abstract_differential.jl

#####
##### `AbstractDifferential`
#####

"""
The subtypes of `AbstractDifferential` define a custom \"algebra\" for chain
rule evaluation that attempts to factor various features like complex derivative
support, broadcast fusion, zero-elision, etc. into nicely separated parts.

All subtypes of `AbstractDifferential` implement the following operations:

`+(a, b)`: linearly combine differential `a` and differential `b`

`*(a, b)`: multiply the differential `a` by the differential `b`

`Base.conj(x)`: complex conjugate of the differential `x`

`extern(x)`: convert `x` into an appropriate non-`AbstractDifferential` type for
use outside of `ChainContext`.

Valid arguments to these operations are `T` where `T<:AbstractDifferential`, or
where `T` has proper `+` and `*` implementations.

Additionally, all subtypes of `AbstractDifferential` support `Base.iterate` and
`Base.Broadcast.broadcastable(x)`.
"""
abstract type AbstractDifferential end

Base.:+(x::AbstractDifferential) = x

"""
    extern(x)

Return `x` converted to an appropriate non-`AbstractDifferential` type, for use
with external packages that might not handle `AbstractDifferential` types.

Note that this function may return an alias (not necessarily a copy) to data
wrapped by `x`, such that mutating `extern(x)` might mutate `x` itself.
"""
@inline extern(x) = x

@inline Base.conj(x::AbstractDifferential) = x