Le 01/11/2011 12:00, Axel Dörfler a écrit : > François Revol<revol@xxxxxxx> wrote: >> Le 31/10/2011 18:39, Axel Dörfler a écrit : >>>> + Transaction transaction(this, fRootNode->BlockNumber()); >>>> + fRootNode->WriteLockInTransaction(transaction); > [...] >>> Besides, this is all wrong: it should get the transaction from its >>> caller, or else this won't work right when called from the initializer. >> That I was wondering, but since in one place it has a transaction opened >> already and another not... > > On second thought (after another look at the code), the transaction should > actually automatically become a child transaction of the current transaction > (if any). So this should actually work fine -- I assume you have tested the > outcome of the patch, at least to found my understanding of the code I wrote > a while back with more substance? :-) > Well both versions seemed to work fine. François.