//! shredder
//! ========
//! `shredder` is a library providing a garbage collected smart pointer: `Gc`.
//! This is useful for times when you want shared access to some data, but the structure
//! of the data has unpredictable cycles in it. (So Arc would not be appropriate.)
//!
//! `shredder` has the following features:
//! - safe: detects error conditions on the fly, and protects you from undefined behavior
//! - ergonomic: no need to manually manage roots, just a regular smart pointer
//! - deref support: `DerefGc` gives you a garbage collected and `Deref` smart pointer where possible
//! - ready for fearless concurrency: works in multi-threaded contexts, with `AtomicGc` for cases where you need atomic operations
//! - limited stop-the world: regular processing will rarely be interrupted
//! - seamless destruction: regular `drop` for `'static` data
//! - clean finalization: optional `finalize` for non-`'static` data
//! - concurrent collection: collection happens in the background, improving performance
//! - concurrent destruction: destructors are run in the background, improving performance
//!
//! `shredder` has the following limitations:
//! - guarded access: accessing `Gc` data requires acquiring a guard (although you can use `DerefGc` in many cases to avoid this)
//! - multiple collectors: only a single global collector is supported
//! - can't handle `Rc`/`Arc`: requires all `Gc` objects have straightforward ownership semantics
//! - collection optimized for speed, not memory use: `Gc` and internal metadata is small, but there is bloat during collection (will fix!)
//! - no no-std support: The collector requires threading and other `std` features (will fix!)

#![cfg_attr(feature = "nightly-features", feature(unsize, coerce_unsized))]
// We love docs here
#![deny(missing_docs)]
// Clippy configuration:
// I'd like the most pedantic warning level
#![warn(
    clippy::cargo,
    clippy::needless_borrow,
    clippy::pedantic,
    clippy::redundant_clone,
    clippy::use_self,
    rust_2018_idioms
)]
// But I don't care about these ones
#![allow(
    clippy::cast_precision_loss,     // There is no way to avoid this precision loss
    clippy::explicit_deref_methods,  // Sometimes calling `deref` directly is clearer
    clippy::module_name_repetitions, // Sometimes clear naming calls for repetition
    clippy::multiple_crate_versions  // There is no way to easily fix this without modifying our dependencies
)]

#[macro_use]
extern crate crossbeam;

#[macro_use]
extern crate log;

#[macro_use]
extern crate rental;

/// Atomic gc operations
pub mod atomic;
mod collector;
mod concurrency;
mod finalize;
/// Marker types
pub mod marker;
/// Various types used for plumbing, stuff you don't need to care about
pub mod plumbing;
mod r;
mod scan;
mod smart_ptr;
mod std_impls;
/// Helpful wrappers used for convenience methods
pub mod wrappers;

use std::cell::RefCell;
use std::sync::{Mutex, RwLock};

use crate::collector::COLLECTOR;

pub use crate::finalize::{Finalize, FinalizeFields};
pub use crate::r::{RMut, R};
pub use crate::scan::{Scan, Scanner, ToScan};
pub use crate::smart_ptr::{DerefGc, Gc, GcGuard};

/// A convenient alias for `Gc<RefCell<T>>`.
/// Note that `Gc<RefCell<T>>` has additional specialized methods for working with `RefCell`s inside
/// `Gc`s.
pub type GRefCell<T> = Gc<RefCell<T>>;

/// A convenient alias for `Gc<Mutex<T>>`.
/// Note that `Gc<Mutex<T>>` has additional specialized methods for working with `Mutex`s inside
/// `Gc`s.
pub type GMutex<T> = Gc<Mutex<T>>;

/// A convenient alias for `Gc<RwLock<T>>`.
/// Note that `Gc<Mutex<T>>` has additional specialized methods for working with `Mutex`s inside
/// `Gc`s.
pub type GRwLock<T> = Gc<RwLock<T>>;

/// Returns how many underlying allocations are currently allocated.
///
/// # Example
/// ```
/// use shredder::{number_of_tracked_allocations, Gc};
///
/// let data = Gc::new(128);
/// assert!(number_of_tracked_allocations() > 0);
/// ```
#[must_use]
pub fn number_of_tracked_allocations() -> usize {
    COLLECTOR.tracked_data_count()
}

/// Returns how many `Gc`s are currently in use.
///
/// # Example
/// ```
/// use shredder::{number_of_active_handles, Gc};
///
/// let data = Gc::new(128);
/// assert!(number_of_active_handles() > 0);
/// ```
#[must_use]
pub fn number_of_active_handles() -> usize {
    COLLECTOR.handle_count()
}

/// Sets the percent more data that'll trigger collection.
///
/// `shredder`'s collection automatically triggers when:
/// ```text
///     allocations > allocations_after_last_collection * (1 + gc_trigger_percent)
/// ```
/// The default value of `gc_trigger_percent` is 0.75, but `set_gc_trigger_percent` lets you
/// configure it yourself. Only values 0 or greater are allowed.
/// (NaNs and negative values will cause a panic.)
///
/// # Example
/// ```
/// use shredder::set_gc_trigger_percent;
/// set_gc_trigger_percent(0.75); // GC will trigger after data exceeds 1.75x previous heap size
/// ```
pub fn set_gc_trigger_percent(percent: f32) {
    if percent < -0.0 || percent.is_nan() {
        panic!(
            "The trigger percentage cannot be less than zero or NaN! (percent = {})",
            percent
        )
    }
    COLLECTOR.set_gc_trigger_percent(percent)
}

/// A function for manually running a collection, ignoring the heuristic that governs normal
/// garbage collector operations.
///
/// This can be an extremely slow operation, since the algorithm is
/// designed to be run in the background, while this method runs it on the thread that calls the
/// method. Additionally, you may end up blocking waiting to collect, since `shredder` doesn't allow
/// two collections at once (and if this happens, you'll effectively get two collections in a row).
///
/// # Example
/// ```
/// use shredder::collect;
/// collect(); // Manually run GC
/// ```
pub fn collect() {
    COLLECTOR.collect();
}

/// Block the current thread until the background thread has finished running the destructors for
/// all data that was marked as garbage at the point this method was called.
///
/// This method is most useful for testing, as well as cleaning up at the termination of your
/// program.
/// # Example
/// ```
/// use shredder::{collect, synchronize_destructors};
/// // Create some data
/// // <SNIP>
/// // Gc happens
/// collect();
/// // We cleanup
/// synchronize_destructors();
/// // At this point all destructors for garbage will have been run
/// ```
pub fn synchronize_destructors() {
    COLLECTOR.synchronize_destructors()
}

/// A convenience method for helping ensure your destructors are run.
///
/// In Rust you can never assume that destructors run, but using this method helps `shredder` not
/// contribute to that problem.
/// # Example
/// ```
/// use shredder::{run_with_gc_cleanup};
///
/// // Generally you'd put this in `main`
/// run_with_gc_cleanup(|| {
///     // Your code goes here!
/// })
/// ```
pub fn run_with_gc_cleanup<T, F: FnOnce() -> T>(f: F) -> T {
    let res = f();

    collect();
    synchronize_destructors();

    res
}

// Re-export the Scan derive, at the bottom cause the documentation is long
/// The `Scan` derive, powering `#[derive(Scan)]`. Important details here!
///
/// Doing `#[derive(Scan)]` will in fact derive `Scan`. However, it actually can auto-implement
/// four traits:
/// - `Scan`   (automatic) [requires all fields be `Scan`]
/// - `GcSafe` (automatic) [requires all fields be `GcSafe`]
/// - `GcDrop` (opt-out)   [requires all fields be `GcDrop`]
/// - `GcDeref`(opt-in)    [requires all fields be `GcDeref`]
///
/// `Scan` and `GcSafe` are the fundamental things that this derive implements. There is no way
/// to opt out.
///
/// To opt out of `GcDrop`, supply the `cant_drop` flag. Ex:
/// ```
/// use shredder::DerefGc;
/// use shredder::Scan;
///
/// #[derive(Scan)]
/// #[shredder(cant_drop)]
/// struct WontBeDrop {
///     v: DerefGc<u32>
/// }
/// ```
///
/// To opt into `GcDeref`, supply the `can_deref` flag:
/// ```
/// use shredder::DerefGc;
/// use shredder::Scan;
///
/// #[derive(Scan)]
/// #[shredder(can_deref, cant_drop)]
/// struct WillBeDeref {
///     v: DerefGc<u32>
/// }
/// ```
///
/// Now there are also field flags, which will remove the recursive checks for that trait:
/// - `skip_scan` (skips check for `Scan`)
/// - `unsafe_skip_gc_deref` (skips check for `GcDeref`. Unsafe!)
/// - `unsafe_skip_gc_drop` (skips check for `GcDrop`. Unsafe!)
/// - `unsafe_skip_gc_safe` (skips check for `GcSafe`. Unsafe!)
/// - `unsafe_skip_all` (skips check for all traits. Unsafe!)
///
/// For safety, if you use unsafe flags, you must ensure those fields satisfy the trait requirements
/// anyway.
///
/// Here is an example of skip usage
/// ```
/// use shredder::DerefGc;
/// use shredder::Scan;
///
/// #[derive(Scan)]
/// struct WillBeDeref {
///     #[shredder(skip_scan, unsafe_skip_gc_safe, unsafe_skip_gc_drop)]
///     v: *const u32 // Assume this is always a valid ptr
/// }
/// ```
pub use shredder_derive::Scan;

/// The `Finalize` derive, powering `#[derive(Finalize)]`
///
/// Much more straightforward than the `Scan` derive, it implements `Finalize` by delegating
/// to each field's `finalize` method.
///
/// A simple example would be:
/// ```
/// use shredder::DerefGc;
/// use shredder::Finalize;
///
/// #[derive(Finalize)]
/// struct WillBeFinalize {
///     v: DerefGc<u32>
/// }
/// ```
///
/// You can also skip fields with `skip_finalize`:
/// ```
/// use shredder::DerefGc;
/// use shredder::Finalize;
///
/// struct NotFinalize;
///
/// #[derive(Finalize)]
/// struct StillFinalize {
///     #[shredder(skip_finalize)]
///     v: NotFinalize
/// }
/// ```
/// `unsafe_skip_all` also includes `skip_finalize`
pub use shredder_derive::Finalize;

/// The `FinalizeFields` derive, powering `#[derive(FinalizeFields)]`
///
/// Works exactly the same as `#[derive(Finalize)]`, but derives `FinalizeFields` instead. This is
/// primarily useful if you want to implement `Finalize` yourself, but don't want to manually
/// finalize each field. This way that logic is autogenerated, and you just need to call
/// `finalize_fields` at the end of your `finalize` method.
pub use shredder_derive::FinalizeFields;