Expand description
A library for doing live-reloading game development.
This is inspired by the article “Interactive Programming in C” by Chris Wellons, and the video “Loading Game Code Dynamically” from Handmade Hero by Casey Muratori.
The general idea is that your main host program is a wrapper around a dynamic library that does all the interesting work of your game. This means that you can simply reload the library while the game is still running, and have your game update live. As a consequence however, you can’t have any global state in your library, everything must be owned by the host in order to avoid getting unloaded with the library.
In order to call back into the host program, you specify a Host
type
containing function pointers to services in the host program. This Host
struct will be passed along with your program state. The Host
type should
always be defined in a separate module so that both the host program and the
reloadable library use a consistent host type. The current recommended
method is to define your host type in its own module. Then use that module
from both the host and the reloadable library. If your project organization
puts the common module in a parent, you can always use the #[path=...]
meta on the module, for example:
#[path="../host_api.rs"]
mod host_api;
While designing your host and library, keep in mind the role of the two communication types:
-
State is isolated to the reloadable library, the main program knows nothing about it except for its size so that it can keep it allocated. This lets you change the contents of the State struct without restarting the whole program. This is intended to handle the state of the game world, independent of the host program.
-
Host is defined by the host program and the layout is visible to both sides of the bridge. This means that it has to remain the same during a run of the game engine. This should hold resources that can only be produced by the host program, and function pointers to services that can only be provided by the host program. (Anything that requires global state like system allocators, graphics contexts, input handling, etc etc.)
See the Host Example and Library Example sections for instructions on how to build a reloadable application.
§Host Example
A program that hosts a reloadable library will need to load the library, and
then periodically reload it. The Reloadable
automatically installs a
filesystem watcher for you so that it knows when the library file has been
updated or replaced, and the reload
method will only actually perform
a reload if the file has changed. The core of your main loop will therefore
usually look something like this:
use std::thread;
mod host_api {
// This should always be in a different file
pub struct HostApi {
pub print: fn(&str),
}
}
use host_api::HostApi;
type App = live_reload::Reloadable<HostApi>;
fn print(msg: &str) {
print!("{}", msg);
}
fn main() {
let mut prog = App::new(
"target/debug/libreload.dylib",
HostApi { print: print },
).expect("Should successfully load");
'main: loop {
if prog.update() == live_reload::ShouldQuit::Yes {
break 'main;
}
prog.reload().expect("Should successfully reload");
}
}
§Library Example
A live-reloadable library needs to register its entry-points so that the
host program can find them. The live_reload!
macro lets you do this
conveniently.
The lifecycle of your reloadable library will happen in a few stages:
init
gets called at the very beginning of the program, when the host starts for the first time.reload
gets called on each library load, including the first time. This should be usually empty, but when you’re in development, you might want to reset things here, or migrate data, or things like that. The pointer you’re passed will refer to the same struct that you had when the previous library was unloaded, so it might not be properly initialized. You should try to make your struct be#[repr(C)]
, and only add members at the end to minimize the problems of reloading.update
gets called at the host program’s discretion. You’ll probably end up calling this once per frame. In addition to doing whatever work you were interested in,update
also returns a value indicating whether the host program should quit.unload
gets called before a library unloads. This will probably be empty even more often thanreload
, but you might need it for some debugging or data migration purpose.deinit
gets called when the host program is actually shutting down–it’s called on the drop of theReloadable
.
Here’s an example of a live-reloadable library that handles a counter.
#[macro_use] extern crate live_reload;
use live_reload::ShouldQuit;
mod host_api {
// This should always be in a different file.
pub struct Host {
pub print: fn(&str),
}
}
use host_api::Host;
live_reload! {
host: Host;
state: State;
init: my_init;
reload: my_reload;
update: my_update;
unload: my_unload;
deinit: my_deinit;
}
struct State {
counter: u64,
}
fn my_init(host: &mut Host, state: &mut State) {
state.counter = 0;
(host.print)("Init! Counter: 0.");
}
fn my_reload(host: &mut Host, state: &mut State) {
(host.print)(&format!("Reloaded at {}.", state.counter));
}
fn my_update(host: &mut Host, state: &mut State) -> ShouldQuit {
state.counter += 1;
(host.print)(&format!("Counter: {}.", state.counter));
ShouldQuit::No
}
fn my_unload(host: &mut Host, state: &mut State) {
(host.print)(&format!("Unloaded at {}.", state.counter));
}
fn my_deinit(host: &mut Host, state: &mut State) {
(host.print)(&format!("Goodbye! Reached a final value of {}.", state.counter));
}
Modules§
- internals
- Exported for compilation reasons but not useful, only look if you’re curious.
Macros§
- live_
reload - Declare the API functions for a live-reloadable library.
Structs§
- Reloadable
- A
Reloadable
represents a handle to library that can be live reloaded.
Enums§
- Error
- The errors that can occur while working with a
Reloadable
object. - Should
Quit - Should the main program quit? More self-documenting than a boolean!