Currently, Plugin API doesn't support loading(?)/unloading when plugin is async. Async support for both plugins and onUnload function will fix the plugin unloading, and might make it easier to handle some asynchronous situations.
Revite Async Support Patch
I already made a quick patch about it, but since I'm not familiar with Revite's technologies, I thought it would be better to both ask for the feature if it's in scope, and get a review about the patch. If it's considered to be added and patch looks fine, I can create a PR.
```diff
diff --git a/src/mobx/stores/Plugins.ts b/src/mobx/stores/Plugins.ts
index e6b3bbb3..eb4c50a2 100644
--- a/src/mobx/stores/Plugins.ts
+++ b/src/mobx/stores/Plugins.ts
@@ -58,7 +58,7 @@ type Plugin = {
type Instance = {
format: 1;
- onUnload?: () => void;
+ onUnload?: () => Promise | void;
};
// Example plugin:
@@ -140,7 +140,8 @@ export default class Plugins implements Store, Persistent {
init() {
if (!this.state.experiments.isEnabled("plugins")) return;
this.plugins.forEach(
- ({ namespace, id, enabled }) => enabled && this.load(namespace, id),
+ async ({ namespace, id, enabled }) =>
+ enabled && (await this.load(namespace, id)),
);
}
@@ -148,19 +149,19 @@ export default class Plugins implements Store, Persistent {
* Add a plugin
* @param plugin Plugin Manifest
*/
- add(plugin: Plugin) {
+ async add(plugin: Plugin) {
if (!this.state.experiments.isEnabled("plugins"))
return console.error("Enable plugins in experiments!");
const loaded = this.getInstance(plugin);
if (loaded) {
- this.unload(plugin.namespace, plugin.id);
+ await this.unload(plugin.namespace, plugin.id);
}
this.plugins.set(`${plugin.namespace}/${plugin.id}`, plugin);
if (typeof plugin.enabled === "undefined" || plugin) {
- this.load(plugin.namespace, plugin.id);
+ await this.load(plugin.namespace, plugin.id);
}
}
@@ -169,8 +170,8 @@ export default class Plugins implements Store, Persistent {
* @param namespace Plugin Namespace
* @param id Plugin Id
*/
- remove(namespace: string, id: string) {
- this.unload(namespace, id);
+ async remove(namespace: string, id: string) {
+ await this.unload(namespace, id);
this.plugins.delete(`${namespace}/${id}`);
}
@@ -179,14 +180,14 @@ export default class Plugins implements Store, Persistent {
* @param namespace Plugin Namespace
* @param id Plugin Id
*/
- load(namespace: string, id: string) {
+ async load(namespace: string, id: string) {
const plugin = this.get(namespace, id);
if (!plugin) throw "Unknown plugin!";
try {
const ns = `${plugin.namespace}/${plugin.id}`;
- const instance: Instance = eval(plugin.entrypoint)();
+ const instance: Instance = await eval(plugin.entrypoint)();
this.instances.set(ns, {
...instance,
format: plugin.format,
@@ -207,14 +208,17 @@ export default class Plugins implements Store, Persistent {
* @param namespace Plugin Namespace
* @param id Plugin Id
*/
- unload(namespace: string, id: string) {
+ async unload(namespace: string, id: string) {
const plugin = this.get(namespace, id);
if (!plugin) throw "Unknown plugin!";
const ns = `${plugin.namespace}/${plugin.id}`;
const loaded = this.getInstance(plugin);
if (loaded) {
- loaded.onUnload?.();
+ const entrypoint = loaded.onUnload;
+ if (entrypoint?.constructor.name === "AsyncFunction")
+ await entrypoint?.();
+ else loaded.onUnload?.();
this.plugins.set(ns, {
...plugin,
enabled: false,
```
We're currently rewriting the frontend and the plugin API will likely change significantly in the process; I'll transfer this so we know to keep it in mind when we get to that point
We're currently rewriting the frontend and the plugin API will likely change significantly in the process; I'll transfer this so we know to keep it in mind when we get to that point
That's great, but as far as I know, there's no ETA for both new client and plugin API. So maybe adding support to Revite as a quick patch a great idea, until the new client arrives?
There aren't any ETAs, correct. We're currently only doing important bug fixes for Revite/small tweaks that don't carry much risk - in my opinion, something like this would need more testing than we intend to do for Revite.
TOML frontmatter on top of JS
This will be executed within a function and the return will be a set of event handlers to be registered, and later cleaned up.
I don't know if this is an already accepted proposal, but here's my recommendations:
Instead of frontmatter styled plugin info, why not using a JSDoc solution, like TamperMonkey and such does, or having them to configure programmatically?
AFAIK, top level returns aren't a thing. What about export { onSomeHandler: () => {} }, or having a function or class type (which will return the hooks and other useful data like plugin name etc.) and exporting it, similar to this.
What do you want to see?
Currently, Plugin API doesn't support loading(?)/unloading when plugin is async. Async support for both plugins and
onUnload
function will fix the plugin unloading, and might make it easier to handle some asynchronous situations.Revite Async Support Patch
I already made a quick patch about it, but since I'm not familiar with Revite's technologies, I thought it would be better to both ask for the feature if it's in scope, and get a review about the patch. If it's considered to be added and patch looks fine, I can create a PR. ```diff diff --git a/src/mobx/stores/Plugins.ts b/src/mobx/stores/Plugins.ts index e6b3bbb3..eb4c50a2 100644 --- a/src/mobx/stores/Plugins.ts +++ b/src/mobx/stores/Plugins.ts @@ -58,7 +58,7 @@ type Plugin = { type Instance = { format: 1; - onUnload?: () => void; + onUnload?: () => PromiseWe're currently rewriting the frontend and the plugin API will likely change significantly in the process; I'll transfer this so we know to keep it in mind when we get to that point
That's great, but as far as I know, there's no ETA for both new client and plugin API. So maybe adding support to Revite as a quick patch a great idea, until the new client arrives?
There aren't any ETAs, correct. We're currently only doing important bug fixes for Revite/small tweaks that don't carry much risk - in my opinion, something like this would need more testing than we intend to do for Revite.
Hijacking this issue, proposal for new plugin format:
TOML frontmatter on top of JS This will be executed within a function and the return will be a set of event handlers to be registered, and later cleaned up.
I don't know if this is an already accepted proposal, but here's my recommendations:
Instead of frontmatter styled plugin info, why not using a JSDoc solution, like TamperMonkey and such does, or having them to configure programmatically?
AFAIK, top level returns aren't a thing. What about
export { onSomeHandler: () => {} }
, or having a function or class type (which will return the hooks and other useful data like plugin name etc.) and exporting it, similar to this.wrt (2), the idea would be that the plugin is implicitly wrapped in a function when evaluated, and then we can take the return at the end