David Glasser wrote:
> OK, sure, so this prevents "two files with the same brand-new
> contents", but does this really prevent "two files with the same
> contents, which already existed in a previous revision"? ie, both
> files have the same rep as a *previously-cached* third rep.
Oh, good point. So we need some kind of intra-txn uniquifier as well. You had
mentioned a 'next-id' file in the txn directory, but I'm wondering if we can't
just use a in-memory counter of some kind, perhaps something. The counter
doesn't need to be per-transaction, so we could through it in
fs_fs_sharded_data_t, but that would require protecting it with a mutex. Still
better than the I/O to go update a file on disk, I'd think.
But this is where my knowledge of the FSFS internals gets a bit rusty. Any
suggestions?
-Hyrum
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=462&dsMessageId=981160
Received on 2008-12-08 15:45:10 CET