Ability to load pre-compiled modules

[hirofa]

Ability to load pre-compiled modules

[andrieshiemstra]

I instantiate only one runtime and create a new realm on every new call (for security reasons and to keep all the calls isolated & stateless).

I'm not sure how the cached annotation would work here as the parameters would include a realm.

i'll look into this later but to answer your question, you can create a custom key in that annotation

#[cached(
    key = "String",
    size = 10000,
    time = 3600,
    result = true,
    convert = r#"{ format!("{}://{}/{:?}", scheme, domain_port, session.get_env()) }"#
)]
async fn get_info(
    scheme: &str,
    domain_port: &str,
    session: GrecoSession,
) -> Result<OpenApi, JsError> {
   todo!()
}

[andrieshiemstra]

so the advantage would be the ability to load a compiled module from disk? for faster startup?

allthough i gues there are practical uses for that i would like to point out that while you say "loading and compiling on every invocation." modules are actually kept cached in every JSContext (Realm) in a loaded_modules list (see quickjs.c) and once compiled they are not compiled again unless you init a new Realm..

Havinbg said that i am still very much open to the idea of a CompiledModuleLoader (or as an option in scriptmodueloader)

[SreeniIO]

In my case, a new context is created for every call. Hence, this would be of great help. Thanks

[andrieshiemstra]

ok, i slammed the code in without testing or even trying to run it.. (so it needs some QC :)) but you could test to see if it works..

[andrieshiemstra]

todo: make from_bytecode accept a &Vec so we can alter CompiledModuleLoader to return a &Vec so you don't have to clone or create thge buffer every time you eval a compiled module

[andrieshiemstra]

todo, check if module (eval and execution) still works if we drop the buffer

[SreeniIO]

Great. Will test it out and let you know.

[SreeniIO]

Made the below changes

1. Pass the current realm ref to CompiledModuleLoader functions (required to compile the source when loading for the first time)
2. Return the bytes as Arc<Vec> instead of &Vec

[andrieshiemstra]

this issue does make me wonder if we're not overcomplicating things... we could also just make an option in the module loading logic to allways first compile modules to bytecode and cache that so it gets reused in every new realm (do you create a new realm for every new call or a new runtime?)

using a simple #[cached(size=100)] annotated function we could probably get the same performance benefit without the extra CompiledModuleLoader

[SreeniIO]

I instantiate only one runtime and create a new realm on every new call (for security reasons and to keep all the calls isolated & stateless).

I'm not sure how the cached annotation would work here as the parameters would include a realm.

[SreeniIO]

Thanks, closing this for now... as we have CompiledModuleLoader in place.