- From: Jamie Lokier <jamie@shareable.org>
- Date: Mon, 17 Nov 2008 02:38:44 +0000
- To: Henrik Nordstrom <henrik@henriknordstrom.net>
- Cc: Werner Baumann <werner.baumann@onlinehome.de>, Yves Lafon <ylafon@w3.org>, ietf-http-wg@w3.org
Henrik Nordstrom wrote: > > Sensible applications, on unix, replace files by renaming over the > > original. When a file is served, it is always the old contents or the > > new contents, never a mix. > > Which would render a competely new weak ETag in Apache (on Unix), even > if the file timestamp and size is identical. Nearly always true, yes. It's good enough but not cast iron. It's is not guaranteed even on unix. Firstly some filesystems have synthetic inode numbers, which are generated at the time of stat() call. Secondly, if you modified the file *twice* in the same second using rename-over, the original inode number can be recycled on any unix filesystem. Only by using the inode generation number (which NFS uses) can you avoid this, and that's not generally available to user processes. -- Jamie
Received on Monday, 17 November 2008 02:39:22 UTC