MediaWiki/reverse engineering

Ok, enough of trying to find the right queries after the upgrade to 1.5 MediaWiki.

Lets just log the sql queries from the server.

A page being requested

get page_id

SELECT page_id FROM `page` WHERE page_namespace = '0' AND page_title = 'Ogg' LIMIT 1;

get page_latest

SELECT page_id,page_namespace,page_title,page_restrictions,page_counter,page_is_redirect,page_is_new,page_random,page_touched,page_latest,page_len FROM `page` WHERE page_id = '17369' LIMIT 1;

after having page_latest (32808) ,which is = rev_id in revision we can:

get revision_text_id

SELECT rev_text_id from revision where rev_id=32808;

after having rev_text_id (32770), which is = old_id, we can finally:

get the actual content

SELECT old_text,old_flags FROM `text` WHERE old_id = '32770';

brion: old_id=rev_text_id
brion: rev_id=page_latest
brion: rev_page=page_id

Searching for content

To seach for content in pages and return page titles that match, we would have to revert the process:

get old_id of matching page

select old_id from text where old_text LIKE "%vollmilch%" limit 1;

+--------+
| old_id |
+--------+
|  32778 |
+--------+

get rev_page from old_id

select rev_page from revision where rev_text_id=32778;

+----------+
| rev_page |
+----------+
|    10425 |
+----------+

result / page_title

select page_title from page where page_id=10425;

+------------+
| page_title |
+------------+
| Sandbox    |
+------------+

Multiple search results

select old_id from text where old_text LIKE "%finn%" LIMIT 5;

+--------+
| old_id |
+--------+
|  19561 |
|  12621 |
|  12622 |
|  12682 |
|  12683 |
+--------+

select distinct rev_page from revision where rev_text_id IN (19561,12621,12622,12682,12683);

+----------+
| rev_page |
+----------+
|     9559 |
|     9609 |
|    13622 |
+----------+

select page_title from page where page_id IN (9559,9609,13622);

+----------------+
| page_title     |
+----------------+
| Debian/Posters |
| Donations      |
| Forum          |
+----------------+

...OR, one could as well just use the searchindex table:

select si_page,si_title from searchindex where si_text LIKE "%Vorbis%";

+---------+-----------+
| si_page | si_title  |
+---------+-----------+
|    9746 | gnump3d   |
|   17369 | ogg       |
|   17368 | zinf      |
|   17367 | icecast   |
|   15735 | converter |
+---------+-----------+

but we want the nice page titles, not this lowercase version, so we join the table with page, since si_page = page_id

select page_title from searchindex join page on page_id=si_page where si_text LIKE "%Vorbis%";

+------------+
| page_title |
+------------+
| Converter  |
| GnuMp3d    |
| Icecast    |
| Ogg        |
| Zinf       |
+------------+

dddooh, so much easier. ;)

ok, got it

< mutante> ok,with newer pages i can use rev_id=page_latest and old_id=rev_text_id, but with older pages i have equal values in rev_id and rev_text_id and i get an empty set if i then try to select page_title from page where page_latest=

<brion> mutante: rev_page=page_id

< brion> page_latest points to the latest rev_id in the page for convenience