//! Library for idiomatic URID support.
//!
//! In the world of [RDF](https://en.wikipedia.org/wiki/Resource_Description_Framework), resources are described using [URIs](https://en.wikipedia.org/wiki/Uniform_Resource_Identifier). In Rust, this concept can be adapted to describe types with URIs using the [`UriBound`](trait.UriBound.html) trait. Then, other crates might use these URIs to describe relationships between types or values using URIs.
//!
//! However, comparing URIs isn't necessarily fast. Therefore, another concept was introduced: The [URID](struct.URID.html). A URID is basically a `u32` which represents a URI. These URIDs are assigned by a [`Map`](trait.Map.html) and can be "dereferenced" by an [`Unmap`](trait.Unmap.html).
//!
//! This library also supports connecting URIDs to their `UriBound` via a generics argument. This can be used, for example, to request the URID of a certain bound as a parameter of a function. If someone would try to call this function with the wrong URID, the compiler will raise an error before the code is even compiled.
//!
//! This may seem a bit minor to you now, but the audio plugin framework [rust-lv2](https://github.com/RustAudio/rust-lv2) heavily relies on this crate for fast, portable and dynamic data identification and exchange.
//!
//! # Example
//!
//! ```
//! use urid::*;
//!
//! // Some types with URIs. The attribute implements `UriBound` with the given URI.
//! #[uri("urn:urid-example:my-struct-a")]
//! struct MyStructA;
//!
//! #[uri("urn:urid-example:my-struct-b")]
//! struct MyStructB;
//!
//! // A collection of URIDs that can be created by a mapper with one method call.
//! #[derive(URIDCollection)]
//! struct MyURIDCollection {
//!     my_struct_a: URID<MyStructA>,
//!     my_struct_b: URID<MyStructB>,
//! }
//!
//! // A function that checks whether the unmapper behaves correctly.
//! // Due to the type argument, it can not be misused.
//! fn test_unmapper<M: Unmap, T: UriBound>(unmap: &M, urid: URID<T>) {
//!     assert_eq!(T::uri(), unmap.unmap(urid).unwrap());
//! }
//!
//! // Create a simple mapper. The `HashURIDMapper` is thread-safe and can map and unmap all URIs.
//! let map = HashURIDMapper::new();
//!
//! // Get the URIDs of the structs. You can use the collection or retrieve individual URIDs.
//! let urids: MyURIDCollection = map.populate_collection().unwrap();
//! let urid_a: URID<MyStructA> = map.map_type().unwrap();
//!
//! // You can also retrieve the URID of a single URI without a binding to a type.
//! let urid_b: URID = map.map_str("https://rustup.rs").unwrap();
//!
//! test_unmapper(&map, urids.my_struct_a);
//! test_unmapper(&map, urids.my_struct_b);
//! test_unmapper(&map, urid_a);
//! ```
use std::cmp::{Ordering, PartialEq, PartialOrd};
use std::collections::HashMap;
use std::convert::TryInto;
use std::fmt;
use std::hash::{Hash, Hasher};
use std::marker::PhantomData;
use std::num::NonZeroU32;
use std::sync::Mutex;

pub use urid_derive::*;

/// Representation of a borrowed Uri.
pub type Uri = ::std::ffi::CStr;
/// Representation of an owned Uri.
pub type UriBuf = ::std::ffi::CString;

/// A trait for types that can be identified by a URI.
///
/// Every type that can be identified by a URI implements this trait. In most cases, you can use the `uri` attribute to implement `UriBound` safely and quickly:
///
/// ```
/// use urid::*;
///
/// // Defining the struct
/// #[uri("urn:urid-example:my-struct")]
/// pub struct MyStruct {
///     a: f32,
/// }
///
/// // Retrieving the URI
/// assert_eq!("urn:urid-example:my-struct", MyStruct::uri().to_str().unwrap());
/// ```
///
/// However, in some cases, you need to implement `UriBound` manually, for example if the URI comes from a generated `sys` crate:
///
/// ```
/// use urid::*;
///
/// // This URI is part of a generated `UriBound` crate:
/// const A_FANCY_URI: &'static [u8] = b"urn:urid-example:fancy-uri\0";
///
/// // This struct is part of a safe wrapper crate:
/// struct FancyStruct {
///     _fancy_content: f32,
/// }
///
/// unsafe impl UriBound for FancyStruct {
///     const URI: &'static [u8] = A_FANCY_URI;
/// }
///
/// assert_eq!("urn:urid-example:fancy-uri", FancyStruct::uri().to_str().unwrap())
/// ```
///
/// # Unsafety
///
/// The [`URI`](#associatedconstant.URI) constant has to contain a null terminator (The `\0` character at the end), which is used by C programs to determine the end of the string. If you omit it, other parts of your program may violate memory access rules, which is considered undefined behaviour. Since this can not be statically checked by the compiler, this trait is unsafe to implement manually.
pub unsafe trait UriBound {
    /// The URI of the type, safed as a byte slice
    ///
    /// Currently, there is no way to express a `CStr` in a constant way. Therefore, the URI has to be stored as a null-terminated byte slice.
    ///
    /// The slice must be a valid URI and must have the null character, expressed as `\0`, at the end. Otherwise, other code might produce a segmentation fault or read a faulty URI while looking for the null character.
    const URI: &'static [u8];

    /// Construct a `CStr` reference to the URI.
    ///
    /// Assuming that [`URI`](#associatedconstant.URI) is correct, this method constructs a `CStr` reference from the byte slice referenced by `URI`.
    fn uri() -> &'static Uri {
        unsafe { Uri::from_bytes_with_nul_unchecked(Self::URI) }
    }
}

/// Representation of a URI for fast comparisons.
///
/// A URID is basically a number which represents a URI, which makes the identification of other features faster and easier. The mapping of URIs to URIDs is handled by a something that implements the [`Map`](trait.Map.html) trait. A given URID can also be converted back to a URI with an implementation of the [`Unmap`](trait.Unmap.html) trait. However, these implementations should obviously be linked.
///
/// This struct has an optional type parameter `T` which defaults to `()`. In this case, the type can represent any URID at all, but if `T` is a `UriBound`, the instance of `URID<T>` can only describe the URID of the given bound. This makes creation easier and also turns it into an atomic [`URIDCollection`](trait.URIDCollection.html), which can be used to build bigger collections.
#[repr(transparent)]
pub struct URID<T = ()>(NonZeroU32, PhantomData<T>)
where
    T: ?Sized;

/// A store of pre-mapped URIDs
///
/// This trait can be used to easily cache URIDs. The usual way of creating such a collection is to define a struct of `URID<T>`s, where `T` implements `UriBound`, and then using the derive macro to implement `URIDCollection` for it. Then, you can populate it with a map and access it any time, even in a real-time-sensitive context.
///
/// # Usage example:
///
///     # use urid::*;
///     // Defining all URI bounds.
///     #[uri("urn:my-type-a")]
///     struct MyTypeA;
///     
///     #[uri("urn:my-type-b")]
///     struct MyTypeB;
///
///     // Defining the collection.
///     #[derive(URIDCollection)]
///     struct MyCollection {
///         my_type_a: URID<MyTypeA>,
///         my_type_b: URID<MyTypeB>,
///     }
///
///     // Creating a mapper and collecting URIDs.
///     let map = HashURIDMapper::new();
///     let collection = MyCollection::from_map(&map).unwrap();
///
///     // Asserting.
///     assert_eq!(1, collection.my_type_a);
///     assert_eq!(2, collection.my_type_b);
pub trait URIDCollection: Sized {
    /// Construct the collection from the mapper.
    fn from_map<M: Map + ?Sized>(map: &M) -> Option<Self>;
}

impl URID<()> {
    /// Creates a new URID from a raw number.
    ///
    /// URIDs may never be zero. If the given number is zero, `None` is returned.
    #[inline]
    pub fn new(raw_urid: u32) -> Option<Self> {
        NonZeroU32::new(raw_urid).map(|inner| Self(inner, PhantomData))
    }
}

impl<T: ?Sized> URID<T> {
    /// Create a URID without checking for type or value validity.
    ///
    /// This value may only be a URID the mapper actually produced and that is recognised by a compatible unmapper. Therefore, it should only be used by [`Map::map_uri`](trait.Map.html#tymethod.map_uri) or [`Map::map_type`](trait.Map.html#method.map_type).
    ///
    /// # Safety
    ///
    /// A URID may not be 0 since this value is reserved for the `None` value of `Option<URID<T>>`, which therefore has the same size as a `URID<T>`. If `T` is also a URI bound, the URID may only be the one that is mapped to the bounded URI.
    ///
    /// Since these constraints aren't checked by this method, it is unsafe. Using this method is technically sound as long as `raw_urid` is not zero, but might still result in bad behaviour if its the wrong URID for the bound `T`.
    pub unsafe fn new_unchecked(raw_urid: u32) -> Self {
        Self(NonZeroU32::new_unchecked(raw_urid), PhantomData)
    }

    /// Return the raw URID number.
    pub fn get(self) -> u32 {
        self.0.get()
    }

    /// Transform the type-specific URID into a generalized one.
    pub fn into_general(self) -> URID<()> {
        unsafe { URID::new_unchecked(self.get()) }
    }
}

impl<T: UriBound + ?Sized> URIDCollection for URID<T> {
    fn from_map<M: Map + ?Sized>(map: &M) -> Option<Self> {
        map.map_type()
    }
}

impl<T: ?Sized> fmt::Debug for URID<T> {
    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
        self.0.fmt(f)
    }
}

impl<T: ?Sized> Clone for URID<T> {
    fn clone(&self) -> Self {
        Self(self.0, PhantomData)
    }
}

impl<T: ?Sized> Copy for URID<T> {}

impl<T1: ?Sized, T2: ?Sized> PartialEq<URID<T1>> for URID<T2> {
    fn eq(&self, other: &URID<T1>) -> bool {
        self.0 == other.0
    }
}

impl<T: ?Sized> PartialEq<u32> for URID<T> {
    fn eq(&self, other: &u32) -> bool {
        self.get() == *other
    }
}

impl<T: ?Sized> PartialEq<URID<T>> for u32 {
    fn eq(&self, other: &URID<T>) -> bool {
        *self == other.get()
    }
}

impl<T: ?Sized> Eq for URID<T> {}

impl<T1: ?Sized, T2: ?Sized> PartialOrd<URID<T1>> for URID<T2> {
    fn partial_cmp(&self, other: &URID<T1>) -> Option<Ordering> {
        self.0.partial_cmp(&other.0)
    }
}

impl<T: ?Sized> PartialOrd<u32> for URID<T> {
    fn partial_cmp(&self, other: &u32) -> Option<Ordering> {
        self.get().partial_cmp(other)
    }
}

impl<T: ?Sized> PartialOrd<URID<T>> for u32 {
    fn partial_cmp(&self, other: &URID<T>) -> Option<Ordering> {
        self.partial_cmp(&other.get())
    }
}

impl<T: ?Sized> Ord for URID<T> {
    fn cmp(&self, other: &Self) -> Ordering {
        self.0.cmp(&other.0)
    }
}

impl<T: ?Sized> Hash for URID<T> {
    fn hash<H: Hasher>(&self, state: &mut H) {
        self.0.hash(state)
    }
}

impl std::convert::TryFrom<u32> for URID {
    type Error = ();

    fn try_from(value: u32) -> Result<URID, ()> {
        if value == 0 {
            Err(())
        } else {
            Ok(unsafe { URID::new_unchecked(value) })
        }
    }
}

#[cfg(test)]
mod tests {
    use crate::URID;

    #[test]
    fn test_urid_size() {
        use std::mem::size_of;

        let size = size_of::<u32>();

        assert_eq!(size, size_of::<URID>());
        assert_eq!(size, size_of::<Option<URID>>());
    }
}

/// A handle to map URIs to URIDs.
pub trait Map {
    /// Maps an URI to a `URID` that corresponds to it.
    ///
    /// If the URI has not been mapped before, a new URID will be assigned.
    ///
    /// # Errors
    /// This method may return `None` in the exceptional case that an ID for that URI could not be
    /// created for whatever reason.
    /// However, implementations SHOULD NOT return `None` from this function in non-exceptional
    /// circumstances (i.e. the URI map SHOULD be dynamic).
    ///
    /// # Realtime usage
    /// This action may not be realtime-safe since it may involve locking mutexes or allocating dynamic memory. If you are working in a realtime environment, you should cache mapped URIDs in a [`URIDCollection`](trait.URIDCollection.html) and use it instead.
    fn map_uri(&self, uri: &Uri) -> Option<URID>;

    /// Map an URI, encoded as a `str` to a `URID` that corresponds to it.
    ///
    /// This function copies the string into a vector, adds a null terminator and calls [`map_uri`](#tymethod.map_uri) with it. Therefore, the rules of `map_uri` apply here too.
    ///
    /// # Additional Errors
    /// This method has the same error cases as `map_uri`, but also returns `None` if the string isn't an ASCII string or if the string can not be converted to a `Uri`.
    fn map_str(&self, uri: &str) -> Option<URID> {
        if !uri.is_ascii() {
            return None;
        }
        let mut bytes: Vec<u8> = uri.as_bytes().to_owned();
        bytes.push(0);
        self.map_uri(Uri::from_bytes_with_nul(bytes.as_ref()).ok()?)
    }

    /// Retrieve the URI of the bound and map it to a URID.
    ///
    /// The rules of [`map_uri`](#tymethod.map_uri) apply here too.
    fn map_type<T: UriBound + ?Sized>(&self) -> Option<URID<T>> {
        self.map_uri(T::uri())
            .map(|urid| unsafe { URID::new_unchecked(urid.get()) })
    }

    /// Populate a URID collection.
    ///
    /// This is basically an alias for [`T::from_map(self)`](trait.URIDCollection.html#tymethod.from_map) that simplifies the derive macro for `URIDCollection`.
    fn populate_collection<T: URIDCollection>(&self) -> Option<T> {
        T::from_map(self)
    }
}

/// A handle to map URIDs to URIs.
pub trait Unmap {
    /// Get the URI of a previously mapped URID.
    ///
    /// This method may return `None` if the given `urid` is not mapped to URI yet.
    ///
    /// # Realtime usage
    /// This action may not be realtime-safe since it may involve locking mutexes or allocating dynamic memory. If you are working in a realtime environment, you should cache mapped URIDs in a [`URIDCollection`](trait.URIDCollection.html) and use it instead.
    fn unmap<T: ?Sized>(&self, urid: URID<T>) -> Option<&Uri>;
}

/// A simple URI → URID mapper, backed by a standard `HashMap` and a `Mutex` for multi-thread
/// access.
#[derive(Default)]
pub struct HashURIDMapper(Mutex<HashMap<UriBuf, URID>>);

impl Map for HashURIDMapper {
    fn map_uri(&self, uri: &Uri) -> Option<URID<()>> {
        let mut map = self.0.lock().ok()?; // Fail if the Mutex got poisoned
        match map.get(uri) {
            Some(urid) => Some(*urid),
            None => {
                let map_length: u32 = map.len().try_into().ok()?; // Fail if there are more items into the HashMap than an u32 can hold
                let next_urid = map_length.checked_add(1)?; // Fail on overflow when adding 1 for the next URID

                // This is safe, because we just added 1 to the length and checked for overflow, therefore the number can never be 0.
                let next_urid = unsafe { URID::new_unchecked(next_urid) };
                map.insert(uri.into(), next_urid);
                Some(next_urid)
            }
        }
    }
}

impl Unmap for HashURIDMapper {
    fn unmap<T: ?Sized>(&self, urid: URID<T>) -> Option<&Uri> {
        let map = self.0.lock().ok()?;
        for (uri, contained_urid) in map.iter() {
            if *contained_urid == urid {
                // Here we jump through some hoops to return a reference that bypasses the mutex.
                // This is safe because the only way this reference might become invalid is if an
                // entry gets overwritten, which is not something that we allow through this
                // interface.
                return Some(unsafe {
                    let bytes = uri.as_bytes_with_nul();
                    Uri::from_bytes_with_nul_unchecked(std::slice::from_raw_parts(
                        bytes.as_ptr(),
                        bytes.len(),
                    ))
                });
            }
        }

        None
    }
}

impl HashURIDMapper {
    /// Create a new URID map store.
    pub fn new() -> Self {
        Default::default()
    }
}