On Mon, Jun 11, 2018 at 11:02:42AM -0700, Junio C Hamano wrote: > > Aside from us not having to worry about emulating the umask, another > reason why we prefer "create, complete the write, and then finally > rename" over "overwrite and let it fail in the middle" is that the > former makes sure we never leave the path in a bad state. But the current checkout implementation does not do this. It writes directly to the target location. The only difference to in-place checkout is that files are unlinked before they are opened for writing.
