Last modified: 2014-06-17 17:02:37 UTC
Per reports about the Thanks extension (https://en.wikipedia.org/wiki/Wikipedia:Village_pump_(technical)#Getting_rid_of_.22thank.22) which uses Vector's experiments preference as an opt-out, it seems like this preference might not be working? I am not wholly sure whether it's a bug in the Thanks extension or Vector, but either way the preference should be preserved, since multiple features rely on it for opt-out purposes.
Related to bug 54852?
With https://gerrit.wikimedia.org/r/#/c/87023/, there is no longer a user preference that can be used to opt-out of Thanks.
(In reply to comment #0) > I am not wholly sure whether it's a bug in the Thanks extension or Vector, > but either way the preference should be preserved, since multiple features > rely on it for opt-out purposes. I'm not sure this is still true. This bug may be invalid. I'm not sure a user preference ever made sense here. Users wishing to hide thanks can use CSS or JS or a local gadget, surely?
(In reply to comment #2) > With https://gerrit.wikimedia.org/r/#/c/87023/, there is no longer a user > preference that can be used to opt-out of Thanks. Okay, I thought this might be the issue. Thanks for the heads up guys. Did anyone talk to Kaldari, Fabrice, et al before merging this? I don't think it was intentional to remove the preference without at least notifying the community that opt-out was being taken away. You can see from the link I provided that the community and the primary author of the feature (Kaldari) seem to be taken off-guard here. I agree with bug 54852 that the experimental opt-out should not be housed in the Vector extension, but it may still be a useful preference, or at least Thanks maintainers might like to explain why it was okay to remove the opt-out at this point. Plus, did we check that there were no other currently deployed extensions depending on the preference? I am unaware of how we can check for dependencies on preferences (other than say, grepping for the preference name). In general, it seems like this preference was removed prematurely.
(In reply to comment #4) > In general, it seems like this preference was removed prematurely. I can't imagine the number of people opting out of using the Thanks extension is more than a few dozen users across Wikimedia wikis, though maybe you can gather proper stats?
(In reply to comment #5) > (In reply to comment #4) > > In general, it seems like this preference was removed prematurely. > > I can't imagine the number of people opting out of using the Thanks extension > is more than a few dozen users across Wikimedia wikis, though maybe you can > gather proper stats? I don't know how preference data such as from the database reports is generated, but my lazy late-night data gathering can say this... Schema:PrefUpdate seems to have been logging preference updates from July until the present. If I am understanding the schema correctly, it looks like since the start of logging, 370 enwiki users have turned on the vector-noexperiments preference while 44 turned it off. The total for all wikis is 944 turned it on, while 97 turned it off. Now, for many of these updates, Thanks was clearly not the reason, since it wasn't deployed everywhere for this duration. But it gives you an idea of how often the preference was being fiddled with by users.
I filed the bug 54852 and asked about it on IRC; I've only become aware of the solution after it was merged. I'm okay with the way this was done (per what MZMcBride said above), but I have no stats handy. Splitting the pref out to a tiny WMF-only extension might be a good idea if we want to keep it, though (as suggested on the bug), but I suggest that we keep it killed and create a separate pref for Thanks if it needs it (and possibly migrate the values). (In reply to comment #4) > Did anyone talk to Kaldari, Fabrice, et al before merging this? I don't think > it was intentional to remove the preference without at least notifying the > community that opt-out was being taken away. You can see from the link I > provided that the community and the primary author of the feature (Kaldari) > seem to be taken off-guard here. Benny (who wrote that patch) is (or was last time I asked) the sole programmer currently working on Echo, and Thanks seems to be basically a part of that (Erik Bernhardson was – again, last time I asked – mostly working on Flow; I don't know of any place where this could be checked without asking everyone separately :) ). Kaldari mentioned leaving the E2 team back when it was called E2, so he's probably not that interested. (In reply to comment #6) > Plus, did we check that there were no other currently deployed extensions > depending on the preference? I assume Benny would know. I am not aware of any myself (except for Thanks). > I am unaware of how we can check for dependencies > on preferences (other than say, grepping for the preference name). There isn't any, which is one of the reasons why this was a bad idea. Other experimental tools like VE use separate preferences, IMO Thanks should too.
I think beta features should help with things like this in the future, but I'm concerned that the decision was made to make it available for all without any kind of notification of the community (that I can see, anyway). I'm going to follow this up internally.
(In reply to comment #8) Sounds like a storm in a teacup. Anyway, if we want to have this work again for now and right now, we'd just have to roll back the Vector version deployed on the cluster a few commits back (to the one before the offending patch linked above).
At least in regards to the Thanks extension, benny has deployed https://gerrit.wikimedia.org/r/#/c/87023/ which essentially makes Thanks no longer experimental(no longer uses vector-noexperiments) and gives it to all users. I'm indifferent on what other extensions want to do/not do here.
Okay, since this issue is specific to Thanks, and I don't think we want to back out the change by Benny et al. (I'd like to remove dependencies on Vector extension just like everyone else). I'll resolve this and we can talk more about how to deal with the Thanks use case in another discussion. Resolving as invalid since a preference that was removed should by definition not work. ;)
