1 character set and replaced them with utf8. That would fix the issue with the upgraded tables having the wrong collation, but not the corrupted data.

So here’s the trick: you need to convert the UTF-8 representation of what was in in the database back to its old binary representation, and then interpret that as UTF-8. I tried to save the file as ISO Latin 1 (ISO 8859-1), but TextWrangler complained about unmappable characters. It turns out that MySQL’s latin1 is actually Windows Latin 1 (code page 1252). So I saved the file in this encoding, and then told TextWrangler to reinterpret it as UTF-8. It all went smoothly, and I had my data back! I could play the SQL dump back on the server, and everything is as it should be.

So what are the morals of the story?

• Always keep backups – particularly when you plan to do something drastic like an upgrade. Even though I didn’t actually need the backup this time, it was comforting to know it was there.
• Don’t trust upgrade/migration scripts – always check the result to ensure it’s actually what you want.
• Store data in appropriate formats – hacks will always come back and bite you. I shouldn’t have been using WordPress when I knew it was doing the wrong thing with my data.
• When you’re writing a migration script, try to ensure that it actually works! Then you don’t risk infuriating and/or losing your users.

Take its text rendering, for example. Since the primary purpose of a web browser is to get text on the screen, you’d think they’d have that right. But no, apparently version 2.0 is still too early to expect decent text rendering. Compare these two snaps:

Firefox	Safari

How has Firefox managed to screw up the fixed-pitch text so badly? It’s just plain illegible! I have absolutely know idea, but however they really should have fixed this kind of thing before version 1.0 – not left it in at 2.0. How about italic text. Maybe they could get that right:

Firefox	Safari

Once again, Safari has rendered it beautifully, but Firefox looks like it’s using a synthetic oblique style, the way System 6 used to when you didn’t have an italic version of the font available – it’s most noticeable in the capital S. And then when you select the text, some of the last italic letter gets cut off. Come on, this is pretty basic stuff, guys!

On the topic of selection, Firefox won’t use Mac conventions in handling double-click and drag. It’s supposed to select whole words, but Firefox selects just one whole word. The same goes for triple-click and drag for lines. (Yes, I know Safari’s selection behaviour isn’t quite standard, and there’s no excuse for that, either.) And speaking of things being non-standard, why can’t Firefox use standard OS widgets? For example, the drop-down menus from items on the bookmark bar don’t respond to clicks in the same way as regular menus (items with submenus, in particular). The controls on forms just look like horrible Windows wannabes:

Firefox	Safari

Safari is more usable, too. In Firefox’s bookmarks window, why can’t I drag object from the tree view in the left-hand pane? Why can’t I rename or edit a link in-place, rather than clicking the Properties or Rename button? Speaking of which, why are there two buttons, when both of them bring up exactly the same sheet, with exactly the same text field selected? And on the topic of bookmarks, when I try to drag the URL to the bookmarks bar, Firefox puts a tooltip in the way to thwart my efforts:

When you have lots of tabs on the same site, Safari cuts off any common prefix it can find in the titles, so you have more chance of knowing which is which:

Firefox
Safari

And on top of this, there are the small things. Like ripping a YouTube video, for example – in Safari, you can easily find it in the activity window, and hold option and double-click. Firefox won’t use the system-wide spelling dictionaries. Firefox takes longer to launch.

So if I hate it, why do I use it? First of all, on Linux and Windows there’s no meaningful competition. On the Mac, Safari has three flaws that are too bad to live with: it can’t deal with table cells spanning multiple rows in XHTML (although it can in regular HTML – this mystifies me), it doesn’t deal with character entities correctly in XHTML and it gives you the spinning pinwheel of death far too often.

While I’m on the topic, I’ll put in a plug for my hosting provider Selpaw Services in Perth. Their support is always speedy and first-rate. Thanks for everything, Luke.

Suppose I’m a bot herder (I hope it’s obvious that this is purely hypothetical). I unleash malware that takes over computers and “calls home,” allowing me to send spam for my paying clients. Since the number of people who read spam is very low, and the number of people who buy products advertised in spam is even lower, I need to send huge volumes of spam to make my services worthwhile for my clients. And sending lots of spam requires lots of compromised computers.

As software vendors patch vulnerabilities in their software, I have to find new vulnerabilities and write new malware to exploit them. This requires considerable effort on my part, and takes away from time I could spend doing things I enjoy. Also, as more users become more security-conscious, there are less machines left open to attack.

Suppose for a moment Windows, Linux and Mac OS X are all equally exploitable, and writing a piece of malware for each takes the same amount of time. What am I going to do? Am I going to write three sets of malware to attack the three platforms, or will I pick one to concentrate on?

Of course, the answer depends on market share. The more even the market share, the more likely I would be to write malware for multiple platforms. Also, it’s worth thinking about where the machines are primarily used.

Linux is used primarily in server and professional environments. Machines that are critical for business operations run by tech-savy operators means the machines are more likely to be secured properly and suspicious software will be removed promptly. So scratch Linux. I want to target home users with DSL or cable internet.

So I’m left with a choice of targeting Windows or OS X. What do I do? I look at market share. I know these figures are probably wrong, but suppose OS X runs on 5% of my target machines and Windows runs on 90%. What am I going to target?

The answer should be obvious. I’ll target Windows. I could target OS X as well, but then I’d be spending twice as much time writing malware for less than 6% more compromised machines to send spam from. It just doesn’t make business sense.