Last modified: 2010-05-15 14:35:47 UTC
Playing with the dumps on zedler, I just found out that a lot of pages have a page_random value of 0, which causes them to never show up as a "random page". This seems to be the case mainly for image description pages (namespace 6). This not only causes the the "random page" and "random image" features to behave incorrectly, it also effects tools that rely on a random selection of pages. Please have a look at what is causing this. Below are some of the results I found when playing with this: mysql> use commonswiki; mysql> select count(*) from page where page_random = 0; +----------+ | 55832 | +----------+ mysql> select count(*) from page where page_random = 0 and page_namespace = 6; +----------+ | 54409 | +----------+ mysql> use enwiki; mysql> select count(*) from page where page_random = 0; +----------+ | 62879 | +----------+ mysql> select count(*) from page where page_random = 0 and page_namespace = 6; +----------+ | 56830 | +----------+ (I appologize in advance for broken formating - MediaZill needs a preview button...)
This is a conversion issue -- since Special:Randompage previously only applied to non- redirects in namespace 0, code which inserted image description pages and redirects tended to miss setting cur_random. Because of the consolidation of article insert code in 1.5, this is now resolved for new pages. However there will still be many pages left with page_random=0, as Daniel points out. I've fixed this in Wikimedia databases, but it might be nice to have a query like UPDATE page SET page_random=RAND() WHERE page_random=0 in future installers.
Bear with me if I'm missing the trick or the point, but wouldn't you want that to be UPDATE page SET page_random=RAND() WHERE page_random=0, page_is_redirect=0; in order to exclude redirects?
(In reply to comment #2) > to be UPDATE page SET page_random=RAND() WHERE page_random=0, > page_is_redirect=0; in order to exclude redirects? First, that's invalid SQL, second page_random should be a nonzero value for every row, including for rows where page_is_redirect = 0.
Sorry; I wasn't thinking. Yes, it's invalid SQL and yes I made the glaring balls-up of specifying WHERE page_is_redirect = 0, when in fact my comment on it made it clear I meant the other way round.
Added a database update for this bug to HEAD, marking this as FIXED in CVS HEAD
