Re: master fails to build on FreeBSD when ACL support is on

[Paul Eggert]

Stefan Monnier wrote:
> Talking about what happens when rename-file uses copy-file: shouldn't it
> first copy-file to a temporary file near the destination and then
> rename-file that temp file?  Otherwise the destination will temporarily
> be in a "partially written" state, which is usually not what
> `rename-file` callers want, right?
I suspect rename-file is trying to mimic 'mv' here. With GNU 'mv', the 
destination can temporarily be in a partially-written state when file system 
boundaries are being crossed. This behavior is allowed by POSIX. Although the 
behavior you describe does have advantages, it is not allowed by POSIX for 'mv' 
and it can exhaust the destination file system even when there's room for the 
final result.