On Sun, 2015-03-01 at 22:37 +0100, Samuel Thibault wrote: > Hello, > > With the current hurd libc and tdb built with the attached patches, > making it use whole-file locking on platforms which don't support record > locking, tdb works fine, ./bin/tdbtorture -n 10 says "OK", it takes > 3m to complete, which is not so bad compared to the 30s it takes with > proper record locking in Linux. > > Again, we are fine with a slow tdb: we don't target supporting samba > servers or such, we just want to have tdb basically working, instead > of manually removing it as a dependency from various packages, only to > have to put that back when we have record locking working properly (that > looks like unnecessary work to me). Samba is currently a blocker for > various packages, including mplayer, vlc, kde runtime, etc. >
Samuel, I don't see how you can declare this safe when the patch edits out whole chunks of the testsuite? If this worked with the tests still in place, then perhaps this might be reasonable, but those tests were added for a reason, and to skip them simply risks creating and then debuging (at best) deadlocks in many other applications, perhaps quite divorced from tdb. This is also the wrong place to propose such patches. Please get them accepted (or more likely, clearly rejected) upstream on [email protected] as this is clearly not a simple packaging matter. Or, fix your target OS to have fcntl() locks, like every other posix system we have come across. This would be far, far more productive. Thanks, Andrew Bartlett -- Andrew Bartlett http://samba.org/~abartlet/ Authentication Developer, Samba Team http://samba.org Samba Developer, Catalyst IT http://catalyst.net.nz/services/samba -- To UNSUBSCRIBE, email to [email protected] with a subject of "unsubscribe". Trouble? Contact [email protected] Archive: https://lists.debian.org/1425372392.15962.26.camel@jesse
