Jeff King <peff@xxxxxxxx> writes: > On Mon, Mar 20, 2017 at 08:56:13PM -0400, Kyle Meyer wrote: [...] >> I'm confused about the setup for the "logged by touch" tests. >> d0ab05849 (tests: remove some direct access to .git/logs, 2015-07-27) >> changed the setup to delete the log file itself rather than its >> contents. The reflog was then recreated by using "--create-reflog" in >> the "create $m (logged by touch)" test. What I don't understand is >> how this change fits with d0ab05849, which seems to be concerned with >> loosening the assumption that the logs are stored in .git/logs. > > I suspect the answer is that the conversion was incomplete. That commit > was done for alternate ref backends, which is an ongoing saga. > > I think it's OK to leave it as-is for now. It's not clear what "logged > by touch" will look like for backends that don't use the filesystem. > Probably it will need to call "update-ref --create-reflog" to kickstart > it, and then further updates will automatically write to it. > > At that point the "rm -f" would need to become "tell the backend to > delete this reflog". There's no command for that now, but we can add one > later. Until then, I suspect the "rm -f" would be a noop. That means > that the first --create-reflog test is failing to test what it claims, > but the result passes anyway. > > And that probably answers the question about why the conversion is > half-done. It was enough to get the tests to stop complaining when built > with an alternate ref backend. :) OK, thanks for the background. -- Kyle