On Fri, Mar 15, 2013 at 7:08 PM, Christopher Morrow <morrowc.lists@xxxxxxxxx> wrote: [....] >> >> What I mean is that if there is disk corruption on the server hosting >> the drafts (which can happen post write), rsync will happily send the >> checksum of the corrupted draft. Git's mechanism makes such a >> probability infinitesimal. > > wait, so.. if the disk fails things go bad... I'm confused. > If the disk goes bad so as to provoke a misread of a sector, post write, the file is effectively corrupted. If this happens with git, the checksum calculated on write will fail to match, and the corruption is detected. Not only that, but: * you may then recover the repository from another existing one; * this mechanism detects corruption _even if you have a bare_ (since what is checksummed is not the file in the working tree but its blob representation in the object tree). -- Francis Galiegue, fgaliegue@xxxxxxxxx JSON Schema in Java: http://json-schema-validator.herokuapp.com