import std/strformat
import std/hashes
import ./slice_mem
export slice_mem

const header_size = when not defined(release):
    sizeof(pointer)*2 + sizeof(Natural)
else:
    sizeof(pointer)

type ArrRef*[T] = ref object
    byte_len: int
    when not defined(release):
        # to see in the debugger
        arr_ptr: ptr T
    arr: UncheckedArray[T]

proc newArrRef*[T](size: Natural): ArrRef[T] =
    when defined(TCC):
        var r: ref pointer # workaround for tcc bug
        unsafeNew(r, size * sizeof(T) + header_size)
        result = cast[ArrRef[T]](r)
    else:
        unsafeNew(result, size * sizeof(T) + header_size)
    result.byte_len = size * sizeof(T)
    when not defined(release):
        result.arr_ptr = result.arr[0].addr

template len*[T](a: ArrRef[T]): Natural =
    a.byte_len div sizeof(T)

template byteLen*[T](a: ArrRef[T]): Natural =
    a.byte_len

template low*[T](a: ArrRef[T]): Natural = 0

template high*[T](a: ArrRef[T]): int = a.len - 1

proc `[]`*[T](a: ArrRef[T], i: Natural): var T =
    when compileOption("rangechecks"):
        if i > a.len:
            raise newException(RangeDefect, &"index out of range: {i} >= {a.len}")
    a.arr[i]

proc `[]=`*[T](a: ArrRef[T], i: Natural, v: T) =
    when compileOption("rangechecks"):
        if i > a.len:
            raise newException(RangeDefect, &"index out of range: {i} >= {a.len}")
    a.arr[i] = v

template `[]`*[T](a: ArrRef[T], i: BackwardsIndex): T =
    a[a.len - i.int]

template `[]=`*[T](a: ArrRef[T], i: BackwardsIndex, v: T) =
    a[a.len - i.int] = v

template toPointer*[T](a: ArrRef[T]): pointer = a.arr[0].addr

iterator items*[T](a: ArrRef[T]): T =
    for i in 0 ..< a.len:
        yield a[i]

iterator pairs*[T](a: ArrRef[T]): tuple[key: int, val: T] =
    for i in 0 ..< a.len:
        yield (i, a[i])

iterator mitems*[T](a: ArrRef[T]): var T =
    for i in 0 ..< a.len:
        yield a[i]

iterator mpairs*[T](a: ArrRef[T]): tuple[key: int, val: var T] =
    for i in 0 ..< a.len:
        yield (i, a[i])

proc `$`*[T](a: ArrRef[T]): string =
    result = "ArrRef(["
    if a.byte_len >= sizeof(T):
        let hi = a.high
        for i in 0 ..< hi:
            result &= $a[i] & ", "
        result &= $a[hi]
    result &= "])"

template to*[T](a: ArrRef[T], U: untyped): untyped =
    cast[ArrRef[U]](a)

proc fill*[T](a: ArrRef[T], v: T) =
    for i in a.low .. a.high: a[i] = v

proc newArrRefWith*[T](size: Natural, v: T): ArrRef[T] =
    when defined(TCC):
        var r: ref pointer # workaround for tcc bug
        unsafeNew(r, size * sizeof(T) + header_size)
        result = cast[ArrRef[T]](r)
    else:
        unsafeNew(result, size * sizeof(T) + header_size)
    result.byte_len = size * sizeof(T)
    when not defined(release):
        result.arr_ptr = result.arr[0].addr
    for i in result.low .. result.high: result[i] = v

proc newArrRef*[T](s: seq[T] | ArrRef[T] | openArray[T]): ArrRef[T] =
    result = newArrRef[T](s.len)
    if s.len != 0:
        copyMem(result.arr.addr, s[0].addr, s.len * sizeof(T))

template newArrRef*[T](s: string): ArrRef[T] =
    newArrRef[byte](s.toOpenArrayByte(0, s.high)).to(T)

proc newArrRef*[T; U: not T](s: seq[U]): ArrRef[T] =
    result = newArrRef[T](s.len)
    for i,v in s: result[i] = v.T

when defined(isNimSkull):
    # mainline nim is WAY too buggy to handle this
    proc newSeq*[T: not ref](a: ArrRef[T]): var seq[T] =
        when defined(isNimSkull):
            result = newSeqUninitialized[T](a.len)
        # elif compiles(newSeqUninit[T](a.len)):
        #     result = newSeqUninit[T](a.len)
        # else:
        #     result = newSeqOfCap[T](a.len)
        if a.len != 0:
            copyMem(result[0].addr, a.arr.addr, result.len * sizeof(T))

proc hash*[T](arr: ArrRef[T]): Hash =
    # Make use of stdlib's murmur3
    # NOTE: We use the size of elements in bytes instead of the actual size
    # just in case the actual size is bigger than that.
    hash(cast[ptr UncheckedArray[byte]](arr.arr.addr).toOpenArray(0, arr.len * sizeof(T) - 1))
    # TODO: for bigger elements, would a different algorithm be faster?

template `[]`*[T](a: ArrRef[T], s: Slice[int]): var SliceMem[T] =
    toSliceMem(a, s)

template concatImpl(arrs: untyped) =
    var total = 0
    for a in arrs:
        total += a.len
    result = newArrRef[T](total)
    var offset = 0
    for a in arrs:
        copyMem(result[offset].addr, a.toPointer, a.len * sizeof(T))
        offset += a.len

proc concat*[T](arrs: varargs[ArrRef[T]]): ArrRef[T] =
    concatImpl(arrs)

proc concat*[T](arrs: seq[ArrRef[T]]): ArrRef[T] =
    concatImpl(arrs)

template `&`*[T,U](a: ArrRef[T], b: ArrRef[U]): ArrRef[T] =
    concat(a, b.to(T))