Brion, I think that not only en dumps failed. I found some errors in the most recent de dump (20051017_pages_articles.xml): As far as I inspected the xml file manually, I saw several <title>'s which do not belong to the comtent. I.e.: <title>Vitruvius</title> contains an article about the planet Venus <title>Indianische Flöte</title> contains the history of Poland <title>Marlon Brando</title> contains Madonna (sic!) and so on Besides, there a many articles which exceptionally length in the dump, which are not belonging into Namespace #0. Cheers jo

Ok, confirmed on dewiki dump; here's a fragment example: http://meta.wikimedia.org/wiki/User:Brion_VIBBER/Crap Obviously there's a synchronization bug in my prefetch code. I'll try and debug this today or tonight and restart the dumps tonight/tomorrow. Don't use any of the 20051017 dumps; they're all suspect. -- brion vibber (brion @ pobox.com)

Folke Behrens wrote: I want to keep them around for bug comparison at the moment, but I have gone ahead and renamed them with a ".broken" extension which I hope will discourage people from downloading them. ;) -- brion vibber (brion @ pobox.com)

Jochen Magnus wrote: Yay, something worked! ;) I forgot to include the detail in my prior message, that some of the older dumps may contain duplicate page titles due to the existence of old non-normalized Unicode titles in the databases; that can lead to 'existing key' errors on the unique title indexes when importing with mwdumper. Those have been purged, and so it shouldn't happen anymore. I am interested in details of failure reports with importDump.php; the prior memory leaks with 1.5 prereleases should be fixed as of 1.5.0 but there might be intermittent or specific problems. Unfortunately importDump is relatively slow because it works page-by-page (mwdumper does faster bulk imports, assuming a blank slate empty database to start with) so it takes a while to test or confirm failures. :) -- brion vibber (brion @ pobox.com)