On 3/16/08, Rahul Akolkar wrote:
On 3/15/07, [EMAIL PROTECTED] <[EMAIL PROTECTED]> wrote: > Author: jochen > Date: Thu Mar 15 15:02:46 2007 > New Revision: 518770 > > URL: http://svn.apache.org/viewvc?view=rev&rev=518770 > Log: > Made the FileCleaningTracker serializable. > This is required by commons-fileupload, because the > DiskFileItem's may be part of the HTTP session while > still carrying a reference to the tracker. > <snip/>
I think this particular solution is sub-optimal in a couple of aspects:
* The instances aren't really serializable so its a bit of a truth in advertising infringement * Its introducing a tighter coupling between the next releases of [io] and [fileupload], which is fine if we can buy that there is evidence that the change would be helpful to [io] in isolation.
I think identical behavior (for what this is worth) can be obtained by reverting this commit and having a transient lazily-initialized FileCleaningTracker in DiskFileItem, which addresses the two bullets above, and the NotSerializableExceptions that you may otherwise witness at container shutdown.
WDYT?
I haven't really though about the fileupload part, but you have convinced me that this must be addressed by fileupload and not by io. I have therefore reverted my patch in the 1.3 branch. I will also revert the same patch in the trunk after the 1.3.2 release is out when I merge the release related changes in. Jochen --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
