v0.4.1

Only bugfixes, improved tests and optimizations, and one added method

Fix: Correct types to ccall
Perf: @nospecialize for methods where specializations give nothing
Fix: Add missing GC.@preserves
Fix: Improve static VERSION checks
Improve test coverage
Add 5-arg copyto! method

v0.4.0

Breaking changes

• Removed the Unsafe trait type.
  Instead of MutableMemoryView(::Unsafe, ::MemoryView), use
  unsafe_from_parts(::MemoryRef, ::Int)
  Using the inner constructor MemoryView{T, M}(::Unsafe, ::MemoryRef{T}, ::Int)
  was never documented API and is now removed.

• Matrix and other Array types with a different dimensionality than 1 is now
  NotMemory, since it is not equal to its own memory view, due to shape mismatch.
  The fact that it was previously IsMemory was a bug.

• Out of bounds access now throws a LightBoundsError from the LightBoundsErrors
  package, instead of Base.BoundsError.
  This improves codegen slightly, as it enables escape analysis of the array,
  and outlines error paths slightly more.

• MemoryView(::SubArray) now accepts fewer subarray types. However, it is unlikely
  that any instance that is now no longer accepted worked previously, so it is
  unlikely to be breaking in practice.

Other

• parentindices now works correctly for zero-sized structs.
• Base.memoryref(::MemoryView) obtains the MemoryRef in a MemoryView.

v0.3.7

Various internal bugfixes

v0.3.6

• Allow indexing with UInt range (#29)
• Add optimised fill!(::MutableMemoryView{UInt8}, ::UInt8) (#30)

v0.3.5

0.3.5

• Add method MemoryKind{::Type{<:MemoryView}}
• Add package extension for LibDeflate.jl

v0.3.4

Add new function unsafe_from_parts to construct a memory view from a MemoryRef.

v0.3.3

New features

• Add a new split_each function, which iterates over memory views delimited
  by a single element delimiter.

Other changes

• Add some more fast methods for Memory/Vector/MemoryView interop
• Add a fast method for Base.readbytes! reading into a memory view

v0.3.2

New features

• FixedSizeArrays.jl is now supported through an extension

Other changes

• Various small bugfixes and optimisations

v0.3.1

Various bugfixes.

v0.3.0

Breaking changes

• Change the bounds checking behaviour of the find* functions to match those of
  Vector. In particular, previously, findnext(pred, mem, -5) would be
  equivalent to searching from index 1, and similarly, findprev(pred, mem, lastindex(mem) + 10) would be equialent to searching from lastindex(mem).
  Now, searching from an index before the first valid index throws a BoundsError.
  Findfirst searching from i > lastindex(mem), and findlast searching from
  i < 1 will still simply return nothing, just like searching vectors.

Other changes

• Add optimised versions of findprev and findlast, searching bytes
• Add optimised version of find*(iszero, bytes) methods
• Add optimised generic find* methods
• Add functions split_first, split_last, split_at and split_unaligned
• Add a more correct implementation of Base.mightalias for memory views and
  some types of arrays