Re: Why O_EXCL would make this difference? I am puzzled..

On Thu, 14 Jul 2005, Linus Torvalds wrote:

It's the "Adding" case in git-merge-one-file-script, which does

	git-checkout-cache -u -f -- "$4"

and it's because of this:

	lstat64("DF", {st_mode=S_IFDIR|0775, st_size=4096, ...}) = 0
	unlink("DF")                            = -1 EISDIR (Is a directory)
	.. unpack the object ..
	open("DF", O_WRONLY|O_CREAT|O_TRUNC|O_EXCL, 0666) = -1 EEXIST (File exists)

ie the problem is that we actually _have_ a test for this, but it's:

	if (errno == EISDIR && force) {

but if the directory already exists, we do that wrong.

Btw, this also shows a different problem: the symlink handling doesn't do any of this, so you cannot even force a directory to become a symlink.

I think both problems can be fixed by just moving the (EISDIR && force) test down to the "unlink()".

Will try.

		Linus