This is a step towards making replication a library, such that the 5.6 rpl-lib can be plugged into the 5.5 core [WL#5675]. When LOAD DATA INFILE reads a file, it writes the parts of the file in chunks to the binary log. Currently, it stores information about the LOAD DATA statement, including the THD object, in a special field in struct IO_CACHE*. Then, the IO_CACHE is configured so that it calls log_loaded_block before a new block of data is loaded into memory. The IO_CACHE is passed as the only argument to log_loaded_block. We should not maintain this kind of spaghetti code, even less build interfaces around it. The most obvious problem is that the generic datatype IO_CACHE now contains information specific to LOAD DATA INFILE statements. That's just bad design. If we would put IO_CACHE in a stable interface, it would mean that (1) we cannot fix the bad design of IO_CACHE; (2) we assume IO_CACHE is stable; (3) we assume IO_CACHE is used to load data; but we may want to change to something else in the future. So we should change the interface to: Binlog::log_loaded_block(THD *thd, int length, unsigned char *data); This step is not required for the prototype of the feature to link 5.6 rpl-lib to 5.5 core, but it is required before we announce the interface as stable.