All righty, folks ... anyone looking to build a database backend of their own, please check out the latest batch of changes in the git repository. It basically works like this: * database.c has been renamed to database_bdb.c * database.h now basically declares an API for storage backends to use * All functions beginning with cdb_ were renamed to bdb_ * The cdb_ functions are now just pointers * During initialization, the cdb_ functions are assigned to their corresponding bdb_ functions I've tested it a bit and everything seems to be working just as it was before. I did the whole thing with function pointer assignment because I didn't want to write a giant batch of wrapper functions to marshal calls to the correct backend and then require one more step up the stack for every call. Since only one backend will ever be used at a time, we load the pointers at initialization time and then everything works exactly as it did before. Here's where I expect to go with this next: * A new server/database.c file will come into existence, and it will handle the binding of the selected backend * Backends will move to a separate part of the directory tree, probably server/modules. They will need to have a different initialization convention because the data store is initialized long before server extension modules are initialized.
* Backends that are incompatible with a particular host (for example, LMDB on 32-bit) will simply be skipped in the build. * We'll have to add another function to scan the data directory and detect what backend was used to write any existing database that might already be there. If an existing database is present, we will automatically load the correct backend. * Where no existing database is present, we will choose one in order of preference. A command line switch will of course be available to override the server's decision. Does that sound acceptable to you all? This should make it possible to bring experimental backends into the tree without foisting them upon the user community.
