Ævar Arnfjörð Bjarmason <avarab@xxxxxxxxx> writes: > ...And this seems to conflict with what Junio's summarized in > xmqqvaapb3r1.fsf@xxxxxxxxxxxxxxxxxxxxxxxxx. I.e. (if I'm reading it > correctly) it's not correct to say that we unlink the existing file, > then replace it, don't we create a new one, and then rename it in-place? No, my recollection was incorrect. entry.c::checkout_entry() does an unlink() then write_entry() to the final place without any rename-to-finish phase.
