# Dataset defined in types.jl

# Get the dataspace of a dataset
dataspace(dset::Dataset) = Dataspace(API.h5d_get_space(checkvalid(dset)))

"""
    open_dataset(parent::Union{File, Group}, path::AbstractString; properties...)

Open an existing [`HDF5.Dataset`](@ref) at `path` under `parent`

Optional keyword arguments include any keywords that that belong to
[`DatasetAccessProperties`](@ref) or [`DatasetTransferProperties`](@ref).
"""
open_dataset(
    parent::Union{File,Group},
    name::AbstractString,
    dapl::DatasetAccessProperties,
    dxpl::DatasetTransferProperties
) = Dataset(API.h5d_open(checkvalid(parent), name, dapl), file(parent), dxpl)

function open_dataset(parent::Union{File,Group}, name::AbstractString; pv...)
    dapl = DatasetAccessProperties()
    dxpl = DatasetTransferProperties()
    pv = setproperties!(dapl, dxpl; pv...)
    isempty(pv) || error("invalid keyword options $(keys(pv))")
    open_dataset(parent, name, dapl, dxpl)
end

# Setting dset creation properties with name/value pairs
"""
    create_dataset(parent, path, datatype, dataspace; properties...)

# Arguments
* `parent` - `File` or `Group`
* `path` - `String` describing the path of the dataset within the HDF5 file or
           `nothing` to create an anonymous dataset
* `datatype` - `Datatype` or `Type` or the dataset
* `dataspace` - `Dataspace` or `Dims` of the dataset
* `properties` - keyword name-value pairs set properties of the dataset

# Keywords

There are many keyword properties that can be set. Below are a few select keywords.
* `chunk` - `Dims` describing the size of a chunk. Needed to apply filters.
* `filters` - `AbstractVector{<: Filters.Filter}` describing the order of the filters to apply to the data. See [`Filters`](@ref)
* `external` - `Tuple{AbstractString, Intger, Integer}` `(filepath, offset, filesize)` External dataset file location, data offset, and file size. See [`API.h5p_set_external`](@ref).

Additionally, the initial create, transfer, and access properties can be provided as a keyword:
* `dcpl` - [`DatasetCreateProperties`](@ref)
* `dxpl` - [`DatasetTransferProperties`](@ref)
* `dapl` - [`DatasetAccessProperties`](@ref)

"""
function create_dataset(
    parent::Union{File,Group},
    path::Union{AbstractString,Nothing},
    dtype::Datatype,
    dspace::Dataspace;
    dcpl::DatasetCreateProperties=DatasetCreateProperties(),
    dxpl::DatasetTransferProperties=DatasetTransferProperties(),
    dapl::DatasetAccessProperties=DatasetAccessProperties(),
    pv...
)
    !isnothing(path) &&
        haskey(parent, path) &&
        error(
            "cannot create dataset: object \"", path, "\" already exists at ", name(parent)
        )
    pv = setproperties!(dcpl, dxpl, dapl; pv...)
    isempty(pv) || error("invalid keyword options")
    if isnothing(path)
        ds = API.h5d_create_anon(parent, dtype, dspace, dcpl, dapl)
    else
        ds = API.h5d_create(parent, path, dtype, dspace, _link_properties(path), dcpl, dapl)
    end
    Dataset(ds, file(parent), dxpl)
end
create_dataset(
    parent::Union{File,Group},
    path::Union{AbstractString,Nothing},
    dtype::Datatype,
    dspace_dims::Dims;
    pv...
) = create_dataset(checkvalid(parent), path, dtype, dataspace(dspace_dims); pv...)
create_dataset(
    parent::Union{File,Group},
    path::Union{AbstractString,Nothing},
    dtype::Datatype,
    dspace_dims::Tuple{Dims,Dims};
    pv...
) = create_dataset(
    checkvalid(parent),
    path,
    dtype,
    dataspace(dspace_dims[1]; max_dims=dspace_dims[2]);
    pv...
)
create_dataset(
    parent::Union{File,Group},
    path::Union{AbstractString,Nothing},
    dtype::Type,
    dspace_dims::Tuple{Dims,Dims};
    pv...
) = create_dataset(
    checkvalid(parent),
    path,
    datatype(dtype),
    dataspace(dspace_dims[1]; max_dims=dspace_dims[2]);
    pv...
)
create_dataset(
    parent::Union{File,Group},
    path::Union{AbstractString,Nothing},
    dtype::Type,
    dspace_dims::Dims;
    pv...
) = create_dataset(checkvalid(parent), path, datatype(dtype), dataspace(dspace_dims); pv...)
create_dataset(
    parent::Union{File,Group},
    path::Union{AbstractString,Nothing},
    dtype::Type,
    dspace_dims::Int...;
    pv...
) = create_dataset(checkvalid(parent), path, datatype(dtype), dataspace(dspace_dims); pv...)
create_dataset(
    parent::Union{File,Group},
    path::Union{AbstractString,Nothing},
    dtype::Type,
    dspace::Dataspace;
    pv...
) = create_dataset(checkvalid(parent), path, datatype(dtype), dspace; pv...)

# Get the datatype of a dataset
datatype(dset::Dataset) = Datatype(API.h5d_get_type(checkvalid(dset)), file(dset))

"""
    read_dataset(parent::Union{File,Group}, name::AbstractString)

Read a dataset with named `name` from `parent`. This will typically return an array.
The dataset will be opened, read, and closed.

See also [`HDF5.open_dataset`](@ref), [`Base.read`](@ref)
"""
function read_dataset(parent::Union{File,Group}, name::AbstractString)
    local ret
    obj = open_dataset(parent, name)
    try
        ret = read(obj)
    finally
        close(obj)
    end
    ret
end

refresh(ds::Dataset) = API.h5d_refresh(checkvalid(ds))
Base.flush(ds::Dataset) = API.h5d_flush(checkvalid(ds))

# Array constructor for datasets
Base.Array(x::Dataset) = read(x)

# The next two lines are kept for v"1.4" <= VERSION <= v"1.5"
Base.lastindex(dset::Dataset) = length(dset)
Base.lastindex(dset::Dataset, d::Int) = size(dset, d)

function iscompact(obj::Dataset)
    prop = API.h5d_get_create_plist(checkvalid(obj))
    try
        API.h5p_get_layout(prop) == API.H5D_COMPACT
    finally
        API.h5p_close(prop)
    end
end

function ischunked(obj::Dataset)
    prop = API.h5d_get_create_plist(checkvalid(obj))
    try
        API.h5p_get_layout(prop) == API.H5D_CHUNKED
    finally
        API.h5p_close(prop)
    end
end

function iscontiguous(obj::Dataset)
    prop = API.h5d_get_create_plist(checkvalid(obj))
    try
        API.h5p_get_layout(prop) == API.H5D_CONTIGUOUS
    finally
        API.h5p_close(prop)
    end
end

# Reading with mmap
ismmappable(::Type{<:ScalarType}) = true
ismmappable(::Type{Complex{T}}) where {T<:BitsType} = true
ismmappable(::Type) = false
ismmappable(obj::Dataset, ::Type{T}) where {T} = ismmappable(T) && iscontiguous(obj)
ismmappable(obj::Dataset) = ismmappable(obj, get_jl_type(obj))

function readmmap(obj::Dataset, ::Type{T}) where {T}
    dspace = dataspace(obj)
    stype = API.h5s_get_simple_extent_type(dspace)
    (stype != API.H5S_SIMPLE) && error("can only mmap simple dataspaces")
    dims = size(dspace)

    if isempty(dims)
        return T[]
    end
    if !Sys.iswindows()
        local fdint
        prop = API.h5d_get_access_plist(obj)
        try
            # TODO: Should check return value of API.h5f_get_driver()
            fdptr = API.h5f_get_vfd_handle(obj.file, prop)
            fdint = unsafe_load(convert(Ptr{Cint}, fdptr))
        finally
            API.h5p_close(prop)
        end
        fd = fdio(fdint)
    else
        # This is a workaround since the regular code path does not work on windows
        # (see #89 for background). The error is that "Mmap.mmap(fd, ...)" cannot
        # create create a valid file mapping. The question is if the handler
        # returned by "API.h5f_get_vfd_handle" has
        # the correct format as required by the "fdio" function. The former
        # calls
        # https://gitlabext.iag.uni-stuttgart.de/libs/hdf5/blob/develop/src/H5FDcore.c#L1209
        #
        # The workaround is to create a new file handle, which should actually
        # not make any problems. Since we need to know the permissions of the
        # original file handle, we first retrieve them using the "API.h5f_get_intent"
        # function

        # Check permissions
        intent = API.h5f_get_intent(obj.file)
        flag = intent == API.H5F_ACC_RDONLY ? "r" : "r+"
        fd = open(obj.file.filename, flag)
    end

    offset = API.h5d_get_offset(obj)
    if offset == -1 % API.haddr_t
        # note that API.h5d_get_offset may not actually raise an error, so we need to check it here
        error("Error getting offset")
    elseif offset % Base.datatype_alignment(T) == 0
        A = Mmap.mmap(fd, Array{T,length(dims)}, dims, offset)
    else
        Aflat = Mmap.mmap(fd, Vector{UInt8}, prod(dims) * sizeof(T), offset)
        A = reshape(reinterpret(T, Aflat), dims)
    end

    if Sys.iswindows()
        close(fd)
    end

    return A
end

function readmmap(obj::Dataset)
    T = get_jl_type(obj)
    ismmappable(T) || error("Cannot mmap datasets of type $T")
    iscontiguous(obj) || error("Cannot mmap discontiguous dataset")
    readmmap(obj, T)
end

# Generic write
function Base.write(
    parent::Union{File,Group},
    name1::Union{AbstractString,Nothing},
    val1,
    name2::Union{AbstractString,Nothing},
    val2,
    nameval...
) # FIXME: remove?
    if !iseven(length(nameval))
        error("name, value arguments must come in pairs")
    end
    write(parent, name1, val1)
    write(parent, name2, val2)
    for i in 1:2:length(nameval)
        thisname = nameval[i]
        if !isa(thisname, AbstractString)
            error("Argument ", i + 5, " should be a string, but it's a ", typeof(thisname))
        end
        write(parent, thisname, nameval[i + 1])
    end
end

# Plain dataset & attribute writes
# Due to method ambiguities we generate these explicitly

# Create datasets and attributes with "native" types, but don't write the data.
# The return syntax is: dset, dtype = create_dataset(parent, name, data; properties...)

function create_dataset(
    parent::Union{File,Group}, name::Union{AbstractString,Nothing}, data; pv...
)
    dtype = datatype(data)
    dspace = dataspace(data)
    obj = try
        create_dataset(parent, name, dtype, dspace; pv...)
    finally
        close(dspace)
    end
    return obj, dtype
end

# Create and write, closing the objects upon exit
"""
    write_dataset(parent::Union{File,Group}, name::Union{AbstractString,Nothing}, data; pv...)

Create and write a dataset with `data`. Keywords are forwarded to [`create_dataset`](@ref).
Providing `nothing` as the name will create an anonymous dataset.

See also [`create_dataset`](@ref)
"""
function write_dataset(
    parent::Union{File,Group}, name::Union{AbstractString,Nothing}, data; pv...
)
    obj, dtype = create_dataset(parent, name, data; pv...)
    try
        write_dataset(obj, dtype, data)
    catch exc
        delete_object(obj)
        rethrow(exc)
    finally
        close(obj)
        close(dtype)
    end
    nothing
end

# Write to already-created objects
function Base.write(obj::Dataset, x)
    dtype = datatype(x)
    try
        write_dataset(obj, dtype, x)
    finally
        close(dtype)
    end
end

# For plain files and groups, let "write(obj, name, val; properties...)" mean "write_dataset"
Base.write(parent::Union{File,Group}, name::Union{AbstractString,Nothing}, data; pv...) =
    write_dataset(parent, name, data; pv...)

# Indexing

Base.eachindex(::IndexLinear, A::Dataset) = Base.OneTo(length(A))
Base.axes(dset::Dataset) = map(Base.OneTo, size(dset))
Base.axes(dset::Dataset, d::Integer) = Base.OneTo(size(dset, d))

# Write to a subset of a dataset using array slices: dataset[:,:,10] = array

const IndexType = Union{AbstractRange{Int},Int,Colon}
function Base.setindex!(dset::Dataset, X::Array{T}, I::IndexType...) where {T}
    !isconcretetype(T) && error("type $T is not concrete")
    U = get_jl_type(dset)

    # perform conversions for numeric types
    if (U <: Number) && (T <: Number) && U !== T
        X = convert(Array{U}, X)
    end

    filetype = datatype(dset)
    memtype = _memtype(filetype, eltype(X))
    close(filetype)

    dspace = dataspace(dset)
    stype = API.h5s_get_simple_extent_type(dspace)
    stype == API.H5S_NULL && error("attempting to write to null dataspace")

    indices = Base.to_indices(dset, I)
    dspace = hyperslab(dspace, indices...)

    memspace = dataspace(X)

    if API.h5s_get_select_npoints(dspace) != API.h5s_get_select_npoints(memspace)
        error("number of elements in src and dest arrays must be equal")
    end

    try
        API.h5d_write(dset, memtype, memspace, dspace, dset.xfer, X)
    finally
        close(memtype)
        close(memspace)
        close(dspace)
    end

    return X
end

function Base.setindex!(
    dset::Dataset, X::Array{S}, I::IndexType...
) where {S<:AbstractString}
    !isconcretetype(S) && error("type $S is not concrete")
    U = get_jl_type(dset)

    filetype = datatype(dset)
    memtype = _memtype(filetype, eltype(X))
    close(filetype)

    dspace = dataspace(dset)
    stype = API.h5s_get_simple_extent_type(dspace)
    stype == API.H5S_NULL && error("attempting to write to null dataspace")

    indices = Base.to_indices(dset, I)
    dspace = hyperslab(dspace, indices...)

    memspace = dataspace(X)

    if API.h5s_get_select_npoints(dspace) != API.h5s_get_select_npoints(memspace)
        error("number of elements in src and dest arrays must be equal")
    end

    p = Ref{Cstring}(X)
    try
        API.h5d_write(dset, memtype, memspace, dspace, dset.xfer, p)
    finally
        close(memtype)
        close(memspace)
        close(dspace)
    end

    return X
end

function Base.setindex!(dset::Dataset, x, I::IndexType...)
    indices = Base.to_indices(dset, I)
    X = fill(x, map(length, indices))
    Base.setindex!(dset, X, indices...)
end

function Base.setindex!(dset::Dataset, x::T, I::IndexType...) where {T<:AbstractString}
    indices = Base.to_indices(dset, I)
    X = fill(x, map(length, indices))
    Base.setindex!(dset, X, indices...)
end

function Base.setindex!(dset::Dataset, X::AbstractArray, I::IndexType...)
    Base.setindex!(dset, Array(X), I...)
end

"""
    create_external_dataset(parent, name, filepath, dtype, dspace, offset = 0)

Create an external dataset with data in an external file.
* `parent` - File or Group
* `name` - Name of the Dataset
* `filepath` - File path to where the data is tored
* `dtype` - Datatype, Type, or value where `datatype` is applicable
* `offset` - Offset, in bytes, from the beginning of the file to the location in the file where the data starts.

See also [`API.h5p_set_external`](@ref) to link to multiple segments.
"""
function create_external_dataset(
    parent::Union{File,Group},
    name::AbstractString,
    filepath::AbstractString,
    t,
    sz::Dims,
    offset::Integer=0
)
    create_external_dataset(parent, name, filepath, datatype(t), dataspace(sz), offset)
end
function create_external_dataset(
    parent::Union{File,Group},
    name::AbstractString,
    filepath::AbstractString,
    dtype::Datatype,
    dspace::Dataspace,
    offset::Integer=0
)
    checkvalid(parent)
    create_dataset(
        parent,
        name,
        dtype,
        dspace;
        external=(filepath, offset, length(dspace) * sizeof(dtype))
    )
end

### HDF5 utilities ###

# default behavior
read_dataset(
    dset::Dataset, memtype::Datatype, buf, xfer::DatasetTransferProperties=dset.xfer
) = API.h5d_read(dset, memtype, API.H5S_ALL, API.H5S_ALL, xfer, buf)
write_dataset(
    dset::Dataset, memtype::Datatype, x, xfer::DatasetTransferProperties=dset.xfer
) = API.h5d_write(dset, memtype, API.H5S_ALL, API.H5S_ALL, xfer, x)

# type-specific behaviors
function _check_invalid(dataset::Dataset, buf::AbstractArray)
    num_bytes_dset = Base.checked_mul(sizeof(datatype(dataset)), length(dataset))
    num_bytes_buf = Base.checked_mul(sizeof(eltype(buf)), length(buf))
    num_bytes_buf == num_bytes_dset || throw(
        ArgumentError(
            "Invalid number of bytes: $num_bytes_buf != $num_bytes_dset, for dataset \"$(name(dataset))\""
        )
    )
    stride(buf, 1) == 1 || throw(
        ArgumentError("Cannot read/write arrays with a different stride than `Array`")
    )
end
function read_dataset(
    dataset::Dataset,
    memtype::Datatype,
    buf::AbstractArray,
    xfer::DatasetTransferProperties=dataset.xfer
)
    _check_invalid(dataset, buf)
    API.h5d_read(dataset, memtype, API.H5S_ALL, API.H5S_ALL, xfer, buf)
end

function write_dataset(
    dataset::Dataset,
    memtype::Datatype,
    buf::AbstractArray{T},
    xfer::DatasetTransferProperties=dataset.xfer
) where {T}
    _check_invalid(dataset, buf)
    if isbitstype(T)
        API.h5d_write(dataset, memtype, API.H5S_ALL, API.H5S_ALL, xfer, buf)
    else
        # For non-bitstypes, we need to convert the buffer to a bitstype
        # For mutable structs, this will usually be a NamedTuple.
        jl_type = get_mem_compatible_jl_type(memtype)
        try
            memtype_buf = convert(Array{jl_type}, buf)
            API.h5d_write(dataset, memtype, API.H5S_ALL, API.H5S_ALL, xfer, memtype_buf)
        catch err
            if err isa MethodError
                throw(
                    ArgumentError(
                        "Could not convert non-bitstype $T to $jl_type for writing to HDF5. Consider implementing `convert(::Type{$jl_type}, ::$T)`"
                    )
                )
            else
                rethrow()
            end
        end
    end
end

function write_dataset(
    dataset::Dataset,
    memtype::Datatype,
    buf::Base.ReinterpretArray,
    xfer::DatasetTransferProperties=dataset.xfer
)
    # We cannot obtain a pointer of a ReinterpretArrayin Julia 1.11 and beyond
    # https://github.com/JuliaLang/julia/issues/51962
    buf_copy = copy(buf)
    @assert !(typeof(buf_copy) <: Base.ReinterpretArray) "Copying $(typeof(buf)) resulted in another Base.ReinterpretArray"
    write_dataset(dataset, memtype, buf_copy, xfer)
end

function write_dataset(
    dataset::Dataset,
    memtype::Datatype,
    str::Union{AbstractString,Nothing},
    xfer::DatasetTransferProperties=dataset.xfer
)
    strbuf = Base.cconvert(Cstring, str)
    GC.@preserve strbuf begin
        # unsafe_convert(Cstring, strbuf) is responsible for enforcing the no-'\0' policy,
        # but then need explicit convert to Ptr{UInt8} since Ptr{Cstring} -> Ptr{Cvoid} is
        # not automatic.
        buf = convert(Ptr{UInt8}, Base.unsafe_convert(Cstring, strbuf))
        API.h5d_write(dataset, memtype, API.H5S_ALL, API.H5S_ALL, xfer, buf)
    end
end
function write_dataset(
    dataset::Dataset, memtype::Datatype, x::T, xfer::DatasetTransferProperties=dataset.xfer
) where {T<:Union{ScalarType,Complex{<:ScalarType}}}
    tmp = Ref{T}(x)
    API.h5d_write(dataset, memtype, API.H5S_ALL, API.H5S_ALL, xfer, tmp)
end
function write_dataset(
    dataset::Dataset,
    memtype::Datatype,
    strs::Array{<:AbstractString},
    xfer::DatasetTransferProperties=dataset.xfer
)
    p = Ref{Cstring}(strs)
    API.h5d_write(dataset, memtype, API.H5S_ALL, API.H5S_ALL, xfer, p)
end
write_dataset(
    dataset::Dataset,
    memtype::Datatype,
    ::EmptyArray,
    xfer::DatasetTransferProperties=dataset.xfer
) = nothing

"""
    get_datasets(file::HDF5.File) -> datasets::Vector{HDF5.Dataset}

Get all the datasets in an hdf5 file without loading the data.
"""
function get_datasets(file::File)
    list = Dataset[]
    get_datasets!(list, file)
    list
end
function get_datasets!(list::Vector{Dataset}, node::Union{File,Group,Dataset})
    if isa(node, Dataset)
        push!(list, node)
    else
        for c in keys(node)
            get_datasets!(list, node[c])
        end
    end
end

### Chunks ###

# heuristic chunk layout (return empty array to disable chunking)
function heuristic_chunk(T, shape)
    Ts = sizeof(T)
    sz = prod(shape)
    sz == 0 && return Int[] # never return a zero-size chunk
    chunk = [shape...]
    nd = length(chunk)
    # simplification of ugly heuristic target chunk size from PyTables/h5py:
    target = min(1500000, max(12000, floor(Int, 300 * cbrt(Ts * sz))))
    Ts > target && return ones(chunk)
    # divide last non-unit dimension by 2 until we get <= target
    # (since Julia default to column-major, favor contiguous first dimension)
    while Ts * prod(chunk) > target
        i = nd
        while chunk[i] == 1
            i -= 1
        end
        chunk[i] >>= 1
    end
    return chunk
end
heuristic_chunk(T, ::Tuple{}) = Int[]
heuristic_chunk(A::AbstractArray{T}) where {T} = heuristic_chunk(T, size(A))
heuristic_chunk(x) = Int[]
# (strings are saved as scalars, and hence cannot be chunked)

"""
    do_write_chunk(dataset::Dataset, offset, chunk_bytes::AbstractArray, filter_mask=0)

Write a raw chunk at a given offset.
`chunk_bytes` is an AbstractArray that can be converted to a pointer, Ptr{Cvoid}.
`offset` is a 1-based list of rank `ndims(dataset)` and must fall on a chunk boundary.
"""
function do_write_chunk(dataset::Dataset, offset, chunk_bytes::AbstractArray, filter_mask=0)
    checkvalid(dataset)
    offs = collect(API.hsize_t, reverse(offset)) .- 1
    write_chunk(dataset, offs, chunk_bytes; filter_mask=UInt32(filter_mask))
end

"""
    do_write_chunk(dataset::Dataset, index, chunk_bytes::AbstractArray, filter_mask=0)

Write a raw chunk at a given linear index.
`chunk_bytes` is an AbstractArray that can be converted to a pointer, Ptr{Cvoid}.
`index` is 1-based and consecutive up to the number of chunks.
"""
function do_write_chunk(
    dataset::Dataset, index::Integer, chunk_bytes::AbstractArray, filter_mask=0
)
    checkvalid(dataset)
    index -= 1
    write_chunk(dataset, index, chunk_bytes; filter_mask=UInt32(filter_mask))
end

"""
    do_read_chunk(dataset::Dataset, offset)

Read a raw chunk at a given offset.
`offset` is a 1-based list of rank `ndims(dataset)` and must fall on a chunk boundary.
"""
function do_read_chunk(dataset::Dataset, offset)
    checkvalid(dataset)
    offs = collect(API.hsize_t, reverse(offset)) .- 1
    filters = Ref{UInt32}()
    buf = read_chunk(dataset, offs; filters=filters)
    return (filters[], buf)
end

"""
    do_read_chunk(dataset::Dataset, index::Integer)

Read a raw chunk at a given index.
`index` is 1-based and consecutive up to the number of chunks.
"""
function do_read_chunk(dataset::Dataset, index::Integer)
    checkvalid(dataset)
    index -= 1
    filters = Ref{UInt32}()
    buf = read_chunk(dataset, index; filters=filters)
    return (filters[], buf)
end

struct ChunkStorage{I<:IndexStyle,N} <: AbstractArray{Tuple{UInt32,Vector{UInt8}},N}
    dataset::Dataset
end
ChunkStorage{I,N}(dataset) where {I,N} = ChunkStorage{I,N}(dataset)
Base.IndexStyle(::ChunkStorage{I}) where {I<:IndexStyle} = I()

# ChunkStorage{IndexCartesian,N} (default)

function ChunkStorage(dataset)
    ChunkStorage{IndexCartesian,ndims(dataset)}(dataset)
end

Base.size(cs::ChunkStorage{IndexCartesian}) = get_num_chunks_per_dim(cs.dataset)

function Base.axes(cs::ChunkStorage{IndexCartesian})
    chunk = get_chunk(cs.dataset)
    extent = size(cs.dataset)
    ntuple(i -> 1:chunk[i]:extent[i], length(extent))
end

# Filter flags provided
function Base.setindex!(
    chunk_storage::ChunkStorage{IndexCartesian},
    v::Tuple{<:Integer,AbstractArray},
    index::Integer...
)
    do_write_chunk(chunk_storage.dataset, index, v[2], v[1])
end

# Filter flags will default to 0
function Base.setindex!(
    chunk_storage::ChunkStorage{IndexCartesian}, v::AbstractArray, index::Integer...
)
    do_write_chunk(chunk_storage.dataset, index, v)
end

function Base.getindex(chunk_storage::ChunkStorage{IndexCartesian}, index::Integer...)
    do_read_chunk(chunk_storage.dataset, API.hsize_t.(index))
end

# ChunkStorage{IndexLinear,1}

ChunkStorage{IndexLinear}(dataset)         = ChunkStorage{IndexLinear,1}(dataset)
Base.size(cs::ChunkStorage{IndexLinear})   = (get_num_chunks(cs.dataset),)
Base.length(cs::ChunkStorage{IndexLinear}) = get_num_chunks(cs.dataset)

function Base.setindex!(
    chunk_storage::ChunkStorage{IndexLinear},
    v::Tuple{<:Integer,AbstractArray},
    index::Integer
)
    do_write_chunk(chunk_storage.dataset, index, v[2], v[1])
end

# Filter flags will default to 0
function Base.setindex!(
    chunk_storage::ChunkStorage{IndexLinear}, v::AbstractArray, index::Integer
)
    do_write_chunk(chunk_storage.dataset, index, v)
end

function Base.getindex(chunk_storage::ChunkStorage{IndexLinear}, index::Integer)
    do_read_chunk(chunk_storage.dataset, index)
end

# TODO: Move to show.jl. May need to include show.jl after this line.
# ChunkStorage axes may be StepRanges, but this is not available until v"1.6.0"
# no method matching CartesianIndices(::Tuple{StepRange{Int64,Int64},UnitRange{Int64}}) until v"1.6.0"

function Base.show(io::IO, cs::ChunkStorage{IndexCartesian,N}) where {N}
    println(io, "HDF5.ChunkStorage{IndexCartesian,$N}")
    print(io, "Axes: ")
    println(io, axes(cs))
    print(io, cs.dataset)
end
Base.show(
    io::IO, ::MIME{Symbol("text/plain")}, cs::ChunkStorage{IndexCartesian,N}
) where {N} = show(io, cs)

function get_chunk(dset::Dataset)
    p = get_create_properties(dset)
    local ret
    try
        ret = get_chunk(p)
    finally
        close(p)
    end
    ret
end

struct ChunkInfo{N}
    offset::NTuple{N,Int}
    filter_mask::Cuint
    addr::API.haddr_t
    size::API.hsize_t
end
function Base.show(io::IO, ::MIME"text/plain", info::Vector{<:ChunkInfo})
    print(io, typeof(info))
    println(io, " with $(length(info)) elements:")
    println(io, "Offset    \tFilter Mask                     \tAddress\tSize")
    println(io, "----------\t--------------------------------\t-------\t----")
    for ci in info
        println(
            io,
            @sprintf("%10s", ci.offset),
            "\t",
            bitstring(ci.filter_mask),
            "\t",
            ci.addr,
            "\t",
            ci.size
        )
    end
end

"""
    HDF5.get_chunk_info_all(dataset, [dxpl])

Obtain information on all the chunks in a dataset. Returns a
`Vector{ChunkInfo{N}}`.  The fields of `ChunkInfo{N}` are
* offset - `NTuple{N, Int}` indicating the offset of the chunk in terms of elements, reversed to F-order
* filter_mask - Cuint, 32-bit flags indicating whether filters have been applied to the cunk
* addr - haddr_t, byte-offset of the chunk in the file
* size - hsize_t, size of the chunk in bytes
"""
function get_chunk_info_all(dataset, dxpl=API.H5P_DEFAULT)
    @static if hasmethod(API.h5d_chunk_iter, Tuple{API.hid_t})
        return _get_chunk_info_all_by_iter(dataset, dxpl)
    else
        return _get_chunk_info_all_by_index(dataset, dxpl)
    end
end

"""
    _get_chunk_info_all_by_iter(dataset, [dxpl])

Implementation of [`get_chunk_info_all`](@ref) via [`HDF5.API.h5d_chunk_iter`](@ref).

We expect this will be faster, O(N), than using `h5d_get_chunk_info` since this allows us to iterate
through the chunks once.
"""
@inline function _get_chunk_info_all_by_iter(dataset, dxpl=API.H5P_DEFAULT)
    ds = dataspace(dataset)
    N = ndims(ds)
    info = ChunkInfo{N}[]
    num_chunks = get_num_chunks(dataset)
    sizehint!(info, num_chunks)
    API.h5d_chunk_iter(dataset, dxpl) do offset, filter_mask, addr, size
        _offset = reverse(unsafe_load(Ptr{NTuple{N,API.hsize_t}}(offset)))
        push!(info, ChunkInfo{N}(_offset, filter_mask, addr, size))
        return HDF5.API.H5_ITER_CONT
    end
    return info
end

"""
    _get_chunk_info_all_by_index(dataset, [dxpl])

Implementation of [`get_chunk_info_all`](@ref) via [`HDF5.API.h5d_get_chunk_info`](@ref).

We expect this will be slower, O(N^2), than using `h5d_chunk_iter` since each call to `h5d_get_chunk_info`
iterates through the B-tree structure.
"""
@inline function _get_chunk_info_all_by_index(dataset, dxpl=API.H5P_DEFAULT)
    ds = dataspace(dataset)
    N = ndims(ds)
    info = ChunkInfo{N}[]
    num_chunks = get_num_chunks(dataset)
    sizehint!(info, num_chunks)
    for chunk_index in 0:(num_chunks - 1)
        _info_nt = HDF5.API.h5d_get_chunk_info(dataset, chunk_index)
        _offset = (reverse(_info_nt[:offset])...,)
        filter_mask = _info_nt[:filter_mask]
        addr = _info_nt[:addr]
        size = _info_nt[:size]
        push!(info, ChunkInfo{N}(_offset, filter_mask, addr, size))
    end
    return info
end

# properties that require chunks in order to work (e.g. any filter)
# values do not matter -- just needed to form a NamedTuple with the desired keys
const chunked_props = (; compress=nothing, deflate=nothing, blosc=nothing, shuffle=nothing)