Last modified: 2011-07-19 16:37:20 UTC
The edit http://de.wikipedia.org/w/index.php?title=Nahostkonflikt&diff=prev&oldid=91224972 was flagged automatically, though it is neither a revert nor is the user Tristan-baeu in any group with autoreview right: http://de.wikipedia.org/w/api.php?action=query&list=users&ususers=Tristan-baeu&usprop=groups|rights But for some reason he has autoreview right.
Does the user meet $wgFlaggedRevsAutoconfirm? Before MW 1.19, the 'autoreview' implicit group is not actually *listed* on rights pages or API.
At the time of this edit he had 52 edits in main namespace of which some were reverted. He didn't enter 20 edit summary, but he had more than 20 autosummaries before r69277, so he probably meets that criterion. If 'totalCheckedEdits' => 50, # ...Edits before the stable version of pages really counts every edit to an article that was flagged in a later revision, even if the edit itself was reverted, then he would have met the conditions. But this wouldn't be a sensible configuration. His next edit in main namespace was not flagged automatically, since he has been blocked in the meantime. But the API still lists the autoreview right for him. For me the API even lists the autoreview group: http://de.wikipedia.org/w/api.php?action=query&list=users&ususers=Schnark&usprop=groups|rights Is this because I'm in editor group as well?
(In reply to comment #1) > Does the user meet $wgFlaggedRevsAutoconfirm? Before MW 1.19, the 'autoreview' > implicit group is not actually *listed* on rights pages or API. Scratch that. I looked the 1.17 code again, and it should be listed on the API even for 1.17 (which is why it showed up for you). It makes sense for you that the API list both the group and the right called 'autoreview'. For this user however, he only has the right and not the group. The only way this can normally happen is for users with the "bot" right (which he is not). (In reply to comment #2) > 'totalCheckedEdits' => 50, # ...Edits before the stable version of pages > really counts every edit to an article that was flagged in a later revision, > even if the edit itself was reverted, then he would have met the conditions. > But this wouldn't be a sensible configuration. Reverted edits are not subtracted from 'totalCheckedEdits', so they still count. Maybe there should be a 'maxRevertedEditRatio' setting to at least check the reverted:total ratio (which is fast to check)?
(In reply to comment #3) > It makes sense for you that the API list both the group and the right called > 'autoreview'. For this user however, he only has the right and not the group. > The only way this can normally happen is for users with the "bot" right (which > he is not). I opened bug 29938 for this, since I see nonsense for other rights, too. > Reverted edits are not subtracted from 'totalCheckedEdits', so they still > count. > > Maybe there should be a 'maxRevertedEditRatio' setting to at least check the > reverted:total ratio (which is fast to check)? In projects like the German Wikipedia every page will be flagged sooner or later, so the number of edits before a stable version is essentially the same as the number of edits in namespaces using this feature. It would make sense if this edit wouldn't count: http://de.wikipedia.org/w/index.php?title=Volkswagen&diff=prev&oldid=45866334 since it is not actively flagged, but http://de.wikipedia.org/w/index.php?title=Volkswagen&diff=next&oldid=45866334 (the next edit) should count, because it was flagged by an editor.
(In reply to comment #4) > In projects like the German Wikipedia every page will be flagged sooner or > later, so the number of edits before a stable version is essentially the same > as the number of edits in namespaces using this feature. > It's still useful since it means the edit was addressed one way or another. If many were vandalism, then the user would probably have been blocked already. There used to be a tally for the number of revisions *specifically* flagged, but that's a bit hit or miss when reviewers decide to review diffs covering multiple revisions.
So it looks like the user meet $wgFlaggedRevsAutoconfirm at the time. So this is not a bug.
Yes, he probably was in autoreview group at the time of the edit, which was correct according to the configuration. The only bug was that one in the API, which is fixed now and had nothing to do with flagged revisions. So this bug is invalid. I asked for comments about what the configuration should be on [[:de:Wikipedia Diskussion:Gesichtete Versionen]], if there is a consensus on changing the configuration or the wish for some other criteria to promote a user automatically to autoreview I'll open another bug for it.
On the topic of config changes: I can make it so that the *total* count of reverted edits are subtracted from the "checkedEdits" count. This would be conservative though, since some of the reverted edits would be to things like Talk pages and such. I can also add a 'maxRevertedEdits' parameter to $wgFlaggedRevsAutoconfirm.