Open bgamari opened 5 years ago
I suspect this is because we don't resolve usernames to their GitLab equivalents. This needs to be fixed.
Hmm, actually it looks like we do resolve gitlab usernames. In this case I have no idea what is going wrong here.
Oh, I see what's going on here: we add/remove the committed subscription changes to the un-mapped CC-list.
Never mind, that's clearly not the issue either. Sigh.
IIUC, the issue is a much more conceptual one. In TicketImport.hs:463
, we change the ticketCC
field such that its new value no longer contains any of the users we have just subscribed or unsubscribed. But this is wrong. What we want instead is:
Update
into per-userID Update
s (that is, we would want to go from Update (Set UserID)
to something like Map UserID (Update ())
)Map UserID (Update ())
into an Update (Set UserID)
Or, alternatively (and this is probably the easier solution): Just cull the user IDs we have handled from both old and new values in the Update
.
936eb4ac77cb8f24ae4ab1098c52bbdd2a14abe6 implements the alternative solution (cull user IDs from both old and new values). Seems to work, though testing against a pristine install seems prudent.
https://gitlab.staging.haskell.org/ghc/ghc/issues/16070 https://gitlab.staging.haskell.org/ghc/ghc/issues/16146