| 30 Sep 2020 |
 m4b | oh but that won't work either hah; because then the client can't add data, since it only has the type erased object, heh | 00:08:31 |
 cwfitzgerald | Why both the Arc and the box | 00:09:02 |
| m4b | arc makes it cloneable and Send | 00:09:31 |
| cwfitzgerald | Huh? | 00:09:45 |
| cwfitzgerald | I'm saying Vec Arc Mutex dyn Plugin | 00:10:01 |
| m4b | oh yea, sorry, that would work; same issue i noted above though, so won't work | 00:10:31 |
| cwfitzgerald | Where's the add data coming from | 00:12:01 |
| m4b | i already told you, the client needs to add whatever it needs to the plugin in order it to render stuff every frame | 00:12:35 |
| m4b | in particular case, this is the prepare method, which passes a boxed imgui callback, for example, which is then consumed internally by imgui to make it ready to render | 00:13:06 |
| m4b | you can't add e.g., a method like fn add_data<T>(d: T) to the trait Plugin right, because it's no longer object safe then | 00:13:41 |
| m4b | you can get around this issue by using something like downcast-rs but that's a heavy hammer | 00:14:16 |
| m4b | it's also too java-android-ui-y for me :P | 00:14:35 |
| cwfitzgerald | so I think you're arriving at the main problem with trying to abstract like this | 00:15:58 |
| m4b | well i think it's perfect storm of trait objects, &mut renderpass<'a>, and mutexes | 00:16:31 |
| cwfitzgerald | As I'm not sure how useful of an abstraction this is, unless you somehow need to accept an arbitrary Plugin from your user | 00:16:50 |
| m4b | well it's extremely useful to me? my 3 applications no longer recompile a 40 second crate using this method, so it's pretty great | 00:17:41 |
| m4b | but yes the idea would be i could add other plugins later on, without hardcoding the plugin or ifdef-ing it in the renderer itself | 00:18:26 |
| m4b | i'm just sad i can't get &mut Renderpass working :/ | 00:18:36 |
| m4b | i think the only way to do it is have Vec<Arc<Mutex<dyn Plugin>>>, lock in loop, render, etc.; then have the user downcast-rs to add whatever data they need (if i want the &mut Renderpass); otherwise i've become convinced the current version internalizing the arc + mutex is actually a better user api | 00:19:46 |
| cwfitzgerald | what is downcast-rs | 00:20:14 |
| m4b | https://crates.io/crates/downcast-rs | 00:20:31 |
| cwfitzgerald | seems a bit heavy handed to me, vs just dyn Any | 00:21:28 |
| m4b | dyn Any is an upcast? downcasting turns an erased/abstract object into it's concrete type | 00:22:09 |
| cwfitzgerald | yeah, Any has downcast_ref and whatev | 00:22:37 |
| m4b | it's like the entire android ui api, (MyActualButton) new Button(foo) haha | 00:22:41 |
| m4b | ok well anyway, it doesn't seem like there's a nice solution to this problem | 00:23:42 |
| m4b | i wonder what bevy does for a wgpu renderer plugin, does anyone know? | 00:23:54 |
| m4b | * i wonder what bevy does for a wgpu renderer plugin, does anyone know? | 00:24:17 |
| cwfitzgerald | I feel like it might just have a proper rendergraph | 00:25:35 |
| m4b | i don't know what that means? if i want to use imgui in bevy, unless it supports it inside it's own renderer, how wouldn't they have the exact problem i just described? | 00:26:46 |
