Older browsers are a different story. Both Firefox 2 and Internet Explorer 7 have trouble handling unknown elements; if you use the above HTML 5 elements to contain child elements as they're intended, both browsers will prematurely close the tags and punt their child elements into sibling positions on the DOM tree. You can verify this by using the DOM Inspector. What you'll see is that something that should look like this:

<header>
    <h1>we're sailing at the edges of time</h1>
    <p>we're drifting at the waterline</p>
</header>

Looks like this to the renderer:

<header/>
<h1>we're sailing at the edges of time</h1>
<p>we're drifting at the waterline</p>

It turns out that there's an easy solution for IE; just use Javascript to programatically create one of each element that you wish to use. Using the example above, the fix is simply to execute the following code:

<script type="text/javascript">document.createElement("header");</script>

You don't even need to append the new element to anything in the document tree; just creating it somehow kicks IE's parser into working order.

So that's easy, but it doesn't work in Firefox 2. In Firefox 2, unknown elements can't contain child elements unless the page is served as XHTML. Serving a page as XHTML doesn't mean just slapping a different DOCTYPE declaration on the page and running it through the validator; this means writing the page using the XHTML serialization of HTML 5 and serving the page using the application/xhtml+xml content-type, which forces your browser to parse the page as XML, which means that you have to be utterly scrupulous about well-formedness, because one error will cause the browser to barf up an error page.

I did this for years with chompy.net and remake/remodel, and it's harder than it looks. I'll probably end up writing scores of unit tests to shake out any XML well-formedness bugs in backwater's codebase.

1. verify that user is allowed to download
2. check that requested file is available
3. output HTTP headers
4. output the file using a call to readfile()

It turns out that the last step no longer works. In PHP 5.0.4 and only PHP 5.0.4, readfile() has a bug that causes it to halt output at about 1.9 MB. This means that only the tiniest of files can be downloaded successfully from Shoplifters.

I sent a note to Dreamhost support and even created a demonstration page for the bug so that they can see what I’m talking about. All they have to do to fix the issue is to upgrade PHP to any subsequent version, but they’re very conservative about PHP upgrades, so there’s no telling when that might happen.

There are things I could do to work around the problem, but I don’t like any of the options that I’ve been able to come up with:

1. link directly to download files instead of passing them through an intermediary script
2. downgrade to PHP 4
3. write Shoplifters 2 in some other language

SELECT DATE_FORMAT(date, '%Y-%m-%d') AS dom, COUNT(id) 
    AS revisions FROM version WHERE DATE_SUB(CURDATE(), 
    INTERVAL 14 DAY) <= date GROUP by dom;

Because a few days during the last two weeks had no activity at all, the results look like this:

+------------+-----------+
| dom        | revisions |
+------------+-----------+
| 2005-10-17 |        11 |
| 2005-10-18 |         2 |
| 2005-10-19 |         9 |
| 2005-10-21 |        11 |
| 2005-10-22 |         2 |
| 2005-10-23 |        13 |
| 2005-10-24 |        11 |
| 2005-10-25 |         3 |
| 2005-10-26 |        10 |
| 2005-10-27 |         4 |
| 2005-10-28 |         3 |
+------------+-----------+
11 rows in set (0.03 sec)

But I need my results to look like this:

+------------+-----------+
| dom        | revisions |
+------------+-----------+
| 2005-10-15 |         0 |
| 2005-10-16 |         0 |
| 2005-10-17 |        11 |
| 2005-10-18 |         2 |
| 2005-10-19 |         9 |
| 2005-10-20 |         0 |
| 2005-10-21 |        11 |
| 2005-10-22 |         2 |
| 2005-10-23 |        13 |
| 2005-10-24 |        11 |
| 2005-10-25 |         3 |
| 2005-10-26 |        10 |
| 2005-10-27 |         4 |
| 2005-10-28 |         3 |
+------------+-----------+
14 rows in set (0.03 sec)

That is, I want the days with no activity to show up in the results. How do I do that?

The "what links here" feature provides a good chance to discuss when AJAX might be used and one way you can do it. My thought process went something like this: I could have put each article's incoming links on a separate page, but since each page's list of incoming links is likely to be small, it made more sense to display them on the same page as the article text. At the same time, I didn't want an additional database query slowing down every page load, so I decided to make the list of incoming links only display when requested by the user. This required making an xmlhttprequest to the backend from the client-side code.

This is the Javascript code that does all the work:

function getIncomingLinks(articleName)
{
    var progress = document.getElementById('progress');
    var loadingMsg = document.createTextNode(" loading...");

    var loadingFunc = function(t) {
        progress.appendChild(loadingMsg);
    }

    var loadedFunc = function(t) {
        progress.removeChild(loadingMsg);
    }

    var errFunc = function(t) {
        alert('Error ' + t.status + ' -- ' + t.statusText);
    }

    var url = 'http://labs.spaceshipnofuture.org/icky/ajax.php';
    var params = 'mode=getIncomingLinks&q=' + articleName;
    var options = {
        method: 'get',
        parameters: params,
        onLoading: loadingFunc,
        onLoaded: loadedFunc,
        onFailure: errFunc
    }
    new Ajax.Updater('incoming-links', url, options);
}

This ought to be easy to follow. When the user clicks the "what links here" link, a call to getIncomingLinks() is made. getIncomingLinks() creates an Ajax.Updater object (part of the Prototype Javascript library). We provide Ajax.Updater with a URL to contact, the HTTP method to use ('get'), and a number of callback functions that are triggered depending on the HTTP request state. When the request completes, Ajax.Updater updates the contents of the HTTP element with the id 'incoming-links' using whatever output the backend sent. That's it.

The client-side code and backend code are rather tightly coupled here. If you don't want the backend to send formatted output, you could use Ajax.Request instead and have the backend send its output as JSON. Your client-side code would then unpack the JSON output and format the data using the DOM.