Ignore force = newer/older if mtimes are equal
#970
Conversation
|
I object to forcing root 2 if there are conflicting changes. The user has asked for newer to be forced, but there is no newer, and thus no basis to force at all. What they are trying to do, and what they actually want, is hard to guess, so I think we should just not. Yes, I realize this leads to interactive, but I think that's better than doing something unexpected that is basically bug-compatible with before. Preferring root2 over root1 in a tie I find bizarre in the first place and it smells like fallthrough instead of intended. |
|
I have no issue with your objection and it's only a minor change to the patch but I can't help but be reminded of https://xkcd.com/1172/ (of course, the PR in its current form does the same, so...). |
|
The xkcd is amusing. However, I feel that force=newer is a kludge to start with, and doesn't really belong in a sync tool. If you want rsync instead of unison, it's much simpler. |
force = newer/older preference works based on the difference of mtime in each replica. If mtimes are equal then it always propagates from root 2 to root 1. This yields unexpected results when propagating changes that don't update the mtime (that's basically all metadata updates), with different roots treated clearly differently. This patch makes it explicit that newer/older are only defined for when mtimes differ, ignoring the force = newer/older preference (and other related preferences, such as prefer and *partial) if mtimes are equal.
tleedjarv
force-pushed
the
force-newer-older
branch
from
8382e21 to
77b96cc
Compare
|
Pushed the new fix which completely ignores "newer" and "older" for equal mtimes. |
|
Thanks. Will hold off until 16Z on 5 October for discussion and then merge as soon as I realize it's after that :-) |
force = newer/olderis defined (and documented as such) only for changes where mtimes are different between the replicas. If mtimes are equal then root 2 is always forced over root 1.This produces unexpected results for changes that don't change mtime (basically all metadata changes). Since
force = newer/olderis undefined for such changes (and, importantly, is very difficult to define), this patch just ignores the preferencefor non-conflicting changesif mtimes are equal.For conflicting changes, the current default of forcing root 2 is kept unchanged.There is further discussion and background in the mailing list.