On Mon, Oct 2, 2017 at 12:30 AM, Platonides platonides@gmail.com wrote:
The database schema is available at https://phabricator.wikimedia.org/source/mediawiki/browse/master/maintenance...
It's not too useful, but I would say that sometimes having the (opaque) rev_text_id could be handy, eg. to detect a rollback.
It is not possible to detect reversions with the rev_text_id, because a new one is generated every time. Eg.: https://en.wikipedia.org/w/index.php?title=Winston_Churchill&type=revisi...
root@dbstore1002:~$ my enwiki -e "SELECT rev_text_id FROM revision where rev_id = 797890286" +-------------+ | rev_text_id | +-------------+ | 807306920 | +-------------+ root@dbstore1002:~$ my enwiki -e "SELECT rev_text_id FROM revision where rev_id = 797893046" +-------------+ | rev_text_id | +-------------+ | 807309755 | +-------------+
rev_text_id are meaningless and prone to change (not stable). Some people use sha hashes to detect reversions. Given it is not useful without the private text table, it is more per formant to remove the whole column than suffering the penalty of conditional filtering on every revision query, plus we avoid the temptation to use it as if they were stable ids, as you suggested :-).