PÁLI Gábor János

Thank you all for your help too!

I have no strong feelings against using the `view_row` tag for the maps. I removed it because I felt the code simpler but I would not say it is future-proof...

For the record, the current version works fine in a mixed-version environment, in old coordinator—new worker, new coordinator—old worker configurations. This was tested with a database of 100K documents, 3...

> Since we're eventually trying to report the work done by a worker up until the row it emitted, would it make sense to have the `completed` messages return the...

> the streaming may also end if we reach the end of the btree. Then, instead of a row we'd emit a `complete` message. I would not call those cases...

> In fact, I am not sure if "abruptly terminated" exactly applies here in general. That part works like that by design. In the majority of cases, the `complete` message...

> It is related to correctness that's why I mentioned the other pr. I am sure you are right about that. But I think we should carry on with the...

> > But is this a real problem here? > It is if we want make any correctness guarantees. It doesn't have as much to do with commutativity and non-associativity...

> I noticed this CI failure in a full CI run. It seems to be in the test code in the replicator Oh, the good old CentOS 8. I will...

I was trying to reproduce this locally (with the CouchDB CI CentOS 8 container) but I could not. Hence it seems to be a flaky case. But I think there...