sqlite memory leak on failed open

[polesapart]

The sqlite::database constructor has a potential memory leak if opening the database fails for whatever reason:

database(std::u16string const & db_name): _db(nullptr) {
			sqlite3* tmp = nullptr;
			auto ret = sqlite3_open16(db_name.data(), &tmp);
			if(ret != SQLITE_OK) exceptions::throw_sqlite_error(ret);

			_db = std::shared_ptr<sqlite3>(tmp, [=](sqlite3* ptr) { sqlite3_close_v2(ptr); }); // this will close the connection eventually when no longer needed.

			//_db.reset(tmp, sqlite3_close); // alternative close. (faster?)
		}

accordingly to https://www.sqlite.org/c3ref/close.html one must close the handle even if open returned an error. The above can be fixed by moving the fail check and the exception throwing after defining the _db shared_ptr, so it will call sqlite3_close_v2 at stack unwinding. One can prefer to do an explicit close instead, which is cheaper but more verbose. I took the 1st approach, below:

database(std::u16string const & db_name): _db(nullptr) {
			sqlite3* tmp = nullptr;
			auto ret = sqlite3_open16(db_name.data(), &tmp);
			_db = std::shared_ptr<sqlite3>(tmp, [=](sqlite3* ptr) { sqlite3_close_v2(ptr); }); // this will close the connection eventually when no longer needed.
   			if(ret != SQLITE_OK) exceptions::throw_sqlite_error(ret);
			//_db.reset(tmp, sqlite3_close); // alternative close. (faster?)
		}

[aminroosta]

@polesapart As always thanks :-)

This should be fixed here aminroosta@b867b5e

[polesapart]

Valgrind isn't complaining anymore, so I think we can close this :-) 👍