Note that the reporting is different for 7.0. If the net is split and each part has a zone, the missing connection will always be reported between the two zone parts (as opposed to between a random pad and a zone part). Additionally, the ratsnest line is optimized to be a reasonable approximation of the shortest line between the two zone parts.
People seem happy enough with it so far, but then the nightly users are more used to KiCad’s quirks, so I’m not sure that’s much data…
Aye, I just created a simple test project for this and just mentioning that parts of a zone are not connected makes me quite happy enough in this regard.
KiCad is in feature freeze for the forthcoming version 7 as of 1st October with, I believe a release date scheduled before the end of Jan 23. New functionality won’t get added to V6 but the new version should be available soon - and can be tried already if you install a nightly version. Generally, after feature freeze, the nightly versions are pretty stable and, although you might not want to use them in a production environment, they are pretty solid. The more eyes on the nightly versions, the sooner bugs get found and fixed so, hopefully, the release version will be fairly bug free.
This was considered a bug fix and so would have been a candidate for back-porting to 6.0. However, it does the optimizations in parallel, and a multi-threading bug was just found & fixed in it. So that resets the clock on its “chill” period in master, and with the release of 7.0 forthcoming it’s unclear if there will be enough time for it to chill and then get back to 6.0.x. (Once 7.0 is released as a stable version we stop back-porting fixes to 6.0.x.)