Clarification of the above commit -- because it's important. When traversing an entire table (as opposed to just fetching one row for which you already know the key) the backend API now returns both a key and a value. This is done by returning a "struct cdbkeyval" which simply contains a pair of "struct cdbdata", one for the key and one for the value. Yes, my discovery that you've been able to return a struct by value from a function in C for the last couple of decades has already reflected itself in Citadel. :) All of this is based on a *new* rule in the backend interface: when you fetch an item from the database, either with cdb_fetch() or cdb_next_item(), the caller does NOT become the owner of the memory it points to. This is how Berkeley DB operates by default (and I changed it back to this mode over the past week) and it's the *only* way LMDB can operate.
Why all the fuss? Well, let me tell you; thanks for asking! ctdldump and ctdlload currently call Berkeley DB directly. This is non-awesome. It means that when we add more backends we would have to write new versions of ctdldump and ctdlload for each one. Not gonna do it. ctdldump and ctdlload will be modified to use the same backend interface that citserver uses. Then we'll be able to dump and load in any format on any machine. But ctdldump needs to know the keys, not just the values. And so, cdb_next_item() now returns both.
