Brian Aker wrote: > Hi! > > On Mar 9, 2009, at 9:59 AM, Monty Taylor wrote: > >> before those propagate to somewhere bad. Of course, to do this properly, >> we need to get all of those new'd pointers in to shared_ptr<>'s so that >> we don't leak by doing that... > > This is my main concern with exceptions... we need to make sure we have > a way to test failure and to make sure that all code is cleaned up to > handle this behavior well (as well as test for performance regression).
Agree... and I think we're a ways away from being able to have a general top-of-thread bad_alloc trap, since we have too many non-owned pointers running around that won't get properly destructed when the stack goes away. > In concept I like it, we just need to do it right. Agree. > On a similar note, Stewart's use of abort() has made me start to wonder > if we need to either make more use of signals to handle shutdown, or if > we should un-require them for the shutdown process. The advantage of not > using them is simple... if we ever need to run on OS'es without them it > makes the porting work simpler. Which OS'es don't have? Honestly - abort() is essentially just throwing an exception without throwing an exception - same overall issues and problems. So would sending more signals. Main problem is that we still need to get more RAII going on with our objects so that out-of-band errors, whether they be proper exceptions or abort() can still actually clean up after themselves... Once it is cleaned up though, we should get rid of the C-workarounds that implement exceptions by hand. Monty _______________________________________________ Mailing list: https://launchpad.net/~drizzle-discuss Post to : [email protected] Unsubscribe : https://launchpad.net/~drizzle-discuss More help : https://help.launchpad.net/ListHelp
