I had a chance this weekend to do a little work on importing a large (4000 or so articles and pages) site into WordPress. It was an interesting bit of work, with a certain amount learning required on my part – which translated into some flailing around on to establish the toolset.
Lesson 1: ALWAYS use a database in preference to anything else when you can.
I wasted a couple hours trying to clean up the data for CSV import using any of a number of WordPress plugins. Unfortunately, CSV import is half-assed at best – more like about quarter-assed, and any cleanup in Excel is excruciatingly slow.
Some of the data came out with mismatched quotes, leaving me with aberrant entries in the spreadsheet that caused Excel to throw an out-of-memory error and refuse to process them when I tried to delete the bad rows or even cells from those bad rows.
Even attempting to work with the CSV data using Text::CSV in Perl was problematic because the site export data (from phpMyAdmin) was fundamentally broken. I chalk that partially up to the charset problems we’ll talk about later.
I loaded up the database using MAMP, which worked perfectly well, and was able to use Perl DBI to pull the pages and posts out without a hitch, even the ones with weirdo character set problems.
Lesson 2: address character set problems first
I had a number of problems with the XMLRPC interface to WordPress (which otherwise is great, see below) when the data contained improperly encoded non-ASCII characters
. I was eventually forced to write code to swap the strings into hex, find the bad 3 and 4 character runs, and replace them with the appropriate Latin-1 substitutes (note that these don’t quite match that table – I had to look for the ”e2ac’ or ‘c3’ delimiter characters in the input to figure out where the bad characters were. Once I hit on this idea, it worked very well.
Lesson 3: build in checkpointing from the start for large import jobs
The various problems ended up causing me to repeatedly wipe the WordPress posts database and restart the import, which wasted a lot of time. I did not count that toward the overall time needed to complete when I charged my client. If I had, it would have been more like 20-24 hours instead of 6. Fortunately the imports were, until a failure occurred, a start-it-and-forget-it process. It was necessary to wipe the database between tried because WordPress otherwise very carefully preserves all the previous versions, and cleaning them out is even slower.
I hit on the expedient of recording the row ID of an item each time one successfully imported and dumping that list out in a Perl END block. If the program fell over and exited due to a charset problem, I got a list of the rows that had processed OK which I could then add to an ignore list. Subsequent runs could simply exclude those records to get me straight to the stuff I hadn’t done yet and and to avoid duplicate entries.
I had previously tried just logging the bad ones and going back to redo those, but it turned out to be easier to exclude than include.
Lesson 4: WordPress::API and WordPress XMLRPC are *great*.
I was able to find the WordPress::API
module on CPAN, which provides a nice object-oriented wrapper around WordPress XMLRPC
. With that, I was able to programmatically add posts and pages about as fast as I could pull them out of the local database.
Lesson 5: XMLRPC just doesn’t support some stuff
You can’t add users or authors via XMLRPC, sadly. In the future, the better thing to do would probably be to log directly in to the server you’re configuring, load the old data into the database, and use the PHP API calls directly to create users and authors as well as directly load the data into WordPress. I decided not to embark on this, this time, because I’m faster and more able in Perl than I am in PHP, and I decided it would be faster to go that way than try to teach myself a new programming language and solve the problem simultaneously.
I’d call this mostly successful. The data made it in to the WordPress installation, and I have an XML dump from WordPress that will let me restore it at will. All of the data ended up where it was supposed to go, and it all looks complete. I have a stash of techniques and sample code to work with if I need to do it again.