ENIGMA Forums
Contributing to ENIGMA => Developing ENIGMA => Topic started by: Josh @ Dreamland on March 28, 2014, 09:49:23 am
-
ENIGMA (Specifically EDL) is full of stateful functions: functions which operate based on a global state. Obvious examples include file_find_first/file_find_next. It also contains functions which operate on resources not owned by ENIGMA. Examples of these include file_bin_open, surface_create, and the list goes on. Still more functions change the overall game state in ways that are conducive to leaks, such as ds_list_create, ds_map_create, ds_grid_create, sprite_add, sound_add, background_add...
You might have figured it out by now, but these functions all have one thing in common: they don't agree with game_save. In Game Maker, game_save and game_load can dump all state information and re-load it in a matter of milliseconds. But this excludes work done by the functions above (or at least the majority of them). I can have ENIGMA generate state dumps for objects, but I can't have it generate them for all functions. Least of all surface_create.
This is problematic because game_save is a fantastic placeholder for custom save mechanisms, and its emulator-like dumping is great for debugging. In fact, that's the major reason I am interested in it right now. Rusky and I were discussing a crackpot lunatic's great idea for game debugging, and for some reason, I'm dumb enough to want to give integrating it into ENIGMA a try. But I can't do that (despite Rusky's insistence otherwise) without a working game_save/game_load.
The easiest way I can devise to handle the situation is to create a namespace for state restoration functions. So in the header that declares file_find_first, you'd see something like this:
string file_find_first(string pattern, int attr);
namespace enigma {
namespace dump_state { vector<char> file_find(); }
namespace load_state { size_t file_find(char*); }
}
The compiler would then include a call to everything in enigma::dump_state in game_save, and a call to everything in enigma::load_state in game_load.
The call to dump_state::*() returns a vector of bytes to write to the save file. The call to load_state::*() accepts a pointer to file content (ideally memory-mapped), reads in the state information in its own dump format, and then returns the number of bytes read. Thus, it's critical that each function write only what it will read and reads everything it might have written. Otherwise, it will ruin all the other state loads.
Now, in the case of file_find_*, it's pretty easy to dump the current filename and just iterate back to it. But in the case of surface_create, dumping all the surfaces to a file might be a large, cumbersome task that a user would want to disable. And in the case of sprite_add, keeping track of all added sprites is more code than some users will want in their game. That said, I think it would be frugal to make sure this logging code can also be toggled off, with a preprocessor. Later on, I'm not opposed to a Game Load event.
With a little luck, game_save will be powerful enough to do what I am planning. If not, we may need to use some more preprocessors and include code to keep track of state changes in a more minimal fashion. I'll worry about that later. For now, game_save is something isolated enough to be concerned with.