PkmnLib_rs/src/script_implementations/wasm/script_resolver.rs

use std::fmt::{Debug, Formatter};
use std::mem::{align_of, size_of};
use std::sync::{Arc, Weak};

use hashbrown::{HashMap, HashSet};
use parking_lot::lock_api::{MappedRwLockReadGuard, RwLockReadGuard};
use parking_lot::{RawRwLock, RwLock};
use unique_type_id::UniqueTypeId;
use wasmer::{
    Cranelift, Exports, Extern, Features, Function, ImportObject, Instance, Memory, Module, NativeFunc, Store,
    Universal, Value, WasmerEnv,
};

use crate::dynamic_data::{ItemScript, Script, ScriptResolver};
use crate::script_implementations::wasm::export_registry::register_webassembly_funcs;
use crate::script_implementations::wasm::extern_ref::ExternRef;
use crate::script_implementations::wasm::script::WebAssemblyScript;
use crate::script_implementations::wasm::script_function_cache::ScriptFunctionCache;
use crate::script_implementations::wasm::temp_wasm_allocator::{AllocatedObject, TempWasmAllocator};
use crate::script_implementations::wasm::WebAssemblyScriptCapabilities;
use crate::static_data::Item;
use crate::{PkmnResult, ScriptCategory, StringKey};

/// A WebAssembly script resolver implements the dynamic scripts functionality with WebAssembly.
pub struct WebAssemblyScriptResolver {
    /// The global state storage of WASM.
    store: Store,
    /// The WASM modules we have loaded.
    modules: Vec<Module>,
    /// Our currently loaded WASM instances. Empty until finalize() is called, after which the loaded modules get turned
    /// into actual instances.
    instances: Vec<Instance>,

    /// This is the WASM function to load a script.
    load_script_fn: Option<NativeFunc<(u8, ExternRef<StringKey>), u32>>,

    /// Script capabilities tell us which functions are implemented on a given script. This allows us to skip unneeded
    /// WASM calls.
    script_capabilities: RwLock<HashMap<ScriptCapabilitiesKey, Arc<HashSet<WebAssemblyScriptCapabilities>>>>,

    /// The data for use in the scripting function calls.
    environment_data: Arc<WebAssemblyEnvironmentData>,
}

/// This struct allows us to index a hashmap with both a category and name of a script.
#[derive(Debug, Clone, Eq, PartialEq, Hash)]
struct ScriptCapabilitiesKey {
    /// The category for the script we're looking for capabilities for.
    category: ScriptCategory,
    /// The name of the script we're looking for capabilities for.
    script_key: StringKey,
}

impl WebAssemblyScriptResolver {
    /// Instantiates a new WebAssemblyScriptResolver.
    pub fn new() -> Box<WebAssemblyScriptResolver> {
        let config = Cranelift::default();
        let mut features = Features::new();
        features.multi_value = true;
        features.reference_types = true;
        let universal = Universal::new(config).features(features);
        let engine = universal.engine();
        let store = Store::new(&engine);
        let s = Self {
            store,
            modules: Default::default(),
            instances: Default::default(),
            load_script_fn: None,
            script_capabilities: Default::default(),
            environment_data: Arc::new(Default::default()),
        };
        Box::new(s)
    }

    /// Load a compiled WASM module.
    pub fn load_wasm_from_bytes(&mut self, bytes: &[u8]) {
        // FIXME: Error handling
        let module = Module::new(&self.store, bytes).unwrap();
        self.modules.push(module);
    }

    /// Initialise all the data we need.
    pub fn finalize(&mut self) {
        let mut imports = ImportObject::new();
        let mut exports = Exports::new();

        let env = WebAssemblyEnv::new(Arc::downgrade(&self.environment_data));
        register_webassembly_funcs(&mut exports, &self.store, env);
        imports.register("env", exports);

        for module in &self.modules {
            for import in module.imports() {
                if imports.get_export("env", import.name()).is_none() {
                    println!(
                        "\x1b[91mMissing import: \"{}\" with type: {:?} \x1b[0m",
                        import.name(),
                        import.ty()
                    );
                }
            }

            let instance = Instance::new(module, &imports).unwrap();
            let exports = &instance.exports;

            let init_fn = exports.get_extern("_init");
            if let Some(Extern::Function(init_fn)) = init_fn {
                init_fn.call(&[]).unwrap();
            }

            let mut exported_functions = self.environment_data.exported_functions.write();
            for export in exports.iter() {
                match export.1 {
                    Extern::Function(f) => {
                        exported_functions.insert(export.0.as_str().into(), f.clone());
                    }
                    Extern::Memory(m) => {
                        let _ = self.environment_data.memory.write().insert(m.clone());
                    }
                    _ => {}
                }
            }
            if let Some(m) = &self.environment_data.memory.read().as_ref() {
                m.grow(32).unwrap();
            }
            if let Some(f) = exported_functions.get::<StringKey>(&"load_script".into()) {
                self.load_script_fn = Some(f.native().unwrap())
            }
            if let Some(f) = exported_functions.get::<StringKey>(&"allocate_mem".into()) {
                let _ = self
                    .environment_data
                    .allocate_mem_fn
                    .write()
                    .insert(f.native().unwrap());

                let temp_memory_slab = self.environment_data.allocate_mem(128, 1);
                let _ = self
                    .environment_data
                    .temp_allocator
                    .write()
                    .insert(TempWasmAllocator::new(temp_memory_slab.0, temp_memory_slab.1));
            }
            self.instances.push(instance);
        }
    }

    /// Gets the data passed to every function as environment data.
    pub fn environment_data(&self) -> &Arc<WebAssemblyEnvironmentData> {
        &self.environment_data
    }
}

impl ScriptResolver for WebAssemblyScriptResolver {
    fn load_script(
        &self,
        owner: *const u8,
        category: ScriptCategory,
        script_key: &StringKey,
    ) -> PkmnResult<Option<Arc<dyn Script>>> {
        let script = self
            .load_script_fn
            .as_ref()
            .unwrap()
            .call(category as u8, ExternRef::new_with_resolver(self, script_key))
            .unwrap();
        if script == 0 {
            return Ok(None);
        }

        let key = ScriptCapabilitiesKey {
            category,
            script_key: script_key.clone(),
        };

        if !self.script_capabilities.read().contains_key(&key) {
            let mut capabilities = HashSet::new();
            unsafe {
                if let Some(get_cap) = self
                    .environment_data
                    .exported_functions
                    .read()
                    .get::<StringKey>(&"get_script_capabilities".into())
                {
                    let res = get_cap.call(&[Value::I32(script as i32)]).unwrap();
                    let ptr = (self.environment_data.memory.read().as_ref().unwrap().data_ptr()
                        as *const WebAssemblyScriptCapabilities)
                        .offset(res[0].i32().unwrap() as isize);
                    let length = res[1].i32().unwrap() as usize;
                    for i in 0..length {
                        capabilities.insert(*ptr.add(i));
                    }
                }
            }
            self.script_capabilities
                .write()
                .insert(key.clone(), Arc::new(capabilities));
        }

        let read_guard = self.script_capabilities.read();
        let capabilities = read_guard.get(&key).unwrap();

        Ok(Some(Arc::new(WebAssemblyScript::new(
            owner as *mut u8,
            script,
            capabilities.clone(),
            Arc::downgrade(&self.environment_data),
            script_key.clone(),
        ))))
    }

    fn load_item_script(&self, _key: &Item) -> PkmnResult<Option<Arc<dyn ItemScript>>> {
        todo!()
    }
}

impl Debug for WebAssemblyScriptResolver {
    fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
        f.write_str("WebAssemblyScriptResolver")
    }
}

/// This data is what is passed to every function that requires access to the global runtime context.
#[derive(Default)]
pub struct WebAssemblyEnvironmentData {
    /// We currently have a hacky implementation of extern refs while we're waiting for ExternRef support to hit the
    /// wasm32-unknown-unknown target of Rust. As we don't want to pass raw memory pointers to WASM for security reasons,
    /// we instead keep track of all the data we've sent to WASM, and pass the ID of that data to WASM. This allows us
    /// to only operate on data we know WASM owns. We currently store this data in this continuous Vec, and give the index
    /// of the data as the ID.
    extern_ref_pointers: RwLock<Vec<*const u8>>,
    /// To make sure we send the same identifier to WASM when we send the same piece of data multiple times, we have a
    /// backwards lookup on extern_ref_pointers. This allows us to get the index for a given piece of data.
    extern_ref_pointers_lookup: RwLock<HashMap<ExternRefLookupKey, u32>>,
    /// As an added security measure on our extern refs, we keep track of the types of the extern ref data we've sent.
    /// This prevents illegal arbitrary memory operations, where we expect type X, but the actual type is Y, which would
    /// allow for modifying memory we might not want to. If we get a type mismatch, we will panic, preventing this.
    extern_ref_type_lookup: RwLock<HashSet<ExternRefLookupKey>>,

    /// Additional security for data slices passed to WASM.
    extern_vec_ref_lookup: RwLock<HashMap<u32, Vec<u32>>>,

    /// The memory inside of the WASM container.
    memory: RwLock<Option<Memory>>,

    /// This is a map of all the functions that WASM gives us.
    exported_functions: RwLock<HashMap<StringKey, Function>>,

    /// A cache of all script functions for faster calls.
    script_function_cache: ScriptFunctionCache,

    /// This is the WASM function to allocate memory inside the WASM container.
    allocate_mem_fn: RwLock<Option<NativeFunc<(u32, u32), u32>>>,

    /// An allocator for quick short lifetime allocations within WASM.
    temp_allocator: RwLock<Option<TempWasmAllocator>>,
}

/// A quick lookup so we can find the extern ref of the value.
#[derive(Clone, Eq, PartialEq, Hash)]
struct ExternRefLookupKey {
    /// The raw pointer to the data
    pub ptr: *const u8,
    /// Whether or not the reference is a Vec
    pub is_vec: bool,
    /// The unique identifier of the type.
    pub t: u64,
}

impl WebAssemblyEnvironmentData {
    /// This returns the memory of the WASM container.
    pub fn memory(&self) -> MappedRwLockReadGuard<'_, RawRwLock, Memory> {
        RwLockReadGuard::map(self.memory.read(), |a| a.as_ref().unwrap())
    }

    /// This returns the functions exported from WASM.
    pub fn exported_functions(&self) -> RwLockReadGuard<'_, RawRwLock, HashMap<StringKey, Function>> {
        self.exported_functions.read()
    }

    ///
    pub fn script_function_cache(&self) -> &ScriptFunctionCache {
        &self.script_function_cache
    }

    /// Allocates memory inside the WASM container with a given size and alignment. This memory is
    /// owned by WASM, and is how we can pass memory references that the host allocated to WASM.
    /// The return is a tuple containing both the actual pointer to the memory (usable by the host),
    /// and the WASM offset to the memory (usable by the client).
    pub fn allocate_mem(&self, size: u32, align: u32) -> (*mut u8, u32) {
        let wasm_ptr = self.allocate_mem_fn.read().as_ref().unwrap().call(size, align).unwrap();
        unsafe {
            (
                self.memory
                    .read()
                    .as_ref()
                    .unwrap()
                    .data_ptr()
                    .offset(wasm_ptr as isize),
                wasm_ptr,
            )
        }
    }

    /// Allocates memory inside the WASM container with a given size and alignment. This memory is
    /// owned by WASM, and is how we can pass memory references that the host allocated to WASM.
    /// The return is a tuple containing both the actual pointer to the memory (usable by the host),
    /// and the WASM offset to the memory (usable by the client).
    pub fn allocate_mem_typed<T>(&self) -> (*mut u8, u32) {
        let wasm_ptr = self
            .allocate_mem_fn
            .read()
            .as_ref()
            .unwrap()
            .call(size_of::<T>() as u32, align_of::<T>() as u32)
            .unwrap();
        unsafe {
            (
                self.memory
                    .read()
                    .as_ref()
                    .unwrap()
                    .data_ptr()
                    .offset(wasm_ptr as isize),
                wasm_ptr,
            )
        }
    }

    /// Allocate a piece of memory inside WASM with a very short lifespan. This is mainly used for
    /// rapid allocation of script function parameters, where WASM needs to write to a specific
    /// pointer.
    pub fn temp_allocate_mem_typed<T>(&self, value: T) -> AllocatedObject<T> {
        self.temp_allocator.read().as_ref().unwrap().alloc::<T>(value)
    }

    /// Get a numeric value from any given value. This is not a true Extern Ref from WASM, as this
    /// is not supported by our current WASM platform (Rust). Instead, this is simply a way to not
    /// have to send arbitrary pointer values back and forth with WASM. Only values WASM can actually
    /// access can be touched through this, and we ensure the value is the correct type. In the future,
    /// when extern refs get actually properly implemented at compile time we might want to get rid
    /// of this code.
    pub fn get_extern_ref_index<T: UniqueTypeId<u64> + ?Sized>(&self, value: &T) -> u32 {
        self.get_extern_ref_from_ptr(value as *const T as *const u8, T::id().0, false)
    }

    /// Get a numeric value from any given value. This is not a true Extern Ref from WASM, as this
    /// is not supported by our current WASM platform (Rust). Instead, this is simply a way to not
    /// have to send arbitrary pointer values back and forth with WASM. Only values WASM can actually
    /// access can be touched through this, and we ensure the value is the correct type. In the future,
    /// when extern refs get actually properly implemented at compile time we might want to get rid
    /// of this code.
    pub fn get_extern_vec_ref_index<T: UniqueTypeId<u64>>(&self, value: &[T]) -> u32 {
        let mut vec = Vec::with_capacity(value.len());
        for v in value {
            vec.push(self.get_extern_ref_index(v));
        }
        let p = self.get_extern_ref_from_ptr(value as *const [T] as *const u8, T::id().0, true);
        self.extern_vec_ref_lookup.write().insert(p, vec);
        p
    }

    /// Get an extern ref belonging to a vector we have passed to WASM.
    pub fn get_extern_vec_ref_extern_ref(&self, extern_vec_ref: u32, index: u32) -> u32 {
        let r = self.extern_vec_ref_lookup.read();
        let v = r.get(&extern_vec_ref).unwrap();
        v[index as usize]
    }

    /// Gets the extern ref index belonging to a specific pointer. If none exists, this will create
    /// a new one.
    fn get_extern_ref_from_ptr(&self, ptr: *const u8, type_id: u64, is_vec: bool) -> u32 {
        if let Some(v) = self.extern_ref_pointers_lookup.read().get(&ExternRefLookupKey {
            ptr,
            is_vec,
            t: type_id,
        }) {
            return *v as u32;
        }
        let index = {
            let mut extern_ref_guard = self.extern_ref_pointers.write();
            extern_ref_guard.push(ptr);
            extern_ref_guard.len() as u32
        };
        self.extern_ref_pointers_lookup.write().insert(
            ExternRefLookupKey {
                ptr,
                is_vec,
                t: type_id,
            },
            index,
        );
        self.extern_ref_type_lookup.write().insert(ExternRefLookupKey {
            ptr,
            is_vec,
            t: type_id,
        });
        index
    }

    /// Gets a value from the extern ref lookup. This turns an earlier registered index back into
    /// its proper value, validates its type, and returns the value.
    pub fn get_extern_ref_value<T: UniqueTypeId<u64>>(&self, index: u32) -> &T {
        let read_guard = self.extern_ref_pointers.read();
        let ptr = read_guard.get((index - 1) as usize).unwrap();
        if self
            .extern_ref_type_lookup
            .read()
            .get(&ExternRefLookupKey {
                ptr: *ptr,
                is_vec: false,
                t: T::id().0,
            })
            .is_none()
        {
            panic!(
                "Extern ref was accessed with wrong type. Requested type {}, but this was not the type the extern ref was stored with.",
                std::any::type_name::<T>()
            );
        }

        unsafe { (*ptr as *const T).as_ref().unwrap() }
    }
}

/// The runtime environment for script execution. This is passed to most of the host functions being called.
#[derive(Clone)]
pub(crate) struct WebAssemblyEnv {
    /// A pointer to the WebAssemblyScriptResolver belonging to the current script environment.
    data: Weak<WebAssemblyEnvironmentData>,
}

impl WebAssemblyEnv {
    /// Instantiates a new Environment with the requested data.
    pub fn new(data: Weak<WebAssemblyEnvironmentData>) -> Self {
        Self { data }
    }

    /// Get the actual data belonging to the current context.
    pub fn data(&self) -> Arc<WebAssemblyEnvironmentData> {
        self.data.upgrade().unwrap()
    }
}

unsafe impl Sync for WebAssemblyEnv {}

unsafe impl Send for WebAssemblyEnv {}

unsafe impl Sync for WebAssemblyEnvironmentData {}

unsafe impl Send for WebAssemblyEnvironmentData {}

impl WasmerEnv for WebAssemblyEnv {}