While investigating T218819, I found that many revision events in mediawiki_history are lacking page and namespace information in revision event. The effect is major: when calculating global monthly active editors (where only content edits are counted), this bug reduced the number by as much as 10 000 per month.
There seem to be at least three different problems at work:
Deleted pages where all the revisions have null page_namespace_historical, page_namespace_is_content and page_namespace_is_content_historical:
- nlwiki page 5160672
- commonswiki page 71797356
- wikidatawiki page 9637937
- arwiki page 4970274
- commonswiki page 73916373
- ruwiki page 7054391
Revisions with null page_namespace_historical and page_namespace_is_content_historical. Most but not all of the revisions to the pages concerned are affected.
- https://en.wikipedia.org/w/index.php?diff=67017781 (page 28408157)
- https://www.wikidata.org/w/index.php?diff=155712677 (page 21524228)
- https://pt.wikipedia.org/w/index.php?diff=1691356 (page 96328)
- https://en.wikipedia.org/w/index.php?diff=820879007 (page 56326900)
Revisions where the join to the page table seems to have failed entirely, because they have have null page_title, page_namespace, and page_namespace_is_content (including historical version) and page_creation_timestamp. Most but not all of the revisions to the pages concerned are affected.
- https://en.wikipedia.org/w/index.php?diff=859361756 (page 40012938)
- https://pt.wikipedia.org/w/index.php?diff=2692528 (page 3177643)
- https://en.wikipedia.org/w/index.php?diff=269025183 (page 29397754)
- https://en.wikipedia.org/w/index.php?diff=347411263 (page 30865452)