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

From: Linus Torvalds <torvalds@osdl.org>
Date: Jul 14, 2005, 16:08 UTC
Message-ID: <Pine.LNX.4.58.0507140904440.19183@g5.osdl.org>
In-Reply-To: <7v8y09g5sq.fsf@assigned-by-dhcp.cox.net>

On Thu, 14 Jul 2005, Junio C Hamano wrote:

Show 7 quoted lines

>
> The bisect search found that the commit
> 
> 	Make "git-checkout" create files with O_EXCL
> 
> makes the test t1005 fail.  But it is getting late so I give up
> to figuire this out tonight.

Ahh, thanks for noticing.

It says

	* expecting success: rm -f .git/index &&
	     rm -fr DF &&
	     mkdir DF &&
	     echo DF/DF >DF/DF &&
	     git-update-cache --add DF/DF &&
	     read_tree_twoway $treeDFDF $treeDF &&
	     git-ls-files --stage >DFDFcheck.out &&
	     diff -u DF.out DFDFcheck.out &&
	     check_cache_at DF clean && # different from pure 2-way
	     :
	100644 052efc3abbc31348f7abd34535b1953d38273257 3       DF
	100644 b90ea14b2dd74b6f377c10870b3757344bbe077c 1       DF/DF
	100644 b90ea14b2dd74b6f377c10870b3757344bbe077c 2       DF/DF
	Adding DF
	error: git-checkout-cache: unable to create file DF (File exists)
	Removing DF/DF
	100644 052efc3abbc31348f7abd34535b1953d38273257 0       DF
	DF: dirty
	* FAIL 24: DF vs DF/DF case test (#2) rm -f .git/index &&

Which is strange. If the path already exists, the "lstat(path, &st)" should have returned 0 in checkout_entry (which is the only caller of "write_entry()"), and that code does an "unlink(path)" before it calls write-entry. So I was sure O_EXCL wouldn't make any difference, but I'm obviously wrong, wrong, wrong.

I'll strace the dang thing.

		Linus