Shattered Crystalline Matrix

This isn’t meant to last,
this is for right now.
— Nine Inch Nails, Last

Spent several hours this morning importing my old old blog posts into WordPress; the import is now complete. Early on in the life of the blog, I began generating an Atom feed directly from my hand-written XHTML; later on, I changed my markup to use the hAtom microformat, and ditched my custom transform in favour of hAtom2Atom. I used this again to run a conversion over each month’s posts, although I had to edit the oldest into hAtom form since they were using a custom class schema; just a trivial matter of assigning the correct classes. I then had to remove the doctype declaration from each page, because Saxon was failing to fetch the XHTMTL 1.1 DTD for some reason; since the DTD was completely unnecessary for this transform, I decided it would be easier to just ditch it.

Now that I had the ability to generate an Atom feed for each page, the next trick was importing this into WordPress somehow. WordPress supports importing an RSS 2.0 feed, so I tracked down another transform to convert the Atom to RSS 2.0: atom2rss-exslt.xsl. After hacking it slightly to run on Saxon 6 (the decode-uri function detection doesn’t work, since it checks for a later version of the Saxon processor), I had what looked like good RSS 2.0 output, which I imported. Unfortunately, this didn’t work out so well; the tags weren’t imported, and WordPress inserts a <br> tag for each newline; since the output of the transform had a bunch of extraneous newlines, this meant that my posts were now littered with extraneous line breaks. I edited the transform to get rid of the newlines, but still wasn’t very happy with this.

So… plan B! WordPress has an import/export format called WordPress eXtended RSS; basically, RSS 2.0 plus a whole whack of custom WordPress extension elements. I spent around an hour hacking the transform to generate WXR output. This was even more painful than it sounds; for example, encoded content is “supported” by stripping any CDATA section start/end markers, and then importing the content as-is. Even if there wasn’t a CDATA section. I guess they only care about reading their own output.

As a final touch, I hacked the transform a little more to insert an invisible anchor tag at the beginning of each post so that even my old permalinks will work. This was fortunately quite easy to do, since my old URL scheme was year/month, which happens to be supported by WordPress too; I just needed the anchors so that you would get taken to the correct post.