Comment 9 for bug 113809

I use Bazaar for a single person project, in theory one of the simplest use cases. But I do make local commits, where dragons seem to lie.

I have now had to go to Andrew Bennetts for help, oh, getting on towards ten times now with problems along the lines of "um, it says it can't add foo.BASE, why would it even want to add foo.BASE, that's not meant to be versioned..." or "why is there a foo.BASE and a foo.BASE.moved" and now (with bug 236724) "why is there a content conflict, text conflict AND path conflict all on the same file, all I did was an update?"

Having my working tree stomped on with spurious conflicts matters. It can take me upwards of 45 minutes each time this happens to double-check that all the files are intact and that stuff I wanted hasn't vanished into foo.BASE.moved.

Unless/until the problems with the checkout model are resolved, I advocate that Bazaar not allow the "update" command to be used as readily in checkouts. Certainly it seems that "update" should essentially never be run in a working tree of a checkout with both unmerged local commits and totally uncommited changes. I would much rather deal with "Please commit all local changes with commit --local before updating from the master branch" than try to remember AGAIN which of foo.{BASE,OTHER,THIS}.moved are likely to be important and which ones I need to call "bzr resolve" on.