site_archiver@lists.apple.com Delivered-To: darwin-dev@lists.apple.com _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev@lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.appl... And the culprit is probably the __put_page function from hash_page.c which is not un-swapping a buffer after writing it to disk. So this leads to a corrupted data base in memory and the next call to a sync is saving corrupted data (header is correct but the rest of the file is incorrect). Adding some un-swapping code after the _write call seems to make things run fine on i386 with a BIG_ENDIAN data base. But I won't bet my right hand on this fix for the moment. AFAIK, the bug is also impacting the PPC platform because as soon as the lorder is not the same as BYTE_ORDER, the swap problem is going to happen. Sadly, this means that if you don't use a fix, hash data base saved by db on disks are not cross-endian compatible. I filed a bug report on the FreeBSD site since this is where the original source code is coming from. This email sent to site_archiver@lists.apple.com